The dual core PC race....

[SamuraiCatJB]

is on!

AMD rumors surfaced about an announcement in April 21st.
http://www.xbitlabs.com/news/cpu/display/20050410210218.html

And they intend to release dual processor and dual core capability at the same time. That's 4 cores on a board.... Not that more than a few people thread that many processors to capability, even I can only pulll about 3.25 cores given my current threads.

Still, Intel followed with their own leak, they will follow the AMD announcement with their own. So the question is.... who will make production first? who will make STABLE and AFFORDABLE dual core products first? Anyone taking bets?

[JackAxe]

AMD of course on the PC side and IBM for us Mac peeps. But then again, Intel did show a "fake" dual-core system, or was it real?...

Rendering and encoding will easilly take advantage of 4 cores on one board and can overwhelm them, tis I why I would still like to get a 2xdual-core XServe this year for all around heavy lifting. But first I still need ot get a Treo. :)

<]=)

[SamuraiCatJB]

It was real, there are even benchmarks.... http://www.legitreviews.com/article.php?aid=188

I love the notes about the tests.... first test... someone complained because he got fragged in Doom3 when his AntiVirus came on automatically and started a weekly scan and his performance dropped and he lost (excuses, excuses)....

cute tests though.... :)

[JackAxe]

I just remember them questioning Intel at one of the press events, because they wouldn't let anyone see inside the box. :)

Any why does it all have to be EXTREME?!! :rolleyes: That is the one word that makes anything lame. :)

Now I must read...

<]=)

[SamuraiCatJB]

I just remember them questioning Intel at one of the press events, because they wouldn't let anyone see inside the box. :)

Any why does it all have to be EXTREME?!! :rolleyes: That is the one word that makes anything lame. :)

Now I must read...

<]=)

Some are Ultra Extreme....

I am waiting for the Double Plus Ultra Extreme. ;p

[JackAxe]

Ultra Extreme my bum. Why don't they just call it the "Excessive Naming of the Extraordinary Extreme addition"?

On this extreme thing, I'm goindg to get the Extreme 3 memory from ScanDisk. :o

<]=)

[Gizmo]

Ultra Extreme my bum. Why don't they just call it the "Excessive Naming of the Extraordinary Extreme addition"?
<]=)
And then put the word "Gold" or "Platinum" on the end.

[zackepceo]

I have an internal copy of the new case badge Intel is handing out to laptop desktop replacement makers..
http://img173.echo.cx/img173/570/p4h0co.jpg

[JackAxe]

Your forgot the Platinum as requested by Gizmo. :D

<]=)

[JackAxe]

If I'm understanding this right, Intel shipped the first Dual-Cores.

http://www.internetnews.com/ent-news/article.php/3496926

When games support DC in abundance, I'm upgrading my PC.

<]=)

[zackepceo]

http://img223.echo.cx/img223/2097/p4p6mc.jpg
:D

[JackAxe]

That's better. :D

You forgot the word "Uber" though. :p

<]=)

[zackepceo]

http://img201.echo.cx/img201/5360/p4p26wq.jpg
You're killing me. :p

[SamuraiCatJB]

and cherry on top? :D ;);) :p

[JackAxe]

Yeah, what about he cherry? :p :D

<]=)

[Karim]

I usually dislike the designs Intel produces (except Pentium-M, my goodness), but once again I will wait until both parties are on the market before I make any who's-better calls. As for who will be first.. AMD will be first at all. Intel will probably be first for consumer-level, though.

[JackAxe]

Intel actually won, I posted a link about 5 up. They did it today those turkeys, shipped out some OEM chips.

I like the peeps that run Intel and their work ethics more then their actual products. (Only going by a show I saw about them the other year.) There's a P4 in my office, but it's not mine and the XScale PXA250 in my Dell Axim is a slug, because it was severly bottlenecked by the memory and other factors. So I haven't really been happy with their products lately. I'm assuming you're refereing to the later Penntium Ms, since for years they were also slugs. :) If the new Treo were using an older XScale, I wouldn't be buying one.

<]=)

[SamuraiCatJB]

Now I want to see one for sale!! :D

[JackAxe]

I'll sell you a dual-corn crop, interested in buying? :)

<]=)

[SamuraiCatJB]

Yeah, what about he cherry? :p :D

<]=)

"he" :eek: :p :D

[JackAxe]

"crry?" :confused: :)

[Karim]

I'm talking about Banias and later specifically... Pentium-M is an incredible design. Sometimes I find it hard to believe it came from Intel. I even wonder if AMD's Turion has a chance at all of matching it.

[Jayman]

I enjoy my Pentium M Laptop --- I have been a loyal AMD user, however the pentium M is a sweet chip.

[Cyker]

I'm talking about Banias and later specifically... Pentium-M is an incredible design. Sometimes I find it hard to believe it came from Intel. I even wonder if AMD's Turion has a chance at all of matching it.

'eh? How so?

From what (admittedly little) I know of it, it's basically a die-shrunk P-III with some battery-saving tech worked into the design...

[Karim]

Yes. They re-released Pentium III and it's taking the notebook platform by storm........

[/sarcasm]

[SamuraiCatJB]

Well, it is official.... AMD is still slated to announce final costs/dates/units on Thursday, the first Dual Core Pentium systems went on sale today. Total units are still low, and only OEM are getting chips, but you can now buy dual core P4 EE based systems from several OEM intel customers, Dell and Alienware are the only ones I checked. Supposedly any OEM can buy them, but only in 1000 unit packages, and first come first serve.... but they are selling and shipping now.

[Karim]

Friggin software developers better knuckle up finally! One thing I've always noticed is poor threading.

[SamuraiCatJB]

Friggin software developers better knuckle up finally! One thing I've always noticed is poor threading.


some of us already do.... I can effectively utilize 3.75 processors, I am working on extending to support up to 8 and 16. :D

but then you are talking about the guy who's accounting software in RPGII (suffers from 2 digit year like Cobol does) was y2k ready in 1994.

[JackAxe]

I look forward to owning a 4 core system, I've never owned more then 2.

Cobal is for "old" nerds. :p And Y2K was a way to scare Christians. Did I just write that out loud. :o

<]=)

[SamuraiCatJB]

ooops it was 1992 anyhow.... I forgot... I left in 1992... the office closed in 1996...

[Karim]

So... looks like Pentium-D is a crock as I expected. P4-64 turned out to be Pentium 4 with "64" slapped on it, now D unsurprisingly turns out to be two P4's slapped on a die. Wow! The future of computing! DAMN I can't wait for the Athlon.

[JackAxe]

I just want all my pro-apps to support 64-bit memory addressing and if It doesn't cost me anymore in the future for a second core, then that's even better. :)

Two cores one one die can make a big difference in performance. I've read up to 210% faster then two seperate procs when it comes to exchanging data back and forth.

<]=)

[SamuraiCatJB]

So... looks like Pentium-D is a crock as I expected. P4-64 turned out to be Pentium 4 with "64" slapped on it, now D unsurprisingly turns out to be two P4's slapped on a die. Wow! The future of computing! DAMN I can't wait for the Athlon.


not sure what you expected... Yes, D is dual core.... we will eventually have 4 and 8 core, etc. systems as time progresses. We are reaching the molecular limitations of silicon dies, so they are increasing outward to keep up with the expectations of Moore's law. Athlon will be the same. Athlon 64 is an Athlon with '64' slapped on it, in fact the exact same as the P4-64 since AMD invented it. And the Athlon 64x2 will be two A64's on one die.

Until we progress into nano chips and ionic storage or holographic storage, we've come near the limit of silicon. In fact the only reason we reached 60nm and below is through some very unique processing methods, including strained silicon.

[JackAxe]

I think that chicken wire is more conductive than strained silicon.

Optical chips are supposed to be available in 2007, not sure when a consumer implication will be here though.

Not good for intel if this peep is correct; :o
http://www.shimon-yanowitz.com/special/Intel/Intel.htm

<]=)

[zackepceo]

Silicon signals degrade to quantum static around the 30nm mark. I think the last silicon production chips will be at most 45nm, but 65nm is also a likely stop.

Artificial diamond transistors can be made reliably to around (I think) 2nm, but after that there's really nothing you can make smaller except fooling around with quantum states.

[SamuraiCatJB]

supposedly strained silicon can almost make the 30nm mark. But it is expensive to widen the molecule gap before make a chip. Although every successive process including the Low K processes used now have come at some expense. but you are correct, without actually building the chip molecule by molecule, there is a physical molecular limit of all silicon processes. However, we all knew this was coming, which was why "some" of us prepared for multi-processor systems. Dual cores have been on the design block for about 4 years now. We knew they would come when the silicon reaches its profitable limit for etching smaller and smaller chip designs.

[SamuraiCatJB]

I think that chicken wire is more conductive than strained silicon.

Optical chips are supposed to be available in 2007, not sure when a consumer implication will be here though.

Not good for intel if this peep is correct; :o
http://www.shimon-yanowitz.com/special/Intel/Intel.htm

<]=)

well, I disagree with some of his assessments. Microsoft survived a direct anti-trust lawsuit with the USA federal court system. They lost, and still survived intact (of course the same threat Microsoft used wouldn't work with Intel, but still, it is hard to judge simply by mistakes).

I disagree with his assessment of dual cores, everyone except him (obviously) knew dual cores were coming. Dual core is not a spin-off of hyperthreading, in fact it is the reverse, hyperthread was a spin-off of dual core. With AMD competing hard with Intel, Intel used the dual core front-end logic with a logical split on processors. It enabled Intel to bypass one of its design flaws to squeeze out a bit more "total" performance.

I do agree Intel is in a bind, and they will have to make a heavy race for it. Everyone said nVidia was dead when the 5800 sucked monkey droppings. Still, they came back with the pressure of competition to up the ante over ATI and now ATI is trying to keep up. Is SLI a fake way of obtaining twice the speed? no it is simply an external dual core rendering system. nVidia and ATI are both expected to tap out dual core chips as well. ATI and nVidia have been attempting dual chip rendering for near 4 years with many failures and minor experimental successes.

The pressure is on Intel to keep up, AMD is expected to follow on the dual core with an entire lineup of Opteron and AMD64 based dual core chips, at least three clock timings for each, and address each market segment from small to large. Intel is not expected to hit the full market on dual core until August, they'd better get their act together and do it well before then.

I agree Intel is in catch-up and keep-up phase, and I agree Intel has really screwed up in the past. Ego will do that, like nVidia with the dentist drill cooler on the 5800 with luke-warm performance. Still nVidia picked up the pieces and came back in the 6000 series with a vengence. Intel needs to do the same. They need to put their nose to the grindstone and kick out some quality stuff to keep up, or yes they are in trouble. Are they guaranteed in trouble now? no, just at risk.

[zackepceo]

Dual-core isn't exactly new, either. IBM has been doing it with the POWER4 since 1999.

[Karim]

I disagree that Intel is in a catch-up phase. Intel simply doesn't care about Pentium 4 anymore, which is why the new releases can be likened to nothingness. That means that I disagree with you about AMD's being the same thing. Even though the core technology is the same, the design using the technology is not. I am positive that AMD will introduce a useful and effective dual-core design as opposed to Intel's "two cores" approach. Yes, two cores is better than one. I get it. But that's not all there is to processing; that's my point. Inefficient designs create unnecessary bottlenecks and overhead. If you don't meet developers (and consequently, consumers) halfway, the stirring will eventually turn into some kind of shitstorm.

Yes, Intel definitely cares little for P4 anymore. Ever since they stumbled onto the beautiful "Pentium-M" design, they've been focusing a lot of energy into it. Intel will probably succeed Pentium 4 with a desktop version of Pentium-M. The mobile version already outperforms P4, making it a natural step in the progression.

[SamuraiCatJB]

it outperforms the P4 for "some" operations... we've had this discussion before. I can quickley and easily bring the P-M to its knees at only 75% of the P4 level performance. Heavy double precision mathematics in rapid succession without letup will floor the PM. It is a great general purpose mobile, but heavy mathematical loads the P4 still rides just below the XEON.

I could be wrong, but I doubt there will be a PM desktop except in a mac-mini competitor (which Intel is already pushing the industry to do). Intel has plans for a full range of dual core systems. The current dual core was only put out to beat AMD to the punch. Further systems will be dual core on a single chip and be indestinguishable without looking at the circuit diagrams to know it is dual core.... well... except for the 1067 pins... :eek:

There will even be dual core PMs for the mobile and mac-mini-type systems.

[SamuraiCatJB]

Dual-core isn't exactly new, either. IBM has been doing it with the POWER4 since 1999.

and now that NDA's are up, my project leader saw Intel's dual core prototype in 2001+ timeframe. In affect, we started running dual cores at the 486 level in PCs, when Intel combined the 80386 chip plus math coprocessor into one chip (still distinct processing cores) and called it the 80486. An error in the etching process flawed many chips, which prompted a suggestion to disable the math processing core by litterally laser cutting the leads, and releasing it as the .... tada... 80486sx. It was popular enough, they actually started building the bloody things as a deliberate design 3 months later.... go figure.... :)

[Karim]

The mobile PMs were what I was talking about, lol. Anyway, the point I'm making is about the direction each line is going in. You say "some operations," but relatively speaking the PM's efficiency beats the P4 in almost everything at all. PM is going up, P4 is going down. Which core do you think has the most influence in upcoming and ongoing designs and research, hmm?

[SamuraiCatJB]

Time will tell, but as far as I can tell the P4 is. Marketing on the PMs are high, but technolgoy advancement in that line is minimal. Until they solve the high end math operations issue it will never replace the P4 as a gaming chip, nor as a workstation chip, nor as a business chip. It will remain as a generic entertainment chip to support mobile and mini systems that play DVDs, low-end to mediocre games, office word processing and the internet.

It's great for Administrative assistants and the boss, lousy for CAD, engineering, mathematics, physics, any advanced game play. As long as you never need anything that is actually technologically advanced, its a great processor.

[SamuraiCatJB]

Here are some real tested benchmarks rather than sales figures....
http://www.gamepc.com/labs/view_content.asp?id=gmso&page=4 CPU use....
http://www.gamepc.com/labs/view_content.asp?id=gmso&page=5 memory throughput...
http://www.gamepc.com/labs/view_content.asp?id=gmso&page=7 Maya rendering And Adobe
http://www.gamepc.com/labs/view_content.asp?id=gmso&page=8 Video Encoding (note Pm wins audio encoding for those folks who only do music and never video)
http://www.gamepc.com/labs/view_content.asp?id=gmso&page=9 Various Physics applications

The only advantage the Pm has is battery/power consumption, which it wins hands down above all competitors. Which leaves it always the best in the mobile market, but very much behind in the desktop market. I fully support the PM in the mobile arena, and even recommend the same. It does NOT however out power a desktop chip, it is a full 33% behind the P4 3.4ghz chip no matter what the salesmen tell you. It is a miracle in low power energy consumption, it will not compete against a desktop.

The PM is like my hybrid, low power, great energy conservation, great for travelling.... But I would never drag race with it, I'd loose every time. On the road, I drive a "hybrid".... At work, I compute on a "sports car."

[Karim]

GamePC.com? Funny, here's something from that same site:

http://www.gamepc.com/labs/view_content.asp?id=dfipm&page=10

Again, I stress the advantage of superior design and potential over raw power. Pentium-M is just plain better. You can bet your *** that just about all future work involves its design. If I had the money, I would put these machines together myself and send them to you for the sole purpose of showing you what I'm talking about. Design sophistication is not a game. It's real life. And it's better.

[SamuraiCatJB]

I also note that although they overclock the PM, they do not similarly overclock the P4s or Athlons, nor do you they choose a point that fills the pipeline of the AGP 4x system in order to give the PM a fighting chance. They do not compare it 533 or 600 DDR2 based memory systems which are available on PCs and not PMs. Every test where the PM wins, they have deliberately chosen older archetecture that is close to the PM rather than taking the latest archetecture available for the other chips. Intel does the same, though Intel does worse by crippling the entire system including disks to provide no point of reference.

With SLI availabe to PCs, memory up to and exceeding DDR2-600, a test utilizing the full range of the AGP 8x or PCI-x archetecture, the PM slacks off rapidly. It is a very good chip compared to where the P3 was, but utilize an advanced mode that requires SSE-3 (P4) command set, wider bus speeds and the like and PM shows its true colors. I still recommend it, just not to the highest end crowd. Because of SLI and extra detail modes in Doom3, bandwidth matters and frame rates are up to twice and thrice the PM if you actually use what is fully available. Step it down to last years technology and yes, PM kicks. Its a good chip, but it competes against what was, not what is new.

[Karim]

http://www.gamepc.com/labs/view_content.asp?id=dfipm&page=7


I hate to be the bearer of bad news, but uh, the configurations overview clearly describes PCI-Express videocards and DDR2 memory on these very very new Pentium 4 chipsets. That was the point of the whole comparison: PM was on this old chipset and smashing the new ones anyway. When it was overclocked, lol, to SUB-P4 SPEEDS.

I understand your points, man, but after reading some of the whitepapers on Pentium-M, there's nothing you can say that will make sense of any P4-superiority-over-PM logic. The thing is fantastic. The things that you say Pentium 4 beats it in.... it doesn't. lol. The Athlon does, but P4 most certainly does not.

SSE3? DDR2? PCI-Ex? It has the latter two now. Yonah has (read: will have) that first item as well. Again, I'm sorry to be the bearer of bad news, but Pentium 4's time has come and it has gone.

[SamuraiCatJB]

You are still not understanding... the PM is nothing new. It is a P3 with P4 Xeon extensions, applying it to the P4 archetecture will not gain you anything. And yes, they have the archetecture, but the specifics of the game-play chosen in the game tests (identifiable by the frame rate on P4 based systems) is on the low-end. You can put a X600 and get the exact same performance because you have throttled down the detail to allow the AGP 4x bottle neck not to overload, and the 533Mhz front side bus not to overload.

If you want PM performance in a P4, buy a Xeon chip, it has the same 128bit floating point processor (only faster) and the same 2gig cache (only faster), and the same long pipelining process. It has 800Mhz FSB and a few models even support faster memory. Therefore, yes, the PM time has come... it is called Xeon, and Intel already has it, only better.

You can deliberately throttle down detail to allow a PM chip to win, but you can ALWAYS fail the PM chip over the P4 and P4 XEON if you do not.

[SamuraiCatJB]

Here is what the frame rates should be if you don't pull detail back.
http://www.pcmoddingmy.com/content.php?review.220.5

Never test a ferrari by pulling 6 spark plugs and then tell me it is slower than my little 4cyl Civic.

[Karim]

It most certainly is not a P3. I implore you to do some reading, sir. It has a very vaguely similar overall structure to the final P3, but it's so vague that even that can't really be considered a "similarity." I don't know where you're getting these ideas, but stop listening to whoever is telling you these things. And PM does not have a long pipelining process, that's Pentium 4 and Xeon. Pentium 4 has basically BEEN nothing but Xeon for the past year or two.

They only "throttled down" the resolution. And I don't see any "specifics chosen," either, just a bunch of recent and demanding games set to XGA. Anyway, why are you talking about video cards and their buses when the subject is CPUs?? The point is, with all other factors normalized, the PM pulls ahead. Why would you point out lowered detail when that factor should allow a beefier system and bus even MORE speed? Come on, dude. And Pentium-M now (recently) has PCI-Ex and DDR2 support, so I don't know why you keep on that.

[Karim]

Here is what the frame rates should be if you don't pull detail back.
http://www.pcmoddingmy.com/content.php?review.220.5

Never test a ferrari by pulling 6 spark plugs and then tell me it is slower than my little 4cyl Civic.

That link is to an Nvidia/ATI bench comparison.

PS: The Ferrari with sparkplugs pulled would be the Pentium-M on three-year-old chipset technology. You can put a CPU on anything you want, really. Let's talk about the CPUs for once, and not the chipset, which can be changed.

[SamuraiCatJB]

That link is to an Nvidia/ATI bench comparison.


correct, but note even an a faster graphics card, the frame rates are lower than the tests in the one you mark as better. What they have done is reduced detail, throttled back the IO path to the graphics processor so that all cards are equal in a AGP 4x environment. They have in affect pulled half the spark plugs in the engine to make the comparison.

Every review I read about the PM says the same thing:
The Pentium-M performed quite well in older games and everyday applications like Microsoft Office and Photoshop. It didn’t do too well in 3D Workstation and Audio/Video Encoding benchmarks and part of the reason for that is its weak FPU which is understandable as most of us tend to do that kind of stuff on our heavy duty desktops.

It is not, nor will ever fully replace the desktop line. It does replace a lot of the low end range, but all the new technology is flowing into the higher end desktops first, and PM last. Yet you said the reverse. This just is not true. The PM gets the changes last in technology, it is still at lower graphics IO, lower FSB, lower memory speeds, lower IO all around. Because it was never intended to replace the entire range of processors. The high end is still at the high end, the PM can fill the low to medium range, and does so surprisingly well. Many Kudos for an excellent low end to medium end chip. The energy and heat savings makes it the absolute king in small environments, hands down a winner for mobile and micro chassis, 150% agree there. But it does not, nor ever has competed against the "new technology" desktops, only the previous generation of desktops, software and gameplay.

It is a great chip, just not the greatest, nor even is the P4 or the P4 Xeon. Intel is in "catch-up" phase trying to compete. The PM opens up some new markets for Intel to pull back market share from Apple, it widens Intel's market share in order to try to pull profit from a larger market since they are having trouble competing. This is a very smart plan, assuming they funnel the wider market into advancements. This is exactly how ATI came to beat nVidia at the 5000 series. ATI has always had a wider market share, nVidia never tried to compete all markets that ATI was in. ATI used that market share to drive profits into R&D and up their technology. nVidia was blind sided, but in response widened their market share to compete with ATI in all areas, and added new areas where there was no competition, again using the market share to drive it to the top again.

This is smart, this is good. But it doesn't make the 5800 suddenly a magic marvel any more than PM. Intel is doing the right thing, let's see where they take it. I expect it will neither be a PM nor a P4 archetecture, but one designed more for scaleability, similar to an IBM design. The future is super-scaler architecture, the P4 nor the PM is super-scaler, if Intel is to compete they had better find a super-scaler architecture as the graphics companies have done.

[SamuraiCatJB]

They only "throttled down" the resolution. And I don't see any "specifics chosen," either, just a bunch of recent and demanding games set to XGA. Anyway, why are you talking about video cards and their buses when the subject is CPUs?? The point is, with all other factors normalized, the PM pulls ahead. Why would you point out lowered detail when that factor should allow a beefier system and bus even MORE speed? Come on, dude. And Pentium-M now (recently) has PCI-Ex and DDR2 support, so I don't know why you keep on that.


No, they throttled down detail. This reduceds physics processing in game-play reducing CPU overhead (how Doom will run on a P3 -- check the minimum requirements, very low). Reducing detail reduces the matrix operations, memory processing, CPU overhead, vertices reproduction, rotation. Everything suddenly speeds up... Wow... surprise. Sure a PM can add as fast as a P4 and P4 Xeon. but if you try to emulate full physics in game play by using high detail, the CPU would bind up, choke and fall apart. Because the PM is not, nor has ever been designed for FPU operations. This is little difference from the 486sx and the 48dx. As long as you don't run any math, the 486sx ran as fast as a dx. Intel made a fortune off off of the 486sx through this marketting strategy. They are simply doing it again.

PM architectures do not support latest technology, therefore the CPU will always be bound by its supporting IO architecture. This is where the CPU design fails, single path, single pipe, low speed IO for all supporting IO to surrounding motherboard makes the CPU fast, but can never compete side by side with against a modern motherboard architecture designed to maximize an Athlon or a P4. They had to overclock the motherboard to get it to support DDR400, DDR400 has been out for ages.

A CPU does not exist in a vaccuum. It must be mounted on a motherboard, tied to an IO package and communicate with it. The entire design of the PM uses low-speed IO to the parent architecture and high speed cached processing locally to make up for it. If you set it side by side and compare it with a CPU that does not suffer from IO problems, you get a clearer picture of the IO capabilities of the chip. Again, I will repeat myself. It is a great chip, but it is no miracle.

[SamuraiCatJB]

Back to the original subject. AMD released their lineup. Six Dual core Opteron's, 4 Athlon-64x2 (dual core). The opterons coming in first in May, and the Athlon 64x2 coming in June.

The big surprise though!! hold on to your hats! AMD’s dual-core processors for desktops and servers are expected to be able to operate within existing Socket 939 and Socket 940 infrastructures provided that the mainboards have BIOS support for them and comply to AMD’s thermal and voltage specifications for all processors made using 90nm process technology.

I "doubt" this means that a reflash of a bios could support dual core, but that does mean minimal cost to change motherboard designs to support dual core. Fewer changes usually means more stability up front. In 60 days we could easily see the entire market populated with dual core AMD designs.

AMD's announcement does say they are going back on their original plan to drop the FX line from the dual core lineup. Another announcement will come later on what FX dual cores will be coming as well as Athlon 64Mx2 dual cores.

The top-of-the-range dual-core desktop processor from Advanced Micro Devices will cost $1001 in 1000-unit quantities, which is $2 higher compared to Intel Pentium Extreme Edition 840 processor priced at $999. But the chip called Athlon 64 X2 4800+ will still be not targeted at gamers, for them AMD plans to release a yet another AMD Athlon 64 FX product.

[Karim]

correct, but note even an a faster graphics card, the frame rates are lower than the tests in the one you mark as better. What they have done is reduced detail, throttled back the IO path to the graphics processor so that all cards are equal in a AGP 4x environment. They have in affect pulled half the spark plugs in the engine to make the comparison.

We're talking about CPUs.

But to humor you: WHA??? Where do you see any notes that IO paths were throttled?? Dude, GamePC.com is not a PM propaganda site out to destroy Pentium 4's reputation, lol. The notes clearly state that the P4 setup is on a PCI-Ex X16 card and that the PM setup is "limited to 4X AGP due to chipset limitations." At the risk of making myself look stupid, WHAT THE HELL ARE YOU TALKING ABOUT?!

I don't normally get the chance to discuss these things, btw. Thank you.


Every review I read about the PM says the same thing:

It is not, nor will ever fully replace the desktop line. It does replace a lot of the low end range, but all the new technology is flowing into the higher end desktops first, and PM last. Yet you said the reverse. This just is not true. The PM gets the changes last in technology, it is still at lower graphics IO, lower FSB, lower memory speeds, lower IO all around. Because it was never intended to replace the entire range of processors.

No, actually that is because Intel initiated this design as a mobile design, and as such, is focusing on its mobile advantages. That means crippling I/O and memory, which translates to better battery.

The high end is still at the high end, the PM can fill the low to medium range, and does so surprisingly well. Many Kudos for an excellent low end to medium end chip.

How does ANY low, or even medium product, best a high-end product at ANY resolution? This simply should not be, unless one of the two has a superior design.


But it does not, nor ever has competed against the "new technology" desktops, only the previous generation of desktops, software and gameplay.

Look at the GamePC.com benches again, and think to yourself:

P4 = Dual-channel DDR2 200MHz core, PCI-Ex X16
PM = Single-channel DDR 166MHz core, AGP 4x

That was last year. Now:

PM = Dual-channel DDR2, PCI-Ex X16.

What do you think THOSE numbers would look like, hmmm?


It is a great chip, just not the greatest, nor even is the P4 or the P4 Xeon. Intel is in "catch-up" phase trying to compete. The PM opens up some new markets for Intel to pull back market share from Apple, it widens Intel's market share in order to try to pull profit from a larger market since they are having trouble competing. This is a very smart plan, assuming they funnel the wider market into advancements. This is exactly how ATI came to beat nVidia at the 5000 series. ATI has always had a wider market share, nVidia never tried to compete all markets that ATI was in. ATI used that market share to drive profits into R&D and up their technology. nVidia was blind sided, but in response widened their market share to compete with ATI in all areas, and added new areas where there was no competition, again using the market share to drive it to the top again.

This is smart, this is good. But it doesn't make the 5800 suddenly a magic marvel any more than PM. Intel is doing the right thing, let's see where they take it. I expect it will neither be a PM nor a P4 archetecture, but one designed more for scaleability, similar to an IBM design. The future is super-scaler architecture, the P4 nor the PM is super-scaler, if Intel is to compete they had better find a super-scaler architecture as the graphics companies have done.

Again, design superiority. That's why ATI beat Nvidia last round. Don't fool yourself. The GeForce 5 series was garbage.

[SamuraiCatJB]

VIA Technologies today announced support for the latest dual-core AMD Opteron processors with its K8 series of core-logic chipsets.

The VIA K8 series chipsets, including the new VIA K8T890, leverage VIA Hyper8 technology for high speed and low latency performance with the latest AMD Opteron processors with dual-core architectures. The VIA K8T890 also integrates 20 lanes of PCI Express connectivity, enabling high bandwidth peripheral connections and support for the latest professional-class 3D graphics solutions for workstation applications.

I was actually hoping for more lanes.... :(
dual 16x is only 32 lanes. Dual 8x means 16 lanes of PCIe graphics leaves only 4 lanes for general IO.... *sigh* They better give me 24 to 32 soon.... grrrrrr....

[SamuraiCatJB]

We're talking about CPUs.

But to humor you: WHA??? Where do you see any notes that IO paths were throttled?? Dude, GamePC.com is not a PM propaganda site out to destroy Pentium 4's reputation, lol. The notes clearly state that the P4 setup is on a PCI-Ex X16 card and that the PM setup is "limited to 4X AGP due to chipset limitations." At the risk of making myself look stupid, WHAT THE HELL ARE YOU TALKING ABOUT?!

I don't normally get the chance to discuss these things, btw. Thank you.


I know this gets confusing. Even in my peers few understand the complexity of balancing pipelines to CPUs with graphics. We'll take it a step at a time. Game complexity is more than resolution. If you look, the link I gave you showed 1024x768 noAA/AF, that is equivalent resolution to your mode. But the total frame rate for the exact same game is 30% lower, but those are actually 15% faster cards than the one in gamepc article. So how can this be? The one I quoted actually says "high mode" -- they enabled full complexity for advanced graphics. This requires wider IO buses (or if you prefer PCIe more "lanes" -- PCIe is actually easier to explain, but I can save that for another day), more texture memory, faster graphics cards, etc. However because the graphics cards are faster and the lanes are wider to feed it, the CPU has less to do, it would idle itself to sleep. So game designers and other performance experts have to increase CPU processing to match the same scale as the GPU, balance the pipeline as it is called. By throttling down the complexity on the gameplay, they increased the frame-rate, but throttled down CPU and IO at the same time. The PM is able to compete, the P4 is probably idling expecting the AGP pipe to be much smaller. So although you think you are making a valid comparison between Apples and oranges by taking away the peel and comparing volume, you aren't taking into account that the peel of an apple is less than a mm and the peel of an orange is much thicker. It sounds reasonable in comparison, but it isn't. There really is no adequate way to compare apples and oranges.... listen to Jack and I go on about G5.... hehehe I tease him, I know it is decent, but until recently it suffered from IO flaws to the graphics system by architecture design. It was by all rights a superior CPU even in the G3/G4 days, but the architecture the CPU was designed for (and therefore had inherent issues in that design -- he's gonna rip me a new on on this hhehehehe), held it back. The new G5 design with PCIe support, wide lanes, fast response from CPU to Bus to memory and back. You can think of this as the "working triangle" in a kitchen design (I took architecture drafting in school, so you will have to forgive my analogy). You can have the worlds greatest stove, but if you put the refridgerator in the bedroom no one will buy the house -- well, except those that want sodas or beer in bed.... The path from the sink to the stove to the refridgerator is the working triangle of a kitchen. The longer the triangle the more effort it is to cook, and the less happy people will be with the overall performance. They might buy the house for its great stove with flame grill and built in overhead microwave... but you never understand why it makes you miserable to walk back and forth across the kitchen to get ingredients from the refridgerator. It is because ther working triangle is so large.

Prior to the Pentium the CPU was far removed from its IO path, trying to compete independantly as CPU to CPU in comparison to RISC chips of the time. This was rather a silly thing to do, if you are going CISC don't hold back, if you are going RISC then you want to, they do not compete on the same footing because the design is different. Pentium started integrating portions of the environment into itself requiring north/south bridges as middle men in the process of integrating to the system. The working triangle of the PC is the CPU to memory to IO. IO includes everything on the front side bus from APG/PCIe graphics to disks. Many times you sacrifice some stages of that IO in preference to others. Servers sacrifice all graphics IO for RAID performance, 3D workstations like mine have to balance the two (thus my grumbling). The CPU design and architecture is how it fits into that working triangle. All CPUs have to do it. You can have the fastest CPU in the world, and partner it with a memory bus that reads 266mhz memory and it will suffer (this was my last P4 Xeon before the current generation P4 Xeon, egad but that first design was screwy). Similarly, if you sacrifice IO path and beef up memory, you have a great design for diskless super-servers (they do exist) that do nothing but work in mass-arrays of processing, no disk IO, no graphics IO, just memory to CPU and the occassional network feedback to the parent.

PM design offers a very narrow IO to memory, single stage with a large cache and a north-bridge with a pre-fetch to memory. For sequential processing this is great, very business like intent. All business models tend to be linear in design, and the pre-fetch on memory in the PM allows it to compete with the big boys. While the processor is working, the bridge chip is trying to predict what the CPU will need next from the last memory request and have it "ready". Spread sheets work this way, images, word processing are all linear models. The PM compares adequately for the intelligent pre-fetch on the northbridge memory controller, NOT the CPU. This is a plus for architecture of the surrounding support system and is good. The single channel memory bus of the CPU actually "looks" like a dual channel of P4 or almost that of dual/quad of an AMD/Xeon. But this has nothing to do with the CPU, that is a smart north bridge set that makes up for the narrow channel memory design of the PM. That northbridge design could be integrated into most any PC design. But it cannot predict random fetches and non-linear mathematical systems. Which is why if you build a table driven physics engine and use linear approximation between samples, you get decent gameplay with predictible memory design and PM has an advantage, but not from the CPU, from the northbridge. A full physics engine is utilizing 3D spatial position using college or even High School physics is a random design. Though a small enough one can fit in the L2 cache of a processor, a complex one will not. Spatial analysis, radiosity rendering, Bi-directional-reflectance-map light physics are so complex with so many constants/variables that they cannot go on the CPU cache, they have to be IOed in from random memory locations and channel through the CPU the hard/long way. Except as graphics designs have increased, freeing up more CPU time, the game developers add more complex physics engines on the CPU. Thus "high detail" mode is more than resolution. you just added CPU complexity. What you like best about the PM is the northbridge intelligent pre-fetch on memory, but that has nothing to do with the CPU.



Again, design superiority. That's why ATI beat Nvidia last round. Don't fool yourself. The GeForce 5 series was garbage.
That we will agree, but nVidia learned from their mistakes and came back strong. I am hoping Intel will also. They have widened their base to pull in more profits, I hope to this end. But they better hurry.

[zackepceo]

The 'highest end crowd' should not be considering buying a Pentium. If their work is critical, they would be using a Xeon, Opteron, or G5. As it is, the Pentium M is poised to become the next flagship processor for Intel, and it's a good thing. It's nothing to Intel for them to bolt on a SSE3 unit, add dual core functions, and call it the Pentium 5.
The Pentium M is very Pentium 3-like in architecture, and this is also a good thing. The Pentium 4 suffered from acute marketing inflation. Netburst is probably the worst mistake that Intel ever made.

[SamuraiCatJB]

The 'highest end crowd' should not be considering buying a Pentium. If their work is critical, they would be using a Xeon, Opteron, or G5. As it is, the Pentium M is poised to become the next flagship processor for Intel, and it's a good thing. It's nothing to Intel for them to bolt on a SSE3 unit, add dual core functions, and call it the Pentium 5.
The Pentium M is very Pentium 3-like in architecture, and this is also a good thing. The Pentium 4 suffered from acute marketing inflation. Netburst is probably the worst mistake that Intel ever made.

very true. I expect to see a draw backwards from the P4 mistakes, back to the solid P3 from which the PM was made (no it is not a P3, it is a P3 "design" with additional design enhancements -- P3 being the level of CISC instructions built into the microcode, SSE2 rather than SSE3, reduction in FPU, etc). Intel has already stated that there will be a dual core PM, everyone has long since admitted that dual cores all around will be here this year.

I expect the P5 will learn from both the PM and the P4 (and probably the competition) and hopefully be a better thought out chip than the P4.

BTW, I peeked at the AMD roadmap.... they take it down to 22nm processes in 2011.... :eek: That may be wishful thinking....

Lets turn the earth into a giant computer to calculate the ultimate question to life the universe and everything. :)

[Karim]

BTW, I peeked at the AMD roadmap.... they take it down to 22nm processes in 2011.... :eek: That may be wishful thinking....

22nm processes could theoretically be controlled on better materials than silicate-based ones, or by ultraviolet laser, without actually using quantum computing models.


Lets turn the earth into a giant computer to calculate the ultimate question to life the universe and everything. :)

I could get my TH55 to say "42" :p

I know this gets confusing. Even in my peers few understand the complexity of balancing pipelines to CPUs with graphics. We'll take it a step at a time. Game complexity is more than resolution. If you look, the link I gave you showed 1024x768 noAA/AF, that is equivalent resolution to your mode. But the total frame rate for the exact same game is 30% lower, but those are actually 15% faster cards than the one in gamepc article. So how can this be? The one I quoted actually says "high mode" -- they enabled full complexity for advanced graphics. This requires wider IO buses (or if you prefer PCIe more "lanes" -- PCIe is actually easier to explain, but I can save that for another day), more texture memory, faster graphics cards, etc. However because the graphics cards are faster and the lanes are wider to feed it, the CPU has less to do, it would idle itself to sleep. So game designers and other performance experts have to increase CPU processing to match the same scale as the GPU, balance the pipeline as it is called. By throttling down the complexity on the gameplay, they increased the frame-rate, but throttled down CPU and IO at the same time. The PM is able to compete, the P4 is probably idling expecting the AGP pipe to be much smaller. So although you think you are making a valid comparison between Apples and oranges by taking away the peel and comparing volume, you aren't taking into account that the peel of an apple is less than a mm and the peel of an orange is much thicker. It sounds reasonable in comparison, but it isn't. There really is no adequate way to compare apples and oranges.... listen to Jack and I go on about G5.... hehehe I tease him, I know it is decent, but until recently it suffered from IO flaws to the graphics system by architecture design. It was by all rights a superior CPU even in the G3/G4 days, but the architecture the CPU was designed for (and therefore had inherent issues in that design -- he's gonna rip me a new on on this hhehehehe), held it back. The new G5 design with PCIe support, wide lanes, fast response from CPU to Bus to memory and back. You can think of this as the "working triangle" in a kitchen design (I took architecture drafting in school, so you will have to forgive my analogy). You can have the worlds greatest stove, but if you put the refridgerator in the bedroom no one will buy the house -- well, except those that want sodas or beer in bed.... The path from the sink to the stove to the refridgerator is the working triangle of a kitchen. The longer the triangle the more effort it is to cook, and the less happy people will be with the overall performance. They might buy the house for its great stove with flame grill and built in overhead microwave... but you never understand why it makes you miserable to walk back and forth across the kitchen to get ingredients from the refridgerator. It is because ther working triangle is so large.

Prior to the Pentium the CPU was far removed from its IO path, trying to compete independantly as CPU to CPU in comparison to RISC chips of the time. This was rather a silly thing to do, if you are going CISC don't hold back, if you are going RISC then you want to, they do not compete on the same footing because the design is different. Pentium started integrating portions of the environment into itself requiring north/south bridges as middle men in the process of integrating to the system. The working triangle of the PC is the CPU to memory to IO. IO includes everything on the front side bus from APG/PCIe graphics to disks. Many times you sacrifice some stages of that IO in preference to others. Servers sacrifice all graphics IO for RAID performance, 3D workstations like mine have to balance the two (thus my grumbling). The CPU design and architecture is how it fits into that working triangle. All CPUs have to do it. You can have the fastest CPU in the world, and partner it with a memory bus that reads 266mhz memory and it will suffer (this was my last P4 Xeon before the current generation P4 Xeon, egad but that first design was screwy). Similarly, if you sacrifice IO path and beef up memory, you have a great design for diskless super-servers (they do exist) that do nothing but work in mass-arrays of processing, no disk IO, no graphics IO, just memory to CPU and the occassional network feedback to the parent.

PM design offers a very narrow IO to memory, single stage with a large cache and a north-bridge with a pre-fetch to memory. For sequential processing this is great, very business like intent. All business models tend to be linear in design, and the pre-fetch on memory in the PM allows it to compete with the big boys. While the processor is working, the bridge chip is trying to predict what the CPU will need next from the last memory request and have it "ready". Spread sheets work this way, images, word processing are all linear models. The PM compares adequately for the intelligent pre-fetch on the northbridge memory controller, NOT the CPU. This is a plus for architecture of the surrounding support system and is good. The single channel memory bus of the CPU actually "looks" like a dual channel of P4 or almost that of dual/quad of an AMD/Xeon. But this has nothing to do with the CPU, that is a smart north bridge set that makes up for the narrow channel memory design of the PM. That northbridge design could be integrated into most any PC design. But it cannot predict random fetches and non-linear mathematical systems. Which is why if you build a table driven physics engine and use linear approximation between samples, you get decent gameplay with predictible memory design and PM has an advantage, but not from the CPU, from the northbridge. A full physics engine is utilizing 3D spatial position using college or even High School physics is a random design. Though a small enough one can fit in the L2 cache of a processor, a complex one will not. Spatial analysis, radiosity rendering, Bi-directional-reflectance-map light physics are so complex with so many constants/variables that they cannot go on the CPU cache, they have to be IOed in from random memory locations and channel through the CPU the hard/long way. Except as graphics designs have increased, freeing up more CPU time, the game developers add more complex physics engines on the CPU. Thus "high detail" mode is more than resolution. you just added CPU complexity. What you like best about the PM is the northbridge intelligent pre-fetch on memory, but that has nothing to do with the CPU.

You don't have to talk down to me, lol, I've understood you perfectly from the start. But AGAIN, most of the things you are discussing here are based on parts that are NOT the CPU itself. As far as the CPU's design to WORK with these other parts, I consider that to be yet relatively untested.

Your points are extremely valid, but in the end are merely theoretical. I point you to a later point in my GamePC.com link, which shows the overlocked PM making better scientific math scores than the P4 Extreme. This cannot be "faked" by programming for optimized or throttled-back I/O. The 3.46GHz P4EE won the molecular dynamics score by 0.9 and lost the Primordia model by 7.2. Neither of them could beat the Athlon. Surprise surprise.

Anyway, my point was never that the current Pentium-M could smash any processor. My point was that it has a better design. I should have pointed out that I was including potential in that judgement. The Pentium 4 has one foot in the grave; the Pentium-M is a newborn that is right on the Pentium 4's heels and already past and above it in many cases. In those (previously unspoken.. sorry) respects, Pentium-M is just plain better, and in philosophy will be the successor of the Pentium 4.

[JackAxe]

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
Not to stir up more debating, but here's an article about Intel and AMD's new Dual Cores. ;)

http://www.cooltechzone.com/index.php?option=content&task=view&id=1232&Itemid=0&limit=1&limitstart=1

<]=)

[SamuraiCatJB]

I finally got back from all my vacations.. catching up on the news.... Intel is loosing hands down in every mode of operations....

and....


Unlike Intel's dual-core Pentium-D and Pentium Extreme Edition processors, which will require a new motherboard that's based on a compatible chipset like the i955, i945 or the nForce 4 Intel Edition, AMD's new dual-core "X2" processors will work with virtually any socket 939 platform as it exists today. The only caveat is that your motherboard will need a BIOS upgrade to support an Athlon 64 X2. We suspect most major motherboard manufactures will have updated BIOS files available in time for the official release of the Athlon 64 X2.

[SamuraiCatJB]

Jack, you might find this more interesting than most....
Additional Applications

The AMD X2 will not see significant advantages in most of these benchmarks due to the lack of multi-threading and more importantly multi-cpu support. In fact according to AMD the key applications that will gain a dual core scaling advantage are:

* 3DS Max 6 ~80% performance gain
* LightWave 8 ~70% to 85% performance gain
* Canopus ProCoder, Procoder Express ~68% faster MPG to DVD
* Windows Media Encoder ~76% faster DV to WMV
* Mandelbrot generator C/C++ runs ~99% faster
* Adobe Premiere Pro ~17% - 60% faster depending on the project
* Cyberlink Power Director ~85% MPEG2, ~90% WMV, ~99% DivX
* Windows MovieMaker ~42%
* Photo Shop depends on the filters used (not all filters are multithreaded)

[JackAxe]

I use simliar, if not the same apps on my Mac with 2 proccessor, so I have an advantage now. :)

But very very impressive. By the time it's out, I'll probably have the itch to assemble a new PC, hopefully the finances also match up.

<]=)

[SamuraiCatJB]

I am eyeing my finances now too.... x2... mmmmmmm

[SamuraiCatJB]

The performance summaries are coming out now... good reviews all around on the dual cores. Of course dual core opterons are the big winners, but they don't really compare well to P4 anyhow, so it was expected. And Intel (who better get their act together) doesn't have a dual core Xeon.

The inclusion of the SSE3 instruction set with the new AMD 64 x2 chips is paying off their performance compared to the P4 creates a landslide victory. As I and a few others knew, the only reason P4s were holding their own against anything was because of the inclusion of SSE3 on the P4 and exclusion of it on AMD. Head to Head with both dual chip designs using SSE3 instructions, AMD leaves the P4 behind. I knew AMD would do this, and glad they did. Now the P4 design flaws are showing through much more clearly.

http://www.motherboards.org/reviews/hardware/1513_10.html

[SamuraiCatJB]

There are also more articles on why dual core: http://www.neoseeker.com/Articles/Hardware/Reports/amd_dualcore/index.html

This is not a makeshift technology gap filler, this is where things will go. The laws of physics said that it will be more difficult to make smaller/faster chips. So now they are growing wider with multiple cores. There will still be wafer-etch shrinkage, but it will be slowed by physics -- in fact it has already been slowed over the last few years.

The question then comes... what comes after silicon? :) sooner or later growing outward will no longer be efficient... any bets?

[JackAxe]

Behold the future, in alllll of its glory;

http://www.bainne.com/fun/intel386dx.jpg

<]=)