An early look at Phenom dual / tri core performance

[yomamafor1]

Ding Ding Ding....we have a winner. Congrats to Yomamafo1 for calling it, here:
http://www.tomshardware.com/forum/page-246375_28_160.html
Posted on 11-20-2007 at 04:30:03 AM
Good call Yomama [:turpit:2]

Thank you thank you. I couldn't have been standing here, accepting this trophy, without the help of others. I want to thank my Mama, my friends, and my dealer.

But don't pop open that champaign just yet... let's see how long before BM made all of the arguments I predicted...

I think.... soon

 

[yomamafor1]

Yomamafor1, will you marry me?

How close do you resemble this lady?

If you're about 1% similar to this, then probably not.

 

[turpit]

You can call that AMD's Achilles's heel. Back in the K8 days, it is AMD's manufacturing capacity that strained AMD's potential into becoming a much bigger company, not Intel's anti-competitive behavior. When AMD took its first step in expansion (ATI buyout, NY fab), BOOM, C2D debuted.

Strained manufacturing capacity might also be the sole reason why AMD continues to wither. If AMD has NY fabs up and running by this year, they could supply enough X2 chips with lower cost, without jeopardizing K10's production. This means Barcelona wouldn't be in such low volume, and further bring in desperately needed revenue and profit for AMD. Its already been two months since Barcelona launched, and very few OEMs have it.

To simply put, it isn't AMD's own product that brought AMD down, but its poor executive decision in not expanding at the right time that crippled AMD. In Rahul Sood's words, "AMD's execution this year has been nothing short of horrible."

No point in bringing that up and falsley raising hopes, fanboy or normal human alike. It takes longer to build and setup a fab (2 1/2 years was the number I believe is saw) than it does to develop a new core Uarch.

And, contrary to popular belief the NY fab is not now nor ever was a done deal. While AMD publically 'confirmed' plans for the fab, back in June of 2006, no binding contracts were ever signed, and as late as this summer(2007)Dresden was still making proposals/offering deals to be the site for the new fab.

In May of this year, at a Saratoga Economic Development Corp.'s dinner meeting to discuss the NY fab, AMD assured that group (SEDC) that they (AMD) had a need for the facility, but it was a matter of "timing". At that meeting, one of the "timing" issues the boys from AMD discussed was the "timing" to decide on the "timeline" for the project. The "timeline" decisions were "timed" to occur....now. 2H07. In otherwords...physical progress = nil. They havent even broken gound, let alone submitted construction/building plans to NYS. Sorry folks.

Also, dont forget the whole 'going fabless' thing.....though as late as this month, that was reported to be a misunderstood 200 mm wafer equipment sale to TMSC that blossomed out of proportion....at least according to fabtech.

Then theres the whole 'asset lite' concept that is married to the fabless rumours, which was also addressed at the May meeting.

In short, AMD is still at least 3 years away from another productive fab, anywhere, and thus any significant capacity increase...unless of course they buy one of Intels old fabs or TMSC devotes itself soley to AMD

(If you were TMSC, or any contract foundry for that matter, would you put all of your eggs in the AMD basket right now? )

Interestingly, at that same May meeting, the good ol' boys from AMD said AMD would be interested in opening an additional 1 or 2 plants in NY.
Now that sounds more like the AMD of 2007 and Hector/Henri!!! Hole not big enough? Open your mouth...Dig deeper!!!
Of course, when the government is subsidisizing you to the tune of $833000 for every $35000 per year job created courtesy of the taxpayers, who cares???, its not your money....spend all you want.

 

[caqde]

I don't get the fascination with scaling. So even if the AMD processors scale better, wouldn't you care more about actual performance and the fact that the Intel processors still kick AMD processors any day of the week and eat them for breakfast?

Quad core market is just plain going to be owned by Intel.

Dual core is up for grabs in my opinion, especially if AMD gets out some high clocked K10 duals, that would be very interesting.


AMD needs to take advantage of the dual core market before software most becomes highly threaded, because like I said, QUAD = INTEL, hands down.

sighhhhh.... I was only bringing up that it did scale better... Heck even the numbers I gave didn't show that big of a margin... Well besides the 20% one.. But understand with a highly threaded program the scaling could tip the scales towards AMD...

Personally I don't get into fascinations and I was trying to sway the conversation away from bashing onto something more interesting that the article showed.. Even though most of us can't read the thing.. Either way I am still wondering why No one has tried to find out what kind of difference AMD's new power management makes in partial loads.... I saw one that brought it up but they couldn't get Cool and Quiet to work on that board even so it was still ignored.

And besides that right now I am more interested in seeing these companies reduce the amount of power a computer uses when doing and not doing things instead of brute performance (Yes I know Penryn is faster). Which makes me interested in products like the RS780 and the HD38X0 series of videocards along with Phenom (when it matures[The PM features/45nm]). Because they will reduce consumption when I am actually doing things and when I am not. But then again this is my current opinion as to the way things stand with me..

 

[caqde]

No point in bringing that up and falsley raising hopes, fanboy or normal human alike. It takes longer to build and setup a fab (2 1/2 years was the number I believe is saw) than it does to develop a new core Uarch.

And, contrary to popular belief the NY fab is not now nor ever was a done deal. While AMD publically 'confirmed' plans for the fab, back in June of 2006, no binding contracts were ever signed, and as late as this summer(2007)Dresden was still making proposals/offering deals to be the site for the new fab.

In May of this year, at a Saratoga Economic Development Corp.'s dinner meeting to discuss the NY fab, AMD assured that group (SEDC) that they (AMD) had a need for the facility, but it was a matter of "timing". At that meeting, one of the "timing" issues the boys from AMD discussed was the "timing" to decide on the "timeline" for the project. The "timeline" decisions were "timed" to occur....now. 2H07. In otherwords...physical progress = nil. They havent even broken gound, let alone submitted construction/building plans to NYS. Sorry folks.

Also, dont forget the whole 'going fabless' thing.....though as late as this month, that was reported to be a misunderstood 200 mm wafer equipment sale to TMSC that blossomed out of proportion....at least according to fabtech.

Then theres the whole 'asset lite' concept that is married to the fabless rumours, which was also addressed at the May meeting.

In short, AMD is still at least 3 years away from another productive fab, anywhere, and thus any significant capacity increase...unless of course they buy one of Intels old fabs or TMSC devotes itself soley to AMD

(If you were TMSC, or any contract foundry for that matter, would you put all of your eggs in the AMD basket right now? )

Interestingly, at that same May meeting, the good ol' boys from AMD said AMD would be interested in opening an additional 1 or 2 plants in NY.
Now that sounds more like the AMD of 2007 and Hector/Henri!!! Hole not big enough? Open your mouth...Dig deeper!!!
Of course, when the government is subsidisizing you to the tune of $833000 for every $35000 per year job created courtesy of the taxpayers, who cares???, its not your money....spend all you want.

AMD needs that FAB badly.. Of course it will hurt now, but think of it this way. Why is AMD not using 45nm? Why does it take so long for them to switch production? From what I can tell AMD needs the new FAB in order for them to be able to switch production lines without shutting down so much of their production. Lets say AMD has 3 Fabs if they switch one to 65nm from 90nm they have to shut down (<=) ~33% of their production capacity for the during of the time it takes to switch to 65nm, but if they have 4 Fabs they only need to shut down (<=) ~25% of their production capacity in order to make the switch to this new process. It hurts AMD alot right now for them to switch and test new production processes, but with a new Fab it won't hurt as much as it used too.

Oh and if I remember my numbers correctly it takes 5 years to make a new Uarch not the short time you are giving.. The new cycle they are doing now is shorter because they are not changing the Architecture as much as they used to so they can give us the benefit of a new Architecture without the large delays we were used too. Although these new Architectures will not be as big of a change as we were used to seeing.. Of course for Intel that might be a good thing (P4 )... BTW the Core 2 series is a product of this new line of thought (P3 -> Pentium M -> Core -> Core Duo -> Core 2 Duo)

Sorry for the Double post (I think..)

 

[spentshells]

hmm interesting...whiny but interesting

 

[turpit]

AMD needs that FAB badly.. Of course it will hurt now, but think of it this way.

Excuse me, why are you saying this to me??? Do you think this is some form of revelation??

Heres a revelation for you.....AMD doesnt need this fab now...they needed it, and at least 2 more on top of what they have, at full manufacturing capacity, 2 years ago

But saying they need it now, next week, or two years ago doesnt change the fact that ground hasnt been broken. Clicking your heels together and saying "I wish AMD had another fab, I wish AMD had another fab, I wish AMD had another fab" isnt going to make it happen. It actually has to be built. And its going to be late 2009, at the very earliest, before it can produce anything...and thats if its not cancelled and if constuction starts tomorrow. Every tried to dig a hole or lay a foundation in Upstate NY?? In the winter?? After the ground has frozen?? No? Well guess what...its december now...the ground hasnt frozen yet, but it will start any day, and the earliest construction could reasonably start at this point (without paying a fortune in premiums/adverse environment penalties) is late february...and thats only if its an easy winter.

Why is AMD not using 45nm? Why does it take so long for them to switch production? From what I can tell AMD needs the new FAB in order for them to be able to switch production lines without shutting down so much of their production.

Before AMD has to worry about shuting down a fab to retool for a different node, they first have to actually develope the process technology for that node. Much like the fab, one can not merely click their heels three times and say "I wish AMD had 45nm, I wish AMD had 45nm, I wish AMD had 45nm". Now, supposedly AMD already has 45nm in qualification....which means exactly nothing. They could have some equipment in place, they could have produced some 45nm wafers, maybe some 45nm ram, heck, they may have an actual functioning 45nm ES. It was reported they had a fuctional 45nm chip last year, but given the source reporting that and the 'quality' of their 65nm process, I find that report dubious at best, In short, no one outside AMD knows the state of their 45nm process, because beyond saying that they are in "qualification", and a few roadmaps notes, AMD hasnt said too much about their 45nm process. Which, considering the fiascos their statements have created the past 23 months is a good thing. But considering their success, or rather lack of at 65nm, worring about 45nm production is putting the cart before the horse. Unless, by some minor miracle, they have admited to themselves that SOI isnt going to make it at 45nm, have abandoned it, secretly developed a new process, and are going to abandon 65nmSOI the instant they have 45nm xxxx qualified....a whole lot of ifs and the long route to get to this....AMD is in no position to worry about full scale 45nm production right now. They seem to be having enough of both yield and design problems at 65nm to keep their plates full for the moment.

Lets say AMD has 3 Fabs if they switch one to 65nm from 90nm they have to shut down (<=) ~33% of their production capacity for the during of the time it takes to switch to 65nm, but if they have 4 Fabs they only need to shut down (<=) ~25% of their production capacity in order to make the switch to this new process. It hurts AMD alot right now for them to switch and test new production processes, but with a new Fab it won't hurt as much as it used too.

Oh hell, lets stick with accuracy in figures and say that AMD has 2 fabs and contracts the rest f its needs out to a foundry, shall we? But dont feel bad...in 2010, AMD will have 3 fabs, at the completion of the of Fab30s conversion to Fab38 in 2008 + the new Dresden expansion finishing in 2010. Hell, by 2010 it may have 4 fabs if the NY deal isnt cancelled. Of course, it may still only have 2, or it may have none as part of its changing 'ass-ette lite' stratedgy, or it may have filed CH11 and no longer exist.

As for the primer on manufacturing processes....uh gee....thanks

Oh and if I remember my numbers correctly it takes 5 years to make a new Uarch not the short time you are giving.. The new cycle they are doing now is shorter because they are not changing the Architecture as much as they used to so they can give us the benefit of a new Architecture without the large delays we were used too. Although these new Architectures will not be as big of a change as we were used to seeing.. Of course for Intel that might be a good thing (P4 )... BTW the Core 2 series is a product of this new line of thought (P3 -> Pentium M -> Core -> Core Duo -> Core 2 Duo)

Yeah, uhhhh, I suggest you go back and check your times, as well as you "new line of thought" concerning Intels Uarch progression...they are wrong enough to fall into the catagory "grossly" wrong.

 

[yomamafor1]

No point in bringing that up and falsley raising hopes, fanboy or normal human alike. It takes longer to build and setup a fab (2 1/2 years was the number I believe is saw) than it does to develop a new core Uarch.

And, contrary to popular belief the NY fab is not now nor ever was a done deal. While AMD publically 'confirmed' plans for the fab, back in June of 2006, no binding contracts were ever signed, and as late as this summer(2007)Dresden was still making proposals/offering deals to be the site for the new fab.

In May of this year, at a Saratoga Economic Development Corp.'s dinner meeting to discuss the NY fab, AMD assured that group (SEDC) that they (AMD) had a need for the facility, but it was a matter of "timing". At that meeting, one of the "timing" issues the boys from AMD discussed was the "timing" to decide on the "timeline" for the project. The "timeline" decisions were "timed" to occur....now. 2H07. In otherwords...physical progress = nil. They havent even broken gound, let alone submitted construction/building plans to NYS. Sorry folks.

Yes, you're right. AMD had the chance to slightly level the playing field by building a FAB back in 2003~2004. Of course, AMD didn't foresee it will become the top dog for the next 2~3 years, thus not making the decision to expand another fab. That's horrible execution on their part. They simply assumed Intel = Crapburst forever, which was not the case. If AMD had made the right decision to expand back in its glory days, they wouldn't be in such dire situation today.

After reading your post, I agree with you. AMD doesn't need the FAB now. They need to improve their process node to take full advantage of the existing FABs. New FABs would only mean new debt to them.

Now, AMD's practically being pushed into a corner, and had to whistle to any tune Intel put out. I'm not sure how K10 will play out, but from the looks of it, it is hardly a success. Spider platform? How is it going to attract your audience if one of the critical components is not performing up to par? Opteron? How are you going to compete if the only advantage you have left will be stripped off you by next year?

People always say, "AMD will survive. AMD had been in much worse condition, and they made it through. They will survive this time". But the question is, how?

Also, dont forget the whole 'going fabless' thing.....though as late as this month, that was reported to be a misunderstood 200 mm wafer equipment sale to TMSC that blossomed out of proportion....at least according to fabtech.

Then theres the whole 'asset lite' concept that is married to the fabless rumours, which was also addressed at the May meeting.

In short, AMD is still at least 3 years away from another productive fab, anywhere, and thus any significant capacity increase...unless of course they buy one of Intels old fabs or TMSC devotes itself soley to AMD

(If you were TMSC, or any contract foundry for that matter, would you put all of your eggs in the AMD basket right now? )

Interestingly, at that same May meeting, the good ol' boys from AMD said AMD would be interested in opening an additional 1 or 2 plants in NY.

I don't quite get their "asset-lite" strategy. They desperately need more manufacturing capacity, and advanced manufacturing node. How can they become "asset-lite", if they need more asset?

 

[yomamafor1]

sighhhhh.... I was only bringing up that it did scale better... Heck even the numbers I gave didn't show that big of a margin... Well besides the 20% one.. But understand with a highly threaded program the scaling could tip the scales towards AMD...

Not to be an @$$ here, but AMD probably only enjoys 10% of scaling advantage, on very limited programs that really take use of multi-threading.

Intel has at least 33% in clockspeed advantage with current 65nm products, and if they like it, they can push it to 100% with 45nm.

EDIT: assuming K10 has the same amount of IPC as Core 2 does, which in reality, K10 is a little behind.

 

[yomamafor1]

Why is AMD not using 45nm? Why does it take so long for them to switch production?

Because AMD doesn't develop them in-house, and have to rely on IBM for such technology? In other words, this means whatever IBM develops, AMD has to use them. They can tweak the process a little bit, but they certainly cannot remove a key piece of a process (e.g. SOI).

I believe AMD is anxious for 45nm, with HK/MG, not with SOI. SOI is giving them heck of problems. But if IBM says SOI for 45nm, AMD has to say SOI for 45nm. If IBM says no HK/MG until 32nm, what can AMD do?

 

[easyg]

I don't get the fascination with scaling. So even if the AMD processors scale better, wouldn't you care more about actual performance and the fact that the Intel processors still kick AMD processors any day of the week and eat them for breakfast?

Quad core market is just plain going to be owned by Intel.

Dual core is up for grabs in my opinion, especially if AMD gets out some high clocked K10 duals, that would be very interesting.


AMD needs to take advantage of the dual core market before software most becomes highly threaded, because like I said, QUAD = INTEL, hands down.

It's funny that you should say this, cause my impression is that AMD will be much further behind in dc than in qc. Phenom x4 has outstanding multicore efficiency relative to Yorkfield, and yet it still falls a little short of the Intel stuff in terms of overall performance. In dual core, Phenom's one big advantage (multi-core scaling) ceases to exist as C2D is already native dual core. I really think all signs point to Wolfdale (45nm) beating the bejeezus out of Phenom x2 at the same frequency. But they won't run at the same frequency. Wolfdale will clock much higher (given everything we've seen so far).

 

[Slobogob]

It seems it is possible to deactivate cores on the regular phenom processors. As a result there are some german sites with benchmarks comparing the performance of 4x,3x and 2x phenoms with intels offerings. I forgot the link but i suppose it can be googled.

 

[elbert]

http://www.erenumerique.fr/test_processeur_amd_phenom_x4_9600_drole_de_rel_ve_-art-1804-11.html

The article is in French, but the benchmarks speak for themselves. It also shows the differences between ganged and unganged memory mode, something most articles have failed to mention, so kudos to the author for the added testing. It also serves as a good yardstick for measuring clock for clock performance between K8 and K10.

In a nutshell, a Phenom X3 @ 2.3GHz is ~= to E6750/E6850 in multithreaded benchmarks, with lower single threaded performance obviously. The Phenom X2 @ 2.3GHz is a good match for a E6400.

What is striking to me is that clock for clock, Phenom is not that much faster than K8. Improvements range between 5 - 10% generally. I was under the impression it was a lot higher than that...

These results make it very obvious why AMD is delaying dual/tri core Phenom until Q2 2008. Unless AMD can get the clockspeed up, a Phenom X3 is no better than current C2Ds, and they may as well forget about Phenom X2 until it reaches 3GHz, otherwise it will be outperformed by an X2 6400+.

I do agree with most of what you said but from a marketing stand point it looks like AMD is coming out ahead. Performance wise they now have CPU's that beat or compete against any CPU below the Q6600. This without a single benchmark from the 4X4 which could spell trouble for Intels top CPU's even at a low of 2.3GHz.

Over all I would call the Phenom X3 2.3GHz a equal to the E6750 as it does manage to beat even the E6850 on a benchmark or 2. Lets face it the dual cores are as optimized as they will get but the X3 will see performance boosts from optimized software of the next year. The X3's may come out ahead after a few months to even beat the E6850 so its to early to tell were its performance will end. Its kind of like when the Kentsfield came out it was well over 6 months before average user could see a performance boost between the Q6600 and the E6850/E6750.

 

[epsilon84]

caqde does bring up a good point in regards to scaling, yes Phenom, being a native quad core, does scale better with more threads, but it is merely a game of playing catch up.

Most reviews put the C2Q about ~10% faster per clock than Phenom X4, compared with the ~17% between C2D and Phenom X2. At this rate of scaling, Phenom may theoretically come out on top once we hit Octocore, but Core 2 would have long been superceded by Nehalem by then, and the IMC advantage is lost.

Elbert, I don't really think optimization will help X3 at all, since software is either single, dual, or massively multithreaded. There is no tri threaded software as far as I know, but programming experts please feel free to correct me if I'm wrong.

The instance where X3 will obviously shine against dual cores is massively multithreaded software like encoding and rendering, where it has 3 threads running as opposed to 2 on a dual core. These benchmarks already show that it takes a C2D clocked 15 - 30% higher in order to compete, and no optimization is going to help the X3 when all cores are already fully utilized by the software.

 

[thepinkpanther]

i usto hate intel because they made the Pentium 4 look way better than it actually was and they advertised them too much. Amd was better back in the day but only gamers and technogeeks knew about them. Nowadays intel is better...while AMD being "theoretically" better with its "native" quad core. Not only is AMD now inferior but they also stink at lying.

 

[ro3dog]

Man you guys get real hostile in here ,so many "burn AMD to the stake"atittude.I regardless to whats being said AMD is burning alot on all ends cpu and gpu.I belive all in this forum are gamer and looking for fps plus beauty in games.Does it reall matter that Intel has the faster processor.If AMDs cpu is 13% slower than Intels cpu of same class how much does it matter.AMD is no longer just cpu,we go through more gpu's than cpu's and since midrange cpu's and gpu's are the same price money will be made by this company.I might be wrong but there gpu's went from 65nm to 55nm to 45nm(Q2 08) in less than a year,not to mention 1 to multi-core(R600 to R700).AMD is still standing folks and I believe they already know the short comings of Phenom to turn ship.If this spider do outperform Intel and Nvidia combined,then who wins

 

[ATIWOLF]

It seems it is possible to deactivate cores on the regular phenom processors. As a result there are some german sites with benchmarks comparing the performance of 4x,3x and 2x phenoms with intels offerings. I forgot the link but i suppose it can be googled.

On not only possible but a reality:
http://www.tomshardware.com/2007/11/19/the_spider_weaves_its_web/page2.html

In an interview in Warsaw, AMD now officially confirmed that the tri-core models are indeed quad-cores with one deactivated core.

😛 AMD is indeed offering circumcised Phenoms and its gonna be awesome
Its going to be awkward when these thing go mainstream and entry-level. Dell are gonna fly off the shelves when they start advertising with "TRi-core" and "Spider"

 

[elbert]

Elbert, I don't really think optimization will help X3 at all, since software is either single, dual, or massively multithreaded. There is no tri threaded software as far as I know, but programming experts please feel free to correct me if I'm wrong.

The instance where X3 will obviously shine against dual cores is massively multithreaded software like encoding and rendering, where it has 3 threads running as opposed to 2 on a dual core. These benchmarks already show that it takes a C2D clocked 15 - 30% higher in order to compete, and no optimization is going to help the X3 when all cores are already fully utilized by the software.

Not all multithreaded software is optimized for 3 threads as most create even numbers of threads as well as this leaves out any advantages with SSE4A. Intels Penryn having SSE4 is a moot point but C2D and C2Q can be questioned.

 

[epsilon84]

Elbert... let me just refer you to caqde's core scaling analysis...

AMD Dual to Tri = AVG Increase = 26%
AMD Tri to Quad = AVG Increase = 20%
AMD Dual to Quad = AVG Increase = 54%

X3 is already only 20% behind X4 in multithreaded applications. This indicates all 3 cores are being fully utilized, and the software is spawning 3 threads. In fact, X3 is running at a higher efficiency per core than X4 in multithreaded software.

 

[caqde]

Yeah, uhhhh, I suggest you go back and check your times, as well as you "new line of thought" concerning Intels Uarch progression...they are wrong enough to fall into the catagory "grossly" wrong.

?? Umm yes it does take ~5 years to make a new Uarch Intel had 2 development teams likely more now but 2 were working on the desktop Architecture's. To make you understand how this works. One team works on the new Architecture the other works on a revision. Say Intel releases Architecture X in 2005 and then releases Architecture X Rev 2 in 2007 Their Next Architecures would come out in ~2009-2010 and ~2011-2012 respectively They are obviously keeping a tight schedule now keeping the Architecture teams and likely have 3 maybe even 4 development teams working on the Core MicroArchitecture.

http://intel-insider.blogspot.com/2007/01/intel-x86.html read the 486 paragraph... Sounds accurate.. (wish I had a better source..)
http://www.eetimes.com/story/OEG20010313S0113 This is when the Cell was being developed It took 4 years to be accurate, but they got ready for 5 years.

It takes a long time to develop something this complex. Computers help us keep track of what we have done and allow us to visually see what we are doing, but it doesn't do the work for us...

Elbert, I don't really think optimization will help X3 at all, since software is either single, dual, or massively multithreaded. There is no tri threaded software as far as I know, but programming experts please feel free to correct me if I'm wrong.

I don't know why developers even bring this up. I sure don't think of my software being threaded for x number of processors I could develop it that way, but I am more likely to thread it as much as I can instead.

 

[TechnologyCoordinator]

It's funny that you should say this, cause my impression is that AMD will be much further behind in dc than in qc. Phenom x4 has outstanding multicore efficiency relative to Yorkfield, and yet it still falls a little short of the Intel stuff in terms of overall performance. In dual core, Phenom's one big advantage (multi-core scaling) ceases to exist as C2D is already native dual core. I really think all signs point to Wolfdale (45nm) beating the bejeezus out of Phenom x2 at the same frequency. But they won't run at the same frequency. Wolfdale will clock much higher (given everything we've seen so far).

I don't feel that way. Reason is:

Intel's best quads are really the best duals. So basically Intel only needs to be able to make a great dual-core. If they can do that, they can make a great quad core. Intel has the ability to pair up sets of cores or to sell them as dual cores. Intel has tons of flexibility and this in part gives them part of the performance advantage.

Now AMD on the other hand, in order to make a great quad core actually has to get all four cores working great. If not the chip is only as good as its weakest core. If three cores clock up to 3ghz but the fourth will only go up to 2.4ghz, then they can only sell it as a 2.4ghz quad core.

Now if AMD was only trying to make two cores the likelihood of hitting high clock speeds is MUCH greater. They only have to get two cores right.

So I feel AMD can probably make a much high clock Phenom DUAL core that will perform better at tasks that are not multi-threaded, such as most games out today. I have no bleeping clue why AMD doesn't have a dual-core K10 out right now, it just doesn't make sense with the current state of the manufacturing, a dual core would be much easier to make and would perform much closer to other Intel dual-cores than X4's perform to Core 2 Quads.

I'm not explaining it well, but I try!

 

[elbert]

I don't feel that way. Reason is:

Intel's best quads are really the best duals. So basically Intel only needs to be able to make a great dual-core. If they can do that, they can make a great quad core. Intel has the ability to pair up sets of cores or to sell them as dual cores. Intel has tons of flexibility and this in part gives them part of the performance advantage.

Now AMD on the other hand, in order to make a great quad core actually has to get all four cores working great. If not the chip is only as good as its weakest core. If three cores clock up to 3ghz but the fourth will only go up to 2.4ghz, then they can only sell it as a 2.4ghz quad core.

Now if AMD was only trying to make two cores the likelihood of hitting high clock speeds is MUCH greater. They only have to get two cores right.

So I feel AMD can probably make a much high clock Phenom DUAL core that will perform better at tasks that are not multi-threaded, such as most games out today. I have no bleeping clue why AMD doesn't have a dual-core K10 out right now, it just doesn't make sense with the current state of the manufacturing, a dual core would be much easier to make and would perform much closer to other Intel dual-cores than X4's perform to Core 2 Quads.

I'm not explaining it well, but I try!

True but the tricky part comes in if AMD does 2 quad cores splice like kentsfield. Putting 2 say dual or quads together is easy but putting 4 duals together Intel has never done. Come this summer it would be easy for AMD to come with a Octal core but Intel would have to wait for a full quad core CPU. AMD may be waiting on the K10 duals to see if its even worth coming to market with a dual core. At current prices its worth little R&D to just replace the X2's already in place.

Come summer a Octal, quad, tri phenoms, with 90nm X2's, and brisbane could be a good marking approach with little R&D costs in the low end. My fellow techno geeks we are in the multicore era. GHz will always take a back seat to more cores.

 

[BaronMatrix]

I thought you said the L3 bug affected at least 20%?

I imagine that if the L3 bug IS affecting general perf (which I think it is) we will see at least 20% and maybe 30%.

How come it becomes "grow to 20%" all of the sudden?

Look Baron, no one wants to prove AMD buyers wrong. AMD buyers know what they got when they made that purchasing decision. We only want to prove YOU wrong. Almost all of your claims are based on that random number generator in your head. NOTHING solid.

And I thought TLB errata lead to BSOD, not performance hit?

I quoted "I imagine that if the L3 bug IS affecting general perf (which I think it is) we will see at least 20% and maybe 30%." There is nothing random about basing this on the excellent David Kanter K10 analysis over at RealWorld tech.

Anand says it's around 15% faster than K8. Some things like HL2 show almost 30%. The fact that a 3DMark run crashed says MAYBE the L3 TLB errata is limiting performance.

Anand has a new review up with Penryn and Barcelona. There are a "variety of workloads that show ~40% over even Penryn." And that's with B2 NOT B3.

 

[elbert]

I don't know why developers even bring this up. I sure don't think of my software being threaded for x number of processors I could develop it that way, but I am more likely to thread it as much as I can instead.

Thats true but its kind of like the chicken and the egg question. Which came first? Programmers have not always had X number of cores. Those programs used on servers and desktops have an advantage as most have the core count first which makes testing and optimizing easy. Looking at the 3dmarks 2005 scores and comparing to the 2006 you can see right away that 2005 was programmed using a dual core CPU or maybe even a dual socket single core. Gaming programmers are really behind in this as they don't even have to program for the dual socket mother boards.

Game programmers are moving to programming on 2 and 4 socket rigs. A game only optimized for the current number of cores will not last long as the amount of time a game stresses top systems the harder it is to replace. Crysis for example may be hard hit by this as only shows a good performance jump moving on dual cores. Once a game comes to market using 4 or 8 cores crysis is dead unless they put major amounts of time into reprogramming as valve with half-life2 is currently in the process of doing.

 

[epsilon84]

True but the tricky part comes in if AMD does 2 quad cores splice like kentsfield. Putting 2 say dual or quads together is easy but putting 4 duals together Intel has never done. Come this summer it would be easy for AMD to come with a Octal core but Intel would have to wait for a full quad core CPU. AMD may be waiting on the K10 duals to see if its even worth coming to market with a dual core. At current prices its worth little R&D to just replace the X2's already in place.

Come summer a Octal, quad, tri phenoms, with 90nm X2's, and brisbane could be a good marking approach with little R&D costs in the low end. My fellow techno geeks we are in the multicore era. GHz will always take a back seat to more cores.

Easy?! Yeah, just slap two 100W 280mm^2 dice in a MCM and call it a day... no, sorry, it doesn't work like that in the real world.

65nm runs FAR too hot, and the die is FAR too big, to support octo core in any form. MCM is also more difficult to implement on an IMC based architecture, and do you honestly think AMD will trumpet MCM just 6 months after hyping up the advantages of a 'true native' quad core design over MCM?!

It just ain't gonna happen.

AMD should focus their R&D on getting 65nm QUADS right, as well as 45nm design&research. NOT octocore in 6 - 9 months, that is just an insane and doomed strategy, and no offence Elbert, but I'm glad you're not at the helm of AMD right now. Hector isn't the greatest but even he wouldn't try something as ludicrous and risky as this.

PS. Here is a funny and well known image with good ol' Henri Richard, but it also shows how big the K10 die is.

I took the liberty to graphically illustrate what you are suggesting:

Now is it just me or do you see something wrong with that CPU on the left?! :lol:

I'm no electrical engineer but here are some of my concerns
1) I'm sure you're not supposed to have the edge of the die THAT close to the PCB edge and
2) It eats into the area of where the capacitors are (is that the correct term?) around the die
3) Where is the heatspreader going to go?
4) How the **** are you going to cool such a monstrosity?! That is 95W per die, or 190W all up. Prescott on steroids?