Official AMD Barcelona Thread

[VulgarHandle]

besides, won't all future Opterons based on K10 be quad-cores? I doubt the "Opterons" in this test were K10

No, you don't understand.

It is 2x performance. The task is to process n pixels. K10 processing 4000+(it is said more than 4000, not exact number. So I assume 4400 pixels/s according to the times) pixels per second and finishes the processing for 56 seconds. For the same task the K8 needed 112 seconds because it was processing 2200 pixels/second.

It is like you have two cars on a drag race and the faster car finishes the race after 56 seconds with average speed of 200km/h. The slower car finishes the race after 112 seconds with average speed of 100km/h. The faster car is 2 times faster than the slower car, not 4 times.

Got it?

the drag strip is not a good example, because they both have 1 engine, but what you are saying is that one car has a better motor, so it not only finishes in a faster time, but accelerates faster which helps it get the better time

so, is the ~4000 pixels/s due to more cores, or better cores?

is the less time due to more cores, or better cores?

i assume the more pixels/s and less time are due to both more cores and better cores

if you were to pit 8 K8 cores against 16 K8 cores, would it still perform more pixels/s at a ratio of 2:1 AND less time at a ratio of 2:1 like this?

[PetNorth]

red, Have you heard in the video to Randy Allen calling the quad core as Barcelona? I'm not. He says dual core solution, and quad core solution. Why do you give credit to written sources (Inq between them), and not to Randy Allen mouth?

VulgarHandle, there will be dual and quad Opteron K10.

[red]

I thought I heard him call the dual solution Opteron and the quad Barcelona... His statement can go both ways but uberpulse clearly says they used a dual core solution shipping today, Inq says Barcelona machine requiring a new BIOS. And I have a birdie telling me it's 16 K10 vs 8 K8. The "poor" scaling could mean
a) News are wrong and it is dual K10 vs quad K10 (decent scaling of 80&#37
b) K10 needs HT3
c) If that demo was with HT3, HT3 is insufficient
d) POV-Ray not a good indication of K10 improvements
e) K10 is a dud

[Sumanji]

I guess this has been discussed already, but what do you make of the guy's analysis:

AMD's official Barcelona show:
http://www.youtube.com/watch?v=VGiv9Dtrc5Q

In POV-Ray, 4 Barcelona (16cores total) got about 4000 pixs /sec.
while the 4 Opteron DC(8 cores total) in the same frequency got about
2200 pixs /sec.

Compare with the test provided by xbitlabs:
http://www.xbitlabs.com/images/cpu/intel-v8/povray.png
http://www.xbitlabs.com/articles/cpu..._11.html#sect0

Two X6365(Quad Core Xeon @3.0GHz), 8 cores total can do 4677 pix/sec.
And two FX74 (Dual core K8 @ 3G, 4Cores total) can do 2020.

combine all the facts above together, we can find out that the test
systems in AMD's show are running at about 2GHz. At these frequency,
the 16 cores Barcelona system even can't beat current 8core Xeon
system.

Remember the Quad core Barcelona will top at 2.5GHz! And Intel's
Harpertown in on the way.

....

[VulgarHandle]

I guess this has been discussed already, but what do you make of the guy's analysis:



....

the problem with that is that all his analysis points to the scaling of POVray to be completely dependent on core speed, not number of cores

so in his analysis, 8 3.0ghz intel beats 16 2.0ghz barcelona, which tells us nothing

based on his analysis we would need either 2.0ghz intel's or 3.0ghz barcelona's....

[red]

I don't think they were running the same POV-Ray bench. There are different scenes and settings and whatnot.

[DeepThought86]

I really wonder what's wrong with the idiot "journalists" and "analysts" who didn't have the sense to ask a simple question to clarify. It just boggles the mind how stupid and ignorant these people are.

I guess this has been discussed already, but what do you make of the guy's analysis:



....

[informal]

Exactly red.The previous results (from Xbit etc.) cannot be compared to these results.To my knowledge,they used 4x opteron HE dual core that ran at 2.2Ghz.

[red]

http://www.dailytech.com/article.aspx?newsid=5992
This conclusion might be a bit off because of recent FUD about better steppings but DailyTech says only 1.9/2.0GHz HE parts for K10. And the Youtube said equal frequency.

[accord99]

I don't think they were running the same POV-Ray bench. There are different scenes and settings and whatnot.

The photo shows the default benchmark image used in Povray 3.7. The image is 512x512, which matches up with the score; 4000+ pixels/s at 62s.

[red]

Core 2 QX6700 gets ~2700. 8 K8 cores get only 2200 and 16 K10 get 4000? lol

[nemrod]

I really wonder what's wrong with the idiot "journalists" and "analysts" who didn't have the sense to ask a simple question to clarify. It just boggles the mind how stupid and ignorant these people are.

Perhaps because AMD doesn't want to answer clearly because there barcelona is buggy...
It 's fantastic, since months we have only bull, AMD doesn't want to provide something else than "estimate" or look this is faster and this will be not our faster cpu... And there is still blind people...
In fact, this bench is 1 barcelona with 3 cores desactivates, underclock at 100 Hz versus 32 opteron K8 overclock at 5 GHz...

Don't you see that they try deseperatly to show something but have nothing to show...

[accord99]

Core 2 QX6700 gets ~2700. 8 K8 cores get only 2200 and 16 K10 get 4000? lol

They were using 68W models, so somewhere around 1.8GHz.

[red]

http://www.lostcircuits.com/cpu/amd_quadfx/11.shtml
From here, Opteron 280 2.4GHz to FX74 3GHz is a 25% difference, which is directly reflected in score...

Lowest clocked 4 socket Opteron HE is 8212 with 2GHz...2/2.4=0.83. 0.83*1711=1426, or 356 per core for a 2GHz deduced POVray score.

Score would be 2900 for 8 K8 2GHz cores or 5700 for 16 K8 2GHz. Nah.

[XeRo]

http://www.hardocp.com/news.html?new...VzaWFzdCwsLDE=

Wouldn't middle of the year be like next month?

[nemrod]

http://www.hardocp.com/news.html?new...VzaWFzdCwsLDE=

Wouldn't middle of the year be like next month?

This is just 65nm facilities...

About barcelona and phenom:

With those high expectations in mind, AMD is preparing to launch its quad-core desktop processors, which will go by the name "Phenom," later this year and just after the company unveils its quad-core Opteron processor "Barcelona" for servers.


And again something look bad with barcelona:

In addition, Sonderman said AMD will start delivering 45-nanometer processors by the middle of 2008 to compete with Intel's "Penryn" family of 45-nanometer chips, which are scheduled to hit the market late in 2007.

[red]

Wow. Comparing 2007 Barcelona with 2006 Clovertown now comparing 2008 45nm Barcelona with 2007 Penryn?

[gOJDO]

No products, no comparison. BTW, if AMD continue this trend of delaying products, then their 45nm K10 will compete against 32nm Nehaleem. This thread is tiresome and I am tired of waiting for K10.

[frankR]

AMD will always be behind Intel in fabs. They just don't have the money to push to the next process as quickly as Intel. Let Intel spend all the money to work out the bugs, then AMD will follow.

[awdrifter]

Hopefully the K10 won't turn out like the R600. That would suck.

[red]

http://www.uberpulse.com/us/2007/05/...e_you_ship.php
In other news, Intel scores 4933 with 8 cores compared to "just over 4000" from the 16 Barcelona cores. This demo didn't have much to do with performance as much as it had to do with... Barcelona existing

[informal]

red,follow this link :http://www.aceshardware.com/forums/r...1703&forumid=1

It shows that seemingly "same" beta can bring diff. results.

[red]

http://www.lostcircuits.com/cpu/amd_quadfx/11.shtml
What are you trying to prove with that link? My point was that the POVray bench wasn't indicative of performance relative to Intel since obviously they could do better. Also, I don't think Intel would have done the stunt unless they knew how Barcelona performed with normal clocks in POVray. And something that has yet to be explained is why 16 K10 only scales 80% against 8 (presumably) K8 cores.

[informal]

Yes that also,but also the fact that there are different betas of the same version that yield different results.So 8 Core(intel) Vs 16 K10 is not an accurate comparison since we know nothing specific about the (sub)version AMD used and the system specs.

[adamsleath]