Official AMD Barcelona Thread

[ted3]

Saw the Cnet article earlier and came here expecting lots of posts with more info about it + the usual fan-boy quarreling =)

Had to dig myself then, QX6800 does 2613 in POV-Ray according to Intel
http://www.intel.com/performance/des...iid=qc_tl+dis3
And X6800 only 1318.

So is this one of the benchmarks where AMD is already doing better than Intel, same as was the case with Spec-FP_rate benches released earlier? No more PER CORE than what current Opterons can do, not so very impressive.

And i dont get it from the Cnet article, is this meassured per processor or per 4 processors?

comparing the performance of two four-socket servers, one using the quad-core Barcelona chip and one using a dual-core Opteron chip.

4-socket board with only 1 processor mounted, maybe thats what it means.

[accord99]

ted3, based on the rough score of 2000 pixels/s for the dual-core Opteron system, it would suggest that it would be a comparison of dual-socket systems.

http://www.2cpu.com/review.php?id=114&page=11
http://www.xbitlabs.com/articles/cpu..._11.html#sect0

[ted3]

ted3, based on the rough score of 2000 pixels/s for the dual-core Opteron system, it would suggest that it would be a comparison of dual-socket systems.

http://www.2cpu.com/review.php?id=114&page=11
http://www.xbitlabs.com/articles/cpu..._11.html#sect0

It sure looks like its dual socket then. Hope for more news on this, the CNet article is messed up.

And if it is 2 sockets, why does dual Clowertown score higher... Maybe its 2 Opterons vs 1 Barcelona.

[nemrod]

It sure looks like its dual socket then. Hope for more news on this, the CNet article is messed up.

And if it is 2 sockets, why does dual Clowertown score higher... Maybe its 2 Opterons vs 1 Barcelona.

no but:

Allen said Barcelona was not running at the fastest clock speed that will be available at launch, although he declined to specify the speeds that will be available.

look like they have probleme to show high performance... That is probably why we wait always and always for something clear...

thinking barcelona is twice as fast as K8 is rather optimistic... even too optimistic for a hardcore amd fanboy like me

You v' probably miss that even "not true" quadcore are twice as fast than dualcore on povray. (perhaps, you doesn't mean 1 vs 1?)

Had to dig myself then, QX6800 does 2613 in POV-Ray according to Intel
http://www.intel.com/performance/des...iid=qc_tl+dis3
And X6800 only 1318.

[ted3]

thinking barcelona is twice as fast as K8 is rather optimistic... even too optimistic for a hardcore amd fanboy like me

Its possible, in some benches C2D can double the score of Yonah. Noone expected it but it still happened =)

[madcho]

thinking barcelona is twice as fast as K8 is rather optimistic... even too optimistic for a hardcore amd fanboy like me

AMD said about 70% faster than K8, This coud be true

[Face]

The Opteron machine on the left was started first and the Barcelona box finished in far less time. The raw results were about 2200 pixels per second on the Opteron while the Barcelona hit "just over" 4000 on the POV Bench.

The systems it showed were identical 4S AMD development platforms running DDR2 memory and HE (65W) chips. Nothing was changed between the two platforms other than the BIOS.

So umm .. 16 cores? i'm confused now.. am I missing anything, or that the Inq is misinformed?
doesen't make sense at all..(Clovertown X5365 8 cores 4677 POV-ray according to Xbitlabs)

http://theinq.com/default.aspx?article=39756

[edit typo]

[Shintai]

AMD said about 70% faster than K8, This coud be true

70% now?

It started with 42%, then 46% then 50% and now 70%?

Sounds like the 2 eggs that became 5 chickens.

[Shintai]

So umm .. 16 cores? i'm confused now.. am I missing anything, or that the Inq is misinformed?
doesen't make sense at all..(Clovertown X5365 8 cores 4667 POV-ray according to Xbitlabs)

http://theinq.com/default.aspx?article=39756

http://www.xbitlabs.com/articles/cpu..._11.html#sect0

But they also state about 2000 for a dual dualcore FX74 setup (4 cores). And these 2000 should be with 8 K8 cores.

Sounds fishy...

[red]

I think they configured the bench differently or used some scene that most sites don't.

http://www.youtube.com/watch?v=VGiv9Dtrc5Q
AMD man claims same GHz. So 16 K10 cores are just 80% faster than 8 K8 cores? Bad scaling anyone? If they compared with 8 K10, that 80% should be even smaller (or maybe K10 not that improved).

[frankR]

Without specific reliable information this is meaningless. AMD does themselves no favors by making tests like this public and not disclosing specifics, then they open to speculation.

[madcho]

70% now?

It started with 42%, then 46% then 50% and now 70%?

Sounds like the 2 eggs that became 5 chickens.

40% between barcelona 2.4 and clovertown 2.66.
50% better at same clock with clovertown.
70% per core vs K8 at same clock.

simple

[gOJDO]

madcho: you are watching too many movies and you are consuming too much AMD marketing.

[PetNorth]

I think they configured the bench differently or used some scene that most sites don't.

http://www.youtube.com/watch?v=VGiv9Dtrc5Q
AMD man claims same GHz. So 16 K10 cores are just 80% faster than 8 K8 cores? Bad scaling anyone? If they compared with 8 K10, that 80% should be even smaller (or maybe K10 not that improved).

It's a little confusing. But I guess, they are comparing scalability: between 4S dual core K10 and 4S quad core K10.

[red]

If it is quad vs dual K10, then 80% isn't bad. But in the video, they call one machine "Opteron" then the quad "Barcelona"..
http://www.uberpulse.com/us/2007/05/...ce_as_fast.php
And this site which I presume has been more informed than us claims that it's "one with the Barcelona quad core chip and the other with a shipping dual core processor."

[XS Janus]

Without specific reliable information this is meaningless. AMD does themselves no favors by making tests like this public and not disclosing specifics, then they open to speculation.

quite true!

I understood it like this:

both systems used 4 socket mobos,
Opreton was dual core, so that means 8core total and scored 2000
K10 was quad core , so that would mean 16cores total and it scored 4000

If this is true it looks very bad considering it is the NEW architecture!!

[red]

To be more specific, according to this vid http://www.youtube.com/watch?v=VGiv9Dtrc5Q
8 "Opteron" got 2200 pixels/second in 112 seconds and 16 Barcelona got 4000 in 56 seconds. About 80% scaling...

[VulgarHandle]

To be more specific, according to this vid http://www.youtube.com/watch?v=VGiv9Dtrc5Q
8 "Opteron" got 2200 pixels/second in 112 seconds and 16 Barcelona got 4000 in 56 seconds. About 80% scaling...

twice the pixels/second in half the time, that's theoretically 4x the performance with only twice as many cores, using same or less thermal envolope, and at possibly less speeds, if what the inq says is true (inq says 65w and not the highest clock to be released)

if that's how the setup ran (4 quads new arc vs. 4 dual old arch), then it sounds good to me

[v_rr]

70% now?

It started with 42%, then 46% then 50% and now 70%?

Sounds like the 2 eggs that became 5 chickens.

50% to C2D
80% to K8

Duh....
At least read something before posting...

[PetNorth]

If it is quad vs dual K10, then 80% isn't bad. But in the video, they call one machine "Opteron" then the quad "Barcelona"..
http://www.uberpulse.com/us/2007/05/...ce_as_fast.php
And this site which I presume has been more informed than us claims that it's "one with the Barcelona quad core chip and the other with a shipping dual core processor."

but 8 cores K8 vs 16 cores K10, and only 80% faster the last machine (with a heavy multithreaded app), doesn´t make sense at all, because we know K10 features, and with them in mind, it's obvious K10 will be a lot faster than K8 with same number of cores and same clock.

[gOJDO]

twice the pixels/second in half the time, that's theoretically 4x the performance with only twice as many cores, using same or less thermal envolope, and at possibly less speeds, if what the inq says is true (inq says 65w and not the highest clock to be released)

if that's how the setup ran (4 quads new arc vs. 4 dual old arch), then it sounds good to me

1. You haven't understood what he said. 4 K10 CPUs are performing a little beat more than 4000 pixels/s, while 4 K8 CPUs are performing 2200 pixels/s. The 4 K10 CPUs finished one task for 56s, while the K8 finished the same task for 112s. That is 2X performance, not 4X.
2. If something is said by the_INQ it is BS or FUD by default.

[VulgarHandle]

1. You haven't understood what he said. 4 K10 CPUs are performing a little beat more than 4000 pixels/s, while 4 K8 CPUs are performing 2200 pixels/s. The 4 K10 CPUs finished one task for 56s, while the K8 finished the same task for 112s. That is 2X performance, not 4X.
2. If something is said by the_INQ it is BS or FUD by default.

again, 16 K10 cores performed a task in half the time as 8 K8 cores, at twice the performance

assuming POVray scales like most say it does, that would mean you would need 32 K8 cores to match the 16 K10 cores (double the K8 cores to match the pixels/s, then double that to match the time)... theoretically

[gOJDO]

again, 16 K10 cores performed a task in half the time as 8 K8 cores, at twice the performance

assuming POVray scales like most say it does, that would mean you would need 32 K8 cores to match the 16 K10 cores (double the K8 cores to match the pixels/s, then double that to match the time)... theoretically

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?

[PetNorth]

and BTW, K10 server will continue be calling "Opteron". So the fact Randy Allen calls Opteron to the dual core machine, doesn't mean it is a K8.
Again, I think tht two machines are K10, one 4S dual core, the other 4S quad core. So they are only showing scalability between them.

[red]

"twice the performance"
?
pixels/second isn't similar to frames per second. I think it just means how much pixels are processed in a second. For example, if a scene has 50,000 pixels and CPU A finishes in 10 seconds, it processed 5000 pixels/second, and if CPU B finished in 30 seconds, it processed 1667 pixels/second.

He labeled one machine "Opteron", one "Barcelona".
http://www.uberpulse.com/us/2007/05/...ce_as_fast.php
"Anyway, the demo compared two AMD 4-sockets systems: one with the Barcelona quad core chip and the other with a shipping dual core processor."
http://www.theinquirer.net/default.aspx?article=39756
Inq mentions everything similar except BIOS... K10 would need a new BIOS. ding