Sign in with
Sign up | Sign in
Your question

INQ: AMD releases more (Bull****) Barcelona "benches"

Tags:
Last response: in CPUs
Share
a b à CPUs
May 22, 2007 11:14:01 AM

And we are supposed to be able to see whats on those monitors...?
May 22, 2007 11:14:34 AM

As you can clearly see from this picture with no support facts, evidence, details, or even really knowing what the benchmark was, Barcelona is 5,000,000% faster. Intel BK 2Q 2008.
Related resources
May 22, 2007 11:14:55 AM

I love how they claim "It will blow away Clovertown," but then don't reveal any frequencies or compare it against Clovertown.

:lol: 
May 22, 2007 11:19:03 AM

Quote:
Not being a POV expert, I can't say how it scales, or really relate this to much. As an exercise for our readership, can someone work this backwards and tell me about what the clocks here are?



I ran this through the Alta Vista BabelFish translator, set to translate BullShitt to english. This is what it gave back:

Quote:
Not having a clue how computers or benchmarks work, I can't say what this means or how it relates to anything. As Im too lazy, could one of the 18 people who read my blog figure this out for me?
May 22, 2007 12:01:16 PM

Task manager shows 8 cores on the left, 16 cores on the right. The rendered scene looks like the standard POVray benchmark which has predefined default settings.

If what Charlie says is true, 8 Opteron cores hit 2200 px/s and 16 Barcelona cores hit >4000 px/s. Both are 4-socket configurations, so the Opteron clocks are probably higher.

For reference, a single Merom core at 3.0 GHz using one channel of DDR2 hits about 520 px/s, but scaling is not quite linear, not to mention (1) Clovertown uses slower FB-DIMMs and (2) a 4S configuration for Clovertown may require separate nodes and thus utilize NUMA with further potential slowdown.

Edit: According to this link - http://www.realworldtech.com/page.cfm?ArticleID=RWT1114... - Clovertown exhibits nearly linear scaling in a 2S SMP configuration. With just 8 cores at 2.33GHz it nearly reaches the claimed benchmark readout from 16 Barcelona cores, clock unspecified. That makes AMD's bench very unimpressive.
May 22, 2007 12:01:24 PM

Quote:
Not having a clue how computers or benchmarks work, I can't say what this means or how it relates to anything. As Im too lazy, could one of the 18 people who read my blog figure this out for me?


:trophy: :lol:  LOL :lol:  :trophy:
May 22, 2007 12:04:42 PM

Quote:
Not having a clue how computers or benchmarks work, I can't say what this means or how it relates to anything. As Im too lazy, could one of the 18 people who read my blog figure this out for me?


:trophy: :lol:  LOL :lol:  :trophy:
May 22, 2007 12:31:31 PM

Quote:
without clock details, this is quite meaningless.


QFT
Anonymous
May 22, 2007 12:42:25 PM

2007 is a lying and lieing year for AMD.
May 22, 2007 1:21:10 PM

Looks like the Barcelona will be faster than an Opteron but we don't know which one. :roll: :roll:
May 22, 2007 1:21:33 PM

Quote:
Task manager shows 8 cores on the left, 16 cores on the right. The rendered scene looks like the standard POVray benchmark which has predefined default settings.

If what Charlie says is true, 8 Opteron cores hit 2200 px/s and 16 Barcelona cores hit >4000 px/s. Both are 4-socket configurations, so the Opteron clocks are probably higher.

For reference, a single Merom core at 3.0 GHz using one channel of DDR2 hits about 520 px/s, but scaling is not quite linear, not to mention (1) Clovertown uses slower FB-DIMMs and (2) a 4S configuration for Clovertown may require separate nodes and thus utilize NUMA with further potential slowdown.

Edit: According to this link - http://www.realworldtech.com/page.cfm?ArticleID=RWT1114... - Clovertown exhibits nearly linear scaling in a 2S SMP configuration. With just 8 cores at 2.33GHz it nearly reaches the claimed benchmark readout from 16 Barcelona cores, clock unspecified. That makes AMD's bench very unimpressive.

Well thanks for clearing that up, and posting up some numbers. Now like mentioned before, Barcelona does not look that impressive at all.
May 22, 2007 1:25:38 PM

Quote:
This was my assessment -- but if this was a 1.8 GHz Barcey doing this then it is impressive, if it were a 2.2 GHz Barcey doing this, then it is not as impressive.... without clock details, this is quite meaningless.


How do you like that! I have to take off for a long weekend and when I come back, not only does Barcy still NOT EVEN REMOTELY EXIST ANYWHERE BUT IN HECTOR'S IMAGINATION, but JACK HAS BECOME A RASTA!!!

AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAaaaaaaaaaaaaaaaaahhh!
May 22, 2007 1:49:52 PM

Quote:
Not being a POV expert, I can't say how it scales, or really relate this to much. As an exercise for our readership, can someone work this backwards and tell me about what the clocks here are?



I ran this through the Alta Vista BabelFish translator, set to translate BullShitt to english. This is what it gave back:

Quote:
Not having a clue how computers or benchmarks work, I can't say what this means or how it relates to anything. As Im too lazy, could one of the 18 people who read my blog figure this out for me?


Man, you had me rolling with that one. LOL!
May 22, 2007 1:55:42 PM

Quote:
http://www.theinquirer.net/default.aspx?article=39756

As usual no clockspeeds given :lol: 


Twice as many cores....twice as fast?
That's assuming POV ray scales well to multi core.
At lower clocks its impressive though.

Not sure what to conclude from this since....
http://www.xbitlabs.com/articles/cpu/display/intel-v8_1...
That's only 8 cores. Was it the same bench.
Bleh AMD speculation, so annoying.

That task manager looks funny with so many thin cpu util bars! :idea:
May 22, 2007 1:58:20 PM

Quote:
This was my assessment -- but if this was a 1.8 GHz Barcey doing this then it is impressive, if it were a 2.2 GHz Barcey doing this, then it is not as impressive.... without clock details, this is quite meaningless.

Clock speed was roughly given when they specified that each socket was a 65W TDP part. This would be in the same power envelope as existing HE-series Opterons.

According to this table from last December, the parts should be around 1.9-2.0GHz:
http://bp0.blogger.com/_Z49NBhppzoY/RcWhUzjKndI/AAAAAAA...
May 22, 2007 2:12:22 PM

You know, the POV sorta-bench may or may not be impressive. But what is truly impressive is watching you guys dig around and work to reverse estimate the clock speeds and substantiate the relation of the results to other processors.

Kudos to all of you for your efforts and results! Well done!


:trophy: :trophy:
May 22, 2007 2:41:42 PM

LMAO!!! this proves nothing! all i see is 4 Optys and 16 Barcey's, and the Barcey's are slower; if they did an equal benchmark then the Optys would have gotten around 4400 and the Barceys-4000. what are they proving? 16 low clocked barceys out perform 8 high clocked Optys?LMAO just throw my E6600 at them which is"high clocked" and we'll see who wins LOLOLOL
May 22, 2007 2:56:17 PM

We know that AMD has managed to outperform their own last gen CPUs.
Better than nothing :lol: 
May 22, 2007 2:56:34 PM

8O Holy S***; this means the Opteron performed equally (or even somehow better than the Barcelona (because doubling the cores in rendering, generally improves performance by a factor of 1.8-1.9X,.... I am starting to see why AMD is sticking with th powerPoint presentations; Barcelona SUCKS
May 22, 2007 3:00:37 PM

Quote:
We know that AMD has managed to outperform their own last gen CPUs.
Better than nothing :lol: 

No man, they didn't! The only way that the barcelona system can somehow conform to their numbers is that the Opteron system features the newest 3.0Ghz Opterons and the barcelona is clocked @ 1.9Ghz.
a b à CPUs
May 22, 2007 3:22:53 PM

I put your THG thread on this topic through my babelfish translater and came up with the following generic response from the hardcore members here:

"We are pissed off that we are last on the list to get any details on Barcelona so lets flame the Inq site out of pure frustration. Um ... the numbers are complex and confusing because we can't see the picture and put the words into some sort of meaningful context".

Grow up ... your flaming responses are childish.

I have a range of PC's with different processors ... suited to different applications. Intel and AMD.

Did you think the world would stand still while you clucked over your C2D cpu's you paid an arm and a leg for?

You should be thankful Intel's major campaign to destroy AMD hasn't yet succeeded ... or you would be kissing goodby to continued performance leaps.

Competition is crutial.

How come the Inq staff got to see a demo and not Dailytech or TGDaily??
That's a question you should be asking and the answer is quite simple ... but I'll let you find out for yourself.
May 22, 2007 4:17:49 PM

Quote:
I put your THG thread on this topic through my babelfish translater and came up with the following generic response from the hardcore members here:

"We are pissed off that we are last on the list to get any details on Barcelona so lets flame the Inq site out of pure frustration. Um ... the numbers are complex and confusing because we can't see the picture and put the words into some sort of meaningful context".

Grow up ... your flaming responses are childish.

I have a range of PC's with different processors ... suited to different applications. Intel and AMD.

Did you think the world would stand still while you clucked over your C2D cpu's you paid an arm and a leg for?

You should be thankful Intel's major campaign to destroy AMD hasn't yet succeeded ... or you would be kissing goodby to continued performance leaps.

Competition is crutial.

How come the Inq staff got to see a demo and not Dailytech or TGDaily??
That's a question you should be asking and the answer is quite simple ... but I'll let you find out for yourself.


Fail.

The Inq sucks. It's a pro-AMD rumor mill, not a site that has even a fraction of the respect, or for that matter, the readership that Tom's or Dailytech has.
May 22, 2007 4:28:53 PM


OUCH! ARE THEY F****** CRAZY; it's K8 performance 8O

I dunno. These are HE and low clock.

I have one question. Are we sure this dual core is a K8? Like i know he said opteron, but what will the dual core server processors be called?
May 22, 2007 4:35:19 PM

One thing i don't like is vaque answers that are talking around the real point. Amd employees are sounding like politiciens...

AMD football team/ spokespeople
May 22, 2007 4:47:18 PM

Quote:
I put your THG thread on this topic through my babelfish translater and came up with the following generic response from the hardcore members here:

"We are pissed off that we are last on the list to get any details on Barcelona so lets flame the Inq site out of pure frustration. Um ... the numbers are complex and confusing because we can't see the picture and put the words into some sort of meaningful context".

Grow up ... your flaming responses are childish.

I have a range of PC's with different processors ... suited to different applications. Intel and AMD.

Did you think the world would stand still while you clucked over your C2D cpu's you paid an arm and a leg for?

You should be thankful Intel's major campaign to destroy AMD hasn't yet succeeded ... or you would be kissing goodby to continued performance leaps.

Competition is crutial.

How come the Inq staff got to see a demo and not Dailytech or TGDaily??
That's a question you should be asking and the answer is quite simple ... but I'll let you find out for yourself.


I put this thru my babelfish translator and came back with:

Quote:
I am an Inq shill.
May 22, 2007 4:59:28 PM

Just found some benches that may help a little in terms of scalability of POV with cpu count and frequency

Core 2 Extreme Qx6700 (2.66Ghz) 110seconds
Core 2 QX6600 (2.4Ghz) 122seconds
FX74 (dual socket) dual core (3Ghz) 134seconds
FX72 (dual socket) dual core (2.8Ghz) 145seconds
FX70 (dual socket) dual core (2.6Ghz) 156seconds
Core 2 X6800 (2.93Ghz) 199seconds

So lets try and normalize everything by dividing by frequency to get seconds per GHZ and then by core to get seconds per ghz per core
So for the Core 2 Quads you get
QX6700 41.4 seconds/Ghz 10.35
QX6600 50.8 seconds/Ghz 12.7
Core 2 67 seconds/Ghz 33.5

Opterons

FX74 44.66 seconds/Ghz 11.165
FX72 51.79 seconds/Ghz 12.94
FX70 60 seconds/Ghz 15

The graphs by frequency are pretty much linear (given the limited amount of data here)

Suprising is that at the effciency of the code goes up as frequency gets better.
And jumping from 2 cores to 4 you get a very rough 60% extra throughput.

So how do you relate that to the current set of barcelona data.

Assuming it were standard opterons going from dual core to quad core at the same frequency, you should (using the very bad maths above) go from score of 2200 to 3520 based on the INQ data. Whereas the barcelona is scoring just over 4000.
At the same frequency that would be a 14% improvement.
So a standard Opteron would have to run at 3.42GHz to compete.
If the barcelona is running at 1.9Ghz that is roughly an 80% improvement over the standard opteron in floating point.

Now my maths may be comepletly off, anybody feel free to correct.

edit: forgot my source for the data.
http://www.hwupgrade.com/articles/print/cpu/10/quad-fx-...
May 22, 2007 5:01:51 PM

I rewatched the video, and at 2:10 he mentions that the dual-core is not the fastest in production, nor is the quad going to be the fastest in production. Unless he misspoke, he clearly implies the system on the left is based on existing K8, while the one on the right is based on unreleased K10.

I'll take back what I mentioned about the systems consuming equal power as he did not mention the TDP of the K8 chips, which are probably lower because the frequency is fixed, but not down to half of the K10 system.

It seems POVRay is not a benchmark for which K10 cores are any better per-clock than K8, except for power consumption.
May 22, 2007 5:20:27 PM

Seconds per GHz is not a valid performance index because as GHz goes up, seconds go down.

GHz * seconds (multiplied by cores) would be a valid index and you'd see a mostly constant number across all the Conroes, again for all the Kentsfields, and so on. This figure represents how many seconds a 1GHz device would take, with perfect downscaling, and lower is better efficiency for a given setup and architecture.

Using your benchmark numbers:

Core 2 Qx6700 (2.66Ghz) 110seconds 1170 Ghz s
Core 2 QX6600 (2.4Ghz) 122seconds 1171 Ghz s
FX74 (dual socket) dual core (3Ghz) 134seconds 1608 Ghz s
FX72 (dual socket) dual core (2.8Ghz) 145seconds 1624 Ghz s
FX70 (dual socket) dual core (2.6Ghz) 156seconds 1622 Ghz s
Core 2 X6800 (2.93Ghz) 199seconds 1166 Ghz s

The % performance-per-clock improvement from FX74 (QuadFX) to X6800 (Conroe) is simply 1608/1166 = 38% bonus. That from Kentsfield to Conroe is 0.34%, or practically perfect scaling.

Figures from the K8 vs. Barcelona video assuming 1.9GHz parts:

K8 4S * 2C (1.9Ghz) 116seconds 1763 Ghz s
K10 4S * 4C (1.9Ghz) 62seconds 1884 Ghz s

As you can see, I'm not impressed with performance-per-clock from those demo figures... they're a far cry from Kentsfield, even with clock down to 1.6Ghz (I suspect they're really at 1.8GHz). Then again, this is only one benchmark and ironically puts Clovertown in the best light - wonder what AMD was thinking.
May 22, 2007 5:47:12 PM

Quote:
I put your THG thread on this topic through my babelfish translater and came up with the following generic response from the hardcore members here:

"We are pissed off that we are last on the list to get any details on Barcelona so lets flame the Inq site out of pure frustration. Um ... the numbers are complex and confusing because we can't see the picture and put the words into some sort of meaningful context".

Grow up ... your flaming responses are childish.

I have a range of PC's with different processors ... suited to different applications. Intel and AMD.

Did you think the world would stand still while you clucked over your C2D cpu's you paid an arm and a leg for?

You should be thankful Intel's major campaign to destroy AMD hasn't yet succeeded ... or you would be kissing goodby to continued performance leaps.

Competition is crutial.

How come the Inq staff got to see a demo and not Dailytech or TGDaily??
That's a question you should be asking and the answer is quite simple ... but I'll let you find out for yourself.



ROLFMAO!!
:lol:  :lol:  :lol:  :lol:  :lol:  :lol: 

Quote:
I put your THG thread on this topic through my babelfish translater and came up with the following generic response from the hardcore members here:

"We are pissed off that we are last on the list to get any details on Barcelona so lets flame the Inq site out of pure frustration. Um ... the numbers are complex and confusing because we can't see the picture and put the words into some sort of meaningful context".

Grow up ... your flaming responses are childish.


Is that you Charlie? It is, isnt it. :lol:  :lol:  :lol:  :lol: 
Hey man, dont get upset with us because the Inq is your own personal blog masquerading as a news site...you made it not us. Our responses are no more childish, if not infinately less so, than your rabid fanboyism. As proof I offer the fact that Ive never seen a THG poster, Horde, Brood or nuetral wearing a bunny costume in public because the lost a bet of the accuracy of information. Can you say the same for yourself?


Quote:
I have a range of PC's with different processors ... suited to different applications. Intel and AMD.

Did you think the world would stand still while you clucked over your C2D cpu's you paid an arm and a leg for?


You mean the C2D in the HP Pavilion Im writing this on....oh wait...its an AMD 64 3200. So much for that theory. But I did buy a C2D for desktop last year. You remeber last year.....When E6600s were taking $359 US while the X2 5000s were taking $550 and up. When you could find them. Umm so what processors cost an arm & a leg again? And who was standing still? Did you mean AMD stood as still as a deer in the headlights when Intel suprised them by releasing something that wasnt another stepping of Crapburst?


Quote:

You should be thankful Intel's major campaign to destroy AMD hasn't yet succeeded ... or you would be kissing goodby to continued performance leaps.

Competition is crutial.
Quick someone, alert the JCS. Charlie finally got one right, its a miracle.

Quote:

How come the Inq staff got to see a demo and not Dailytech or TGDaily??
That's a question you should be asking and the answer is quite simple ... but I'll let you find out for yourself.

LMAO....who gives a crap? Me ...nope. This is a forum Charlie. THG Forumz, not a blog like your Inq, or a test site like THG itself. We discuss news here, and if required, the accuracy of the source. We dont do the testing :roll:

Charlie..you should stick to your blog man, Im worried about you. Playing in the forum doesnt appear to be good for your blood pressure and judging by the photos of you on the net, you shouldnt be taking that risk. Hugs-n-kisses sweety, have a safe trip back to blogquirer
May 22, 2007 6:26:29 PM

1. both cpus are running at the same clock speed.

2. the quad core on the right finished before the dual core even though it was started later.

3. quad core achieved 4000/pixels per second.

4. dual core achieved a little over 2200/pps

5. cpu time for the quad core was a little over 62 seconds.

6. cpu time for the dual core was about 116 seconds.

[my bad]
May 22, 2007 6:38:36 PM

Where does salesman say it isn't new vs old?
May 22, 2007 8:15:06 PM

Quote:
Competition is crutial.


That's why we can't get any K10 benchys! It's not a CPU! It's a crouton!

Might go good with a nice balsamic vinaigrette though... :lol: 
May 22, 2007 8:20:29 PM

Really funny; we all look like KGB spies trying to decode the message :lol: 
So, That's what I understood from that:
1- It's K8 vs K10
2- there are two identical 4-socket systems...
3- however, the guy (when explains at the beginning), says POV Ray is only capable of using two sockets , so it's 4x K8 cores vs 8x K10 cores
4- the system using double the cores renders twice faster
________________________________________________________
Conclusion; it's a SHAME!
May 22, 2007 9:05:02 PM

Quote:
LMAO....who gives a crap? Me ...nope. This is a forum Charlie. THG Forumz, not a blog like your Inq, or a test site like THG itself. We discuss news here, and if required, the accuracy of the source. We dont do the testing :roll:

I (or is it we?) take offense to that.
May 22, 2007 9:12:42 PM

Well... this thread is actually interesting. :lol: 
May 22, 2007 9:25:20 PM

Quote:

I (or is it we?) take offense to that.

What the....Who the hell are you..Oh, Sorry Ninja, its you. Didnt mean to Dis you, it being your week and all. Its just, well, you know, you all look alike to me[/shrug] :wink: :lol: 
May 22, 2007 10:35:13 PM

Quote:
I put your THG thread on this topic through my babelfish translater and came up with the following generic response from the hardcore members here:

"We are pissed off that we are last on the list to get any details on Barcelona so lets flame the Inq site out of pure frustration. Um ... the numbers are complex and confusing because we can't see the picture and put the words into some sort of meaningful context".

Grow up ... your flaming responses are childish.

I have a range of PC's with different processors ... suited to different applications. Intel and AMD.

Did you think the world would stand still while you clucked over your C2D cpu's you paid an arm and a leg for?

You should be thankful Intel's major campaign to destroy AMD hasn't yet succeeded ... or you would be kissing goodby to continued performance leaps.

Competition is crutial.

How come the Inq staff got to see a demo and not Dailytech or TGDaily??
That's a question you should be asking and the answer is quite simple ... but I'll let you find out for yourself.


What the hell are you on about. Nothing to do with the Inq and even if it is their own story shows their lack of knowledge. More so to do with how 16x k10 didnt double the performance of 8x k8
May 22, 2007 10:38:44 PM

Quote:
If the barcelona is running at 1.9Ghz that is roughly an 80% improvement over the standard opteron in floating point.

Now my maths may be comepletly off, anybody feel free to correct.

edit: forgot my source for the data.
http://www.hwupgrade.com/articles/print/cpu/10/quad-fx-...


The video stated that all processors were running at the same clock speed.
May 22, 2007 10:43:09 PM

Quote:
3- however, the guy (when explains at the beginning), says POV Ray is only capable of using two sockets , so it's 4x K8 cores vs 8x K10 cores


That's strange. Very strange. Physical and logical design should be transparent to software? I can't see why a piece of software would care or know how many sockets there are?
May 22, 2007 10:54:05 PM

Quote:
3- however, the guy (when explains at the beginning), says POV Ray is only capable of using two sockets , so it's 4x K8 cores vs 8x K10 cores

Are you sure? I thought he was saying Clovertown was only capable of two-socket operation, which is only partly correct. Number of sockets is determined by the chipset (memory controller), and Intel did demonstrate a 4S Clovertown last year, but it might still be in development or on hold. That 4S demo used POVRay as well, and all 4 sockets were used to provide a ~16x speedup.
May 22, 2007 11:38:26 PM

Quote:
INQ: AMD releases more (Bull****) Barcelona "benches"


Did you step in it?
!