rates!Originally Posted by Nedjo
Oh yes it does:
....
if this is slower, then there is no point to this discusion:
non-rate benchmarks should be available in a couple of days, then we'll talk performance
rates!Oh yes it does:
....
if this is slower, then there is no point to this discusion:
non-rate benchmarks should be available in a couple of days, then we'll talk performance
Not really. The rates are a very bad bench in itself. Because its mainly memory bandwidth dependent.
If rates was a sign of actual speed. K8 would beat Core 2.
Crunching for Comrades and the Common good of the People.
sorry I'll explain a little better.
Surely the figures that are shown, does not put AMD in a bad light. I agree, the benchmark itself is of little importance but saying it makes AMD look bad (by the original Author) is no more than FUD?
Ou has updated September SPEC scores. AMD 2GHz K10 now loses by 4.2 points in int_rate.
page 20/21, I see around 7%+ for int_rate and 9.7% for fp_rate, with them supposedly just isolating fsb increase.
The problem is that you don't buy servers at a certain clock speedOn a 2p system the AMD opteron is 4% slower in SPECint_rate.
I don't understand what you mean by this.
It means it doesn't matter if it's clock for clock equivalent if it's clocked 2GHz to 3GHz
But that would indicate I was after a 2p 3ghz system. Where I work, we have more than just speed to think of.
If AMDs 2ghz cpu is costed equivalent to Intels 2ghz Xeon. If the rest of the system is all based the same, then both CPUs have an equal amount of chance of being bought.
Granted, if I was after pure speed then the 3ghz Xeons are the best bet but other considerations are taken into account before purchasing.
On a clock for Clock basis (and going of the price points), the benchmark does not put the AMD at a disadvantage or am I wrong?
"clock for clock", it looks fine, but it's a pity that 2Ghz quad can only command top price of $372 (it's currently $873 for 2P dual 3GHz k8).
If you look at old numbers, you'll see that AMD wasn't much behind in int_rate per core/clock anyway
http://www.spec.org/cpu2006/results/...604-01203.html
http://www.spec.org/cpu2006/results/...803-01638.html
http://www.spec.org/cpu2006/results/...821-01887.html
http://www.spec.org/cpu2006/results/...820-01831.html
http://www.spec.org/cpu2006/results/...412-00810.html
So AMD has gone from a bit behind in int_rate, to creepily close, bravo! Now if you consider that AMD scales better, on nonrate speccpu... Just comparing Ou's 2P numbers, 88/(6.4 or 7.2 which is 80-90% scaling), gets you 12.2-13.8 nonrate specint score. 92 for Intel/(5.6-6.4) which is 70-80% gets you 14.4 and 16.4.
btw, specint is a single bench suite. specintrate and sorts is running multiple instances of specint, voila
Perhaps someone else can jump in here.
"Clock for clock" it seems on par with intel. If that is the case and the prices are equal. Surely it doesn't make the CPU a bad buy?
Ofcourse not.
Crunching for Comrades and the Common good of the People.
QFT
if this is slower, then there is no point to this discusion:
Using rates benches is like using DiVX encoding with a Conroe vs Penryn and claim average performance of Penryn is double of Conroe.
Imagine if we didnt have to be feed with marketing spins and lies. So many people that would be unemployed.
Crunching for Comrades and the Common good of the People.
Not quite. Believe it or not, there are some people who like to know how a system performs when all cores are loaded. This is what the rate benchmarks test. Not everyone runs a single thread on a quad-core.
Rates doesnt even reflect that either. Try do some rendering or the like. Rates is only useful for heavy bandwidth applications. And rates isn´t even multithreaded as such. Its more a bunch of singlethreaded applications running in its benching method to simulate alot of "clients".Not quite. Believe it or not, there are some people who like to know how a system performs when all cores are loaded. This is what the rate benchmarks test. Not everyone runs a single thread on a quad-core.
So that somehow brings us back to 4x4 and "megatasking".
Last edited by Shintai; 09-08-2007 at 04:08 PM.
Crunching for Comrades and the Common good of the People.
ftp://ftp.software.ibm.com/eserver/b...cpu_091007.pdf
2 socket, 1.9ghz k10s
SPECint2006 11.3
SPECint_rate2006 83.2
SPECint_rate_base2006 72.8
SPECfp2006 11.2
SPECfp_rate2006 73.0
SPECfp_rate_base2006 68.5
(1) Planned availability for the x3455 model using the AMD Opteron Model 2347 processor
(1.9GHz, 512KB L2 cache per core) is November 16, 2007.
------
ftp://ftp.software.ibm.com/eserver/b...yna_091007.pdf
The x3455 server achieved an elapsed time of 31,526 seconds, which is better than the 32,578
seconds of elapsed time achieved by a single-node 8-core blade system using the Quad-Core
Intel® Xeon® X5355 processor at 2.66GHz.
So according to the 2 PDFs,1.9Ghz DP Barcelona(2347) system is a little bit faster than 2.66 DP intel(X5355) quad system in this LS Dyna set of benchmarks.
Last edited by informal; 09-08-2007 at 09:07 PM.
nice.. K10 owned
Interesting... the non-rate scores look like they will fall short of a 2.0 GHz clovertown....
Take for example, a 5335 http://www.spec.org/cpu2006/results/...0301-00593.pdf
Barcey at 1.9 GHz SpecFP2006 -- 11.2
Clovertown at 2.0 GHz SpecFP2006 -- 13.6 (base)
It appears Barcy's key strength is BW driven.
Last edited by JumpingJack; 09-08-2007 at 10:26 PM.
qft, I mean come on people, does it not seem a bit funny that the only semi official leaked benches are rate benches that k8 already fears well in?
Those Specint2006 and Specfp2006 scores are very very weak, well behind what Clovertown offers on a clock for clock basis. This bodes quite badly for what K10 will offer in terms of desktop performance.
So when we stop using bandwidth/rates benches. K10 suddenly drops behind...how funny...and how predictable.
Crunching for Comrades and the Common good of the People.
no one in the industry who buys these CPUs gives a damn about how wellSo when we stop using bandwidth/rates benches. K10 suddenly drops behind...how funny...and how predictable.
a single core performs. System performance is what matters period.
your bashing of the _rate benchmarks and aproval of the nonrates is really
hard to understand and makes no sense. For your information the _rates
run the same nonrate benchmarks but on all the cores of the system, thus
giving us a good clue of how the whole system works on certain type of
applications. hard to pick your logic... seriously...
1 more day to go till our new rendering monster
Except the same cores will be used for the desktop market, and where the non-rate SPEC tests are more relevant than the rate tests.
I don't think it's a bashing of the rate benchmarks specifically, merely a desire for more benchmarks that paint a more complete picture of the performance of Barcelona, vs AMD's continual usage of only SPECfp_rate for months to hype the performance of Barcelona while being silent on the many other available server benchmarks.your bashing of the _rate benchmarks and aproval of the nonrates is really
hard to understand and makes no sense.
Running multiple copies of single-threaded applications is not really how most server applications work.For your information the _rates
run the same nonrate benchmarks but on all the cores of the system, thus
giving us a good clue of how the whole system works on certain type of
applications. hard to pick your logic... seriously...
This whole debate reminds of when the P4 first made its appearance, at that stage memory bandwidth was the buz word and how important it was for Quake 3. Strange how times have changed.
And you know what are the scores of Clowerton @1.9 GHz? Please share them with us!
Posting Permissions
Reply With Quote
Bookmarks