Ipteron 6282 SE SiSoft benchmarks

E

eastmen

Limp Gawd
Joined
Dec 6, 2010
Messages
333
http://www.sisoftware.eu/rank2011d/...a895a086eed3e6c0b885b492f792af9fb9caf7c7&l=en


cessor Arithmetic Benchmark Anonymous World 2x AMD Opteron(tm) Processor 6282 SE (2N 8CU 16C 2GHz, 8x 2MB L2, 6MB L3) 231.971 GOPS 2 / 32 2048 MHz Normal Server/WS Windows x64 6.1.1 7 September 2011
Processor Multi-Media Benchmark Anonymous World 2x AMD Opteron(tm) Processor 6282 SE (2N 8CU 16C 2GHz, 8x 2MB L2, 6MB L3) 585.983 Mpix/s 2 / 32 2048 MHz Normal Server/WS Windows x64 6.1.1 7 September 2011
Multi-Core Efficiency Benchmark Anonymous World 2x AMD Opteron(tm) Processor 6282 SE (2N 8CU 16C 2GHz, 8x 2MB L2, 6MB L3) 21.773 GB/s 2 / 32 2048 MHz Normal Server/WS Windows x64 6.1.1 7 September 2011
Cryptography Benchmark Anonymous World 2x AMD Opteron(tm) Processor 6282 SE (2N 8CU 16C 2GHz, 8x 2MB L2, 6MB L3) 4.661 GB/s 2 / 32 2048 MHz Normal Server/WS Windows x64 6.1.1 7 September 2011
Click to expand...


Seems really good if the 2ghz clock speeds are real


Doh can a mod change the I to an O in the title !
 
click each test and it will show what other setups are getting also. From the one I looked at, it's about 8% behind Intel clock for clock, core for core.
 
So can somebody do some rough math and give an estimate to how Bulldozer will perform based upon these scores?
 
AMD_Gamer said:
So can somebody do some rough math and give an estimate to how Bulldozer will perform based upon these scores?
Click to expand...

theres no way to do rough math on a 16 core processor at 2Ghz that is designed for server loads to compare it to a consumer level processor that won't have all the opteron optimizations and only being 8 cores at almost twice the clock speed.
 
flyk at semiaccurate posted this in comparison to the numbers

http://www.sisoftware.co.uk/rank2011...eb8cbf6c7&l=en

so
MC 2*12core @ 2,5GHz
rocessor Arithmetic Benchmark 2x AMD Opteron 6180 SE 202.098 GOPS 2 / 24 2500 MHz
Processor Multi-Media Benchmark 2x AMD Opteron 6180 SE 331.207 Mpix/s 2 / 24 2500 MHz
Multi-Core Efficiency Benchmark 2x AMD Opteron 6180 SE 8.824 GB/s 2 / 24 2500 MHz

vs 2*16core (2GHz?). added % improvement without clock and core normalization because i'm not sure this one is running @ 2Ghz.. The numbers would give MC a 25% clock advantage.... but an all core turbo of 500MHz would be equal clocks..

Processor Arithmetic Benchmark 2x AMD Opteron(tm) Processor 6282 SE (2N 8CU 16C 2GHz, 8x 2MB L2, 6MB L3) 231.971 GOPS 2 / 32 2048 MHz +15%
Processor Multi-Media Benchmark 2x AMD Opteron(tm) Processor 6282 SE (2N 8CU 16C 2GHz, 8x 2MB L2, 6MB L3) 585.983 Mpix/s 2 / 32 2048 MHz +76%
Multi-Core Efficiency Benchmark 2x AMD Opteron(tm) Processor 6282 SE (2N 8CU 16C 2GHz, 8x 2MB L2, 6MB L3) 21.773 GB/s 2 / 32 2048 MHz +146%

ranging from very little to alot. (side note is that MC did very well in arithmitics even compared to intel..)
Click to expand...
 
So the AMD system has 16CU, 32 "cores", and essentially ties clock for clock with 16 real Xeon cores.
AMD, I am disappoint.
 
Isn't the whole point of BD is that their "real" cores are better than intels "pretend" cores? There's several of these opteron 6k series benchmarks up there now and they're frankly not that flattering... Where is JF to spin this one now? How much performance do they really expect to get out of different bios, drivers and optimizations...

I can't ever think of an instance where a bios upgrade dramatically affected cpu performance... bios is updates generally determine if the cpu works or doesn't. Even if driver, OS, and application optimizations gave 5% improvement each (which seems generous) a 15% boost to these numbers is still bad. Honestly I was expecting BD to at least put a hurting on intel for highly multi-threaded apps by 20-30% at least... but still falling behind for single threaded clock per clock.

This is just disappointing... if this is real, and it looks more and more each day that it either is or close enough, we're in for another 3 or 4 years of intel milking the high end
 
-Dragon- said:
Isn't the whole point of BD is that their "real" cores are better than intels "pretend" cores? There's several of these opteron 6k series benchmarks up there now and they're frankly not that flattering... Where is JF to spin this one now? How much performance do they really expect to get out of different bios, drivers and optimizations...

I can't ever think of an instance where a bios upgrade dramatically affected cpu performance... bios is updates generally determine if the cpu works or doesn't. Even if driver, OS, and application optimizations gave 5% improvement each (which seems generous) a 15% boost to these numbers is still bad. Honestly I was expecting BD to at least put a hurting on intel for highly multi-threaded apps by 20-30% at least... but still falling behind for single threaded clock per clock.

This is just disappointing... if this is real, and it looks more and more each day that it either is or close enough, we're in for another 3 or 4 years of intel milking the high end
Click to expand...
One kind of BIOS glitch is showing that a CPU is running at 3GHz when it is only running at 2GHz. A correctly working BIOS can make a world of difference from one that doesn't work.
 
JMccovery said:
One kind of BIOS glitch is showing that a CPU is running at 3GHz when it is only running at 2GHz. A correctly working BIOS can make a world of difference from one that doesn't work.
Click to expand...

These server cpu's are shipping. Now. The motherboard manufacturer's and OEM's have had month's with engineering samples to validate these CPU's. Even if there were a bios glitch, a glitch of that magnitude is completely unacceptable and highly unlikely this late in the game.
 
CaptNumbNutz said:
These server cpu's are shipping. Now. The motherboard manufacturer's and OEM's have had month's with engineering samples to validate these CPU's. Even if there were a bios glitch, a glitch of that magnitude is completely unacceptable and highly unlikely this late in the game.
Click to expand...

That isn't fully true. Right now the chips are shipping out to huge companys like Cray who are either using their own boards or have a company for it that will work with them and can at some point have daily updates to the bios to work around any bug.

Its far diffrent than us consumers who are lucky if we get a bios revision
 
bwang said:
Yeah, but AMD's "core" is much more of a core than an Intel virtual core.
Click to expand...

If BD's 2way Interlagos 32 core early models slightly beat Intel's 16+16ht cores for the same TDP and transistor count, then it's all good. Intel has be dethroned.
 
sirmo said:
If BD's 2way Interlagos 32 core early models slightly beat Intel's 16+16ht cores for the same TDP and transistor count, then it's all good. Intel has be dethroned.
Click to expand...

Only in the server market (which, to be fair, is the target market of Interlagos).
It appears that clock for clock, a Bulldozer core is slightly less than half a Sandy Bridge core, so a 3.6GHz Zambezi would perform the same as quad-core Nehalem on highly threaded tasks...sadly, that's not the case for most games.
Now, if Zambezi were at the same price point as the Phenom II X6, all would be well; i would definitely get a couple for my silly compute tasks (rendering, factoring, etc)...sadly, this is not the case (~$300 MSRP).
 
Michaelius said:
There should be 230-260$ 8 core FXes.
Click to expand...

AX6FL.png


This item was last known to be back-ordered, however, that may no longer be accurate. You may order this product regardless of availability.


3EU/6T
$188.32 - FX-6100 SIX CORE AM3+ 14MB BOX 95W 3300MHZ
http://www.shopblt.com/cgi-bin/shop...age=01100300U031_BLA5136P.shtml&order_id=!ORD ERID!

4EU/8T
$221.73 - FX-8120 EIGHT CORE AM3+ 16MB BOX 125W 3100MHZ
http://www.shopblt.com/cgi-bin/shop...age=01100300U031_BLA5135P.shtml&order_id=!ORD ERID!

4EU/8T
$266.28 - FX-8150 EIGHT CORE AM3+ 16MB BOX 125W 3600MHZ
http://www.shopblt.com/cgi-bin/shop...age=01100300U031_BLA5134P.shtml&order_id=!ORD ERID!
Click to expand...

Seems like the top of the line will be $266 boxed , prob see it for $240 or less oem.
 
I wonder how Bulldozer's integer performance is...given that it replicates the integer units 8 times, we should hopefully see performance near SB-E's.
 
bwang said:
Only in the server market (which, to be fair, is the target market of Interlagos).
It appears that clock for clock, a Bulldozer core is slightly less than half a Sandy Bridge core, so a 3.6GHz Zambezi would perform the same as quad-core Nehalem on highly threaded tasks...sadly, that's not the case for most games.
Click to expand...

Don't see how you can derive clock for clock performance considering the nature of the shared design in BD. A Core not sharing the FPU with another core may change your figures drastically. Also BD's Turbo appears to be much more aggressive than Intel's.

This is why BD is such an enigma. All we know is that BD's performance is very situational. Multi-threaded performance numbers don't correlate to single threaded performance numbers. Also it depends on the benchmark. Is it FPU heavy, Integer heavy? We really have to see a lot of benchmarks to know what we're dealing with here.
 
sirmo said:
Don't see how you can derive clock for clock performance considering the nature of the shared design in BD. A Core not sharing the FPU with another core may change your figures drastically. Also BD's Turbo appears to be much more aggressive than Intel's.

This is why BD is such an enigma. All we know is that BD's performance is very situational. Multi-threaded performance numbers don't correlate to single threaded performance numbers. Also it depends on the benchmark. Is it FPU heavy, Integer heavy? We really have to see a lot of benchmarks to know what we're dealing with here.
Click to expand...

I was referring to FP performance...media encoding is a pretty good measure of multithreaded floating-point performance, and AMD's architectures have historically had pretty good multicore scaling.
For gaming, where more than four threads are rarely present, BD might be a reasonable choice given the new, lowered price points.
I'm very much hoping that Bulldozer excels at integer workloads..if a Bulldozer core performs anywhere near an Intel core at integer math, then the 4CU FX should be as fast as SB-E for less than half the price =)
Still, this shared nature gives me an icky feeling...Intel does it "right"...four streamlined, well-designed cores with HT to give parallel workloads an extra boost. AMD's "MOAR CORES" paradigm can't work forever; not all of us are running database/compute tasks on our desktops.
On a price/performance scale, though, Interlagos seems to win hands down, considering the 8CU chip is only $850.
 
bwang said:
Still, this shared nature gives me an icky feeling...Intel does it "right"...four streamlined, well-designed cores with HT to give parallel workloads an extra boost. AMD's "MOAR CORES" paradigm can't work forever; not all of us are running database/compute tasks on our desktops.
On a price/performance scale, though, Interlagos seems to win hands down, considering the 8CU chip is only $850.
Click to expand...

Intel's HT couldn't be further from streamlined. BD cores are streamlined. HT is an exercise in redundant circuitry, in fact in some benchmarks Intel's CPUs perform better with HT turned off. Also HT has other issues like security vulnerabilities.

AMD saw this and said for little more transistor count we can just get another core.

Also I am sorry but "MOAR CORES" is the future in case you somehow missed it. Sandy Bridge is about up to 10% faster IPC wise than Nehailem, and Nehailem is about 10% IPC faster than core 2 duo. Only way Intel themselves can keep up is by adding more cores.
 
Last edited:
sirmo said:
Intel's HT couldn't be more from streamlined. BD cores are streamlined. HT is an exercise in redundant circuitry, in fact in some benchmarks Intel's CPUs perform better with HT turned off. Also HT has other issues like security vulnerabilities.

AMD saw this and said for little more transistor count we can just get another core.

Also I am sorry but "MOAR CORES" is the future in case you somehow missed it. Sandy Bridge is about up to 10% faster IPC wise than Nehailem, and Nehailem is about 10% IPC faster than core 2 duo. Only way Intel themselves can keep up is by adding more cores.
Click to expand...

MOAR CORES only works on servers and in HPC, where tasks are easily parallelized.
On desktops, if you are a avid video/photo/rendering type of guy, MOAR CORES is fine. For the typical enthusiast, who builds a nice rig for gaming, additional cores are of questionable use...games don't yet scale across arbitrary numbers of cores yet, and it is uncertain whether they ever will.
Right now, even the 6-core Gulftowns are overkill for typical enthusiast use, let alone the 8/12/16/whatever cores AMD can give you.
Then again, I feel that since Opteron, AMD has been engineering for servers first, then trickling the improvements down to the desktop market. The desktop side costs no R&D (since it has already been done for the server side), and by pricing competitively, AMD can get a market share (even if their offerings are a bit slower than Intel's).
What disappoints me is the lack of floating point performance on Bulldozer. Its understandable for servers (which are mostly integer) but makes Bulldozer pretty sad in HPC applications :(
 
bwang said:
MOAR CORES only works on servers and in HPC, where tasks are easily parallelized.
On desktops, if you are a avid video/photo/rendering type of guy, MOAR CORES is fine. For the typical enthusiast, who builds a nice rig for gaming, additional cores are of questionable use...games don't yet scale across arbitrary numbers of cores yet, and it is uncertain whether they ever will.
Right now, even the 6-core Gulftowns are overkill for typical enthusiast use, let alone the 8/12/16/whatever cores AMD can give you.
Then again, I feel that since Opteron, AMD has been engineering for servers first, then trickling the improvements down to the desktop market. The desktop side costs no R&D (since it has already been done for the server side), and by pricing competitively, AMD can get a market share (even if their offerings are a bit slower than Intel's).
What disappoints me is the lack of floating point performance on Bulldozer. Its understandable for servers (which are mostly integer) but makes Bulldozer pretty sad in HPC applications :(
Click to expand...

As a software developer, fellow software developers are lazy. The future is horizontal scaling and multithreaded apps. The reason why most games are pretty bad at leveraging n-cores is laziness or incompetence.

PS3 developers had to adapt to the multithreaded model long time ago. Software should leverage all cores and I imaging more and more software will in the future if it aims at utilizing all the system resources. 26% of gamers (steam survey) run quad core systems today, and this number will only increase. Otherwise we're stuck with 10%-20% performance improvement every few years.
 
sirmo said:
As a software developer, fellow software developers are lazy. The future is horizontal scaling and multithreaded apps. The reason why most games are pretty bad at leveraging n-cores is laziness or incompetence.

PS3 developers had to adapt to the multithreaded model long time ago. Software should leverage all cores and I imaging more and more software will in the future if it aims at utilizing all the system resources. 26% of gamers (steam survey) run quad core systems today, and this number will only increase. Otherwise we're stuck with 10%-20% performance improvement every few years.
Click to expand...

i'm sure it has to do with the fact that dx 9 had a strangle hold on the market for so long because of the length of time it was on the market.

As windows xp goes bye bye i'm sure we will see tons more apps take advantage of multiple cores
 
I think it also has to do with the fact that a lot of the games we see today (such as Crysis) were released at around the same time as Conroe, back when single-core was still acceptable and dual-core was what your typical enthusiast had.
Remember those days when games were all single threaded?
 
Interesting that:

A. There isn't an Opteron 6282
B. There is no SE part clocked at 2GHz

Why do people make this stuff up?
 
ppl are making crap up because there's no real info out yet to dispute it. Plus it probably helps with ad dollars till everyone finds out they're fake
 
Last edited:
You must log in or register to reply here.
Back
Top