Feeds

back to article Nehalems make like elephants on HPC memory test

Intel's Nehalem EP chip has significantly out-peformed AMD's Istanbul on a set a memory-intensive benchmark tests. The techies at supercomputer cluster maker Advanced Clustering Technologies are at it again, running their own benchmarks on single server nodes using popular high-performance computing tests normally used on …

COMMENTS

This topic is closed for new posts.
Bronze badge

Crime does pay

Not surprising. With Intel having used illegal tactics to limit AMD's sales, AMD just hasn't had as much money to throw at research. Pity the fines the EU will probably be levying won't be partially paid to AMD to make up for Intel's crimes.

0
0
Paris Hilton

good grief

Benchmarks that show that one brand is good at X, one good at Y- you'd almost think these ones weren't done by total shills and fanbois or something, and maybe they even had some clue what they were doing...

Excuse me, I'm going to have to have a little sit down, I've come over all previous...

0
0
Happy

Memories

Reminds me of a quick visit to Motorola Phoenix in Feb 1983, a stopover en route to SFO. The reason need not concern us, but Moto knew that I had been involved in CPU design (including an early MMU) and was by 1983 managing the design of an innovative product that used the 68000. After a very brief meeting in which they answered my question (an answer that had eluded Moto Munich and a UK agent), I was shown into a lecture theatre. The audience was very small, and they showed a video of a new CPU design. We then discussed not features but likely performance - which would be limited by memory bandwidth and the small size of on-chip cache then possible. No NDA. I left to drive to the airport - and it rained.

Some months later I received an invite to the London launch of the 68020. On attending, I realised that the video shown in Phoenix was not the 68020 but the next generation. The 68020 had an instruction cache, and received wisdom later was that its performance was not as good as hoped, and that the next one, the 030 with both data and instruction caches and on-chip MMU, went much better - but by then I was doing something different and never got to grips with the 68K family again.

0
0
E 2
Pint

And then lesson is

...pick your hardware to match the characteristics of your software!

0
0

Time to requote

Well dang - I just got four HP bladeservers with Istanbul chips quoted, guess I'll have to get the Nehalem equivalent quoted from HP and Dell as well!

0
0
Bronze badge
Stop

Useless

So they are testing two systems with different processors (With different core counts), different memory and they come to the conclusion that they perform different, how strange.

I am also getting tired of seeing benchmarks between the two architectures, this is no longer the old days when there was was very little to differentiate the two companies. The processors are very different now and each performs better than the other in some areas and is slower in others. The Nehalems are better at multi-threaded processing (Such as virtualization, running multiple applications, etc.) where the Istanbuls run much better with application that need raw power and have only minimal threads (Such as super-computer), Benchmarking has become very irrelevant in the past few years (and in my opinion, it was never relevant). Benchmarks are now just useless pissing contests.

0
0

No shit, Holmes

using 1.6 times faster memory will yield 1.6 times more throughput. I am surprised they didn´t need Gartner to find this...

0
0
Bronze badge

Metaphor Failure

Surely the elephant metaphor is all about the length of retention of memory and not about it's speed. The Alheimer's metaphor is almost as bad.

Oh - and as many people are no doubt asking, outside the esoteric area of supercomputing, what does this mean in the real world? No doubt something, but benchmarks closer to what is run in the average IT shop might be more useful.

0
0
Bronze badge
Thumb Down

bah

Who cares about the bandwidth, it's high enough already. Gimme small latency, now!

0
0
Anonymous Coward

I wonder....

The AMD has 20k with DDR2 vs 7k on the Intel with DDR2. Now the DDR3 Intels have 37k & 32k and this is impressive why ? They have twice the memory bandwidth with memory that offers even more.

I wonder what the page stall is like. the latency on DDR3 is high.

Oh well. Don't want to sound like a fanboi.

0
0
Boffin

Shocked!

So, in conclusion...

DDR3 is faster than DDR2?

I am totally shocked!

0
0

Fact Check Please?

Doesn't the amount of system RAM used impact the STREAM benchmark results, systems with more RAM (in this case the AMD) end up doing more work to solve a larger problem-space than those with less?

HZ

0
0

when it finally gets a round tuit.

@Bronek Kozicki

Only logged in to make the point that you already have, but then had to post, you know, its a synchronous extended data operation thing.

If I'd had lower latency I could have gotten some coffee instead.

When are we going to get some decent RAM in our computers???

0
0
This topic is closed for new posts.