X-bit labs

Started by: RCooley | Date 12/22/05

Comments: 29 | Last Comment:  10/08/06

[1-20 | 21-23]

1. Interesting article.

Did you try to disable the hyper-threading on the Xeon's? Few of the apps used them, and they will cause cache flushes on the primary thread.

Second, a pity you did not include a cheaper system based on a DFI-Lan Party SLI with an AMD ATHLON 64 3800+ X2. Maybe use 4 one gig memory sticks..

2. Given that most software don't know how to load 8 threads properly, would the picture improve substiantially if you disabled Hyper threading for the Dual Core Xeon? That way it can run 4 threads with the full resources of 4 cores in all of the tests.

In many tests, the loads aren't balanceable:
Only two physical cores work -- 4 physical cores(2 P4 DC@2.8G)
But four physical cores work -- 4 physical cores(2 K8 DC@2.2G)

3. I've seen reports suggesting that Hyperthreading hurts performance in some situations. It would be interesting for you to test the dual core Xeon with HT turned off (4 logical processors) to more accurately emulate the dual core Opteron.

4. Overall, the article was OK.

I don't think there is a need for synthetic benchmarks like Sisoft or PCMark as well as games in a workstation article.

I use FLUENT 6.2.16, (Windows and Linux versions), but I'm curious which one did you folks use...64bit one or 32bit one?

As well, which model, solvers/settings did you use? Sometimes, you see one benefit over another.

I don't think there's anything unique about FLUENT's results though, given that engineering apps like these are often properly written to take advantage of multi-core/CPU solutions.

FLUENT also supports clustering. But you need low-latency (expensive) networking to really benefit.

i'm interested to know if muliple FSBs (Dual Channel FSB)
would performa as well or better than AMD Opteron's solution

(FSB reverted to internal communication @ CPU clock speed)

5. Where is power consumption comparation?

6. AMD FoReVer :) by CZECH

7. what video driver version was used from nvidia?

8. Not surprising as to the victory of AMD; it's practically A64 vs P-D 820.

A couple of issues:
1. As above mentioned (and also, I believe, noted when testing the P-XE 840) HT might hurt performance when the apps don't support 8 threads, since they can't distinguish between physical and virtual cores some of the physical cores will be loaded with 2 threads while others will not be loaded at all.
2. Xeon platforms have been supporting SLI before the NF4-SLI chipset came out; although there might not be any such boards supporting dual-core Xeons

9. Note that Opteron memory controller supports 4GB DIMMs, so you could populate the K8WE (S8295) with 32GB , albeit PC2100 ; hence the Iwill DK88 dual Opteron motherboard allows 64GB with 16 sockets, yet, like the Arima SW* series [Serverworks HT* chipsets] , only features PCI-E x8, unlike the Tyan you tested, which offers PCI-E x16.

10. TYAN MASINBOARD WAS INTRODUCED 12 MONTHS AGO. AWAITING WORKSTATION NEXT MONTH WITH 44 LANES OF PCI EXPRESS ON ONE DUAL CORE AMD. ALTHOUGH NOT AS MASSIVE AS TYAN?, MUCH MORE REASONABLE IN COST & JUST AS POWERFUL IS MY GUESSTIMATE.SIGNED:PHYSICIAN THOMAS STEWART VON DRASHEK M.D.

11. Very very good article, except for this final remark.

-----
So, despite the today’s evident fiasco, the future looks pretty rosy for new Intel workstation CPUs.
-----
How can you even make this remark?
It's stupid as you haven't tested this CPU.....
Nothing rosy about it until they can PROVE it works....

Bas.

12. Where's the most important test?

Playing Quake 4 *while* my video renders :)

Back in the day that's why I had a dualie. A little Diablo or Fallout while my video or 3D rendered. Today a good number of games and applications are multi-threaded (or multi-process) so I have no problems making the most of a dual dual-core Opteron box.

Kinda on topic... I've been working on multi-user systems. A P4 with HT on was able to easily run 3DMark + another user running a Spyware scan + IE. Start bumping up the number of cores (and video cards) and you get a system where multiple people are fragging, surfing the web, DVR, watching pre-recorded video, etc.

No core goes unused here! :D

13. Whats not to love about cheaper and much better?
AMD is the shizzle my nizzles lol

14. Conclusion:
AMD 2.2G > P4 2.8G at the most cases.
But in fluent algorithm simulation, P4 > Opteron

15. If we test the dgemm of linear matrix algorithm with the best code lib of I/A U (MKL or ACML), P4@2.8G > Athlon@2.2G.

16. This test is biased to software not supporting key performance enhancing features which makes it meaning less. The Test was AMD vs. Intel not what present applications MAY run faster. If you are a developer, Intel has more capable performance boosting than AMD does.

Don't agree with any of this personally test HT and it out performance AMD, any day.

What "key performance enhancing features" do you mean? What is missing at the Intel plattform in order to achieve more performance?

Your statement is claiming things without concretely saying what you mean...

17. I bought the Tyan MB, the proper power supply, the ram, the processors, video card. When I powered up the machine, I got 4 beeps, then the 4 beeps repeat, I shut the power off, did some power checking, found there was no -5v dc ??

should there be -5v dc ???

what are the 4 beeps, memory ??

there was no post;
power on, fans started to turn, the leds started flasing on the MB, so I know it was checking out the system..no video.

any clues...???

18. How about compilation?

It would be interesting to see how these different systems fare when compiling a large body of C++ code. I write code for a living, and would be interested to know how they compare, currently using an HP xw6200 Xeon based workstation at work.

19. This is excellent. I would love to see how the dual 254 compare with AMD 4800 overclock to 2800, ie, how a dual core processor perform against dual processors running at the same core speed. Thank you.

20. Im a bit confused.Do dual cores r making things slow?

[1-20 | 21-23]