Unsolved
This post is more than 5 years old
1 Message
0
16849
April 30th, 2003 13:00
hyper-threading degrades performance?
Hi have a Precision 530 with dual 2.5 GHz Xeon processors. I run a few CPU intensive programs. Only a couple of them are written to take advantage of dual cpu machines. I recently enabled hyper-threading and have found that programs that use to run up to 50% cpu now only run up to 25% cpu. Obviously this is not an improvement. My question is how much improvement can I expect to see in the programs that do take advantage of multiple cpu's? Should I maybe just forget about hyper-threading?
John
No Events found!


gsx600racer
1 Message
0
April 30th, 2003 15:00
Hey jdehls,
My company just got me a Dell Precision650 with single 3.06 zeon, 2 gig of ddr, nvida quadro4 900xgl video card, 320 scsi. It was ordered with Win 2000 pro, but I ordered it with the hyper threading on so it came shipped with xp pro instead (2000 didnt support hyperthreading.
Well with that said, I noticed a performance increase when I turned hyperthreading off.
When hyperthreading is on, my files open slower, and it hesitates when opening programs, or while rendering 3d drawings, its half a fast.
I thought it was me at first, but im glad I checked out the forums here. It looks like im not the only one with this issue
I hope dell has some answers for this issue, Its depressing to know that my company spent almost 6K (thats incncluding the 20" lcd screen) for this pc, and my home pc with a P4 2.53, 512 rambus, and a nvidia gforce4 and ide hard drive running xp pro as well benchmarks about 5% faster it it only cost about 1K to build.
Don
DKP
25 Posts
0
May 8th, 2003 12:00
Rich-SD
1 Message
0
May 14th, 2003 17:00
DKP
25 Posts
0
May 15th, 2003 20:00
Your not lowering the true CPU load, you are creating extra Ques for threads to request CPU time thru (overly simplied).
You still only have 1 CPU and it's associated resource (Cache, ALU etc)..
Add into this the P4's long pipe, and it's "Guess what I want done Next" branching, if it tries to predict the next action and fails the operation goes to the end of the line..
So, unoptimized code can slow down, Optimized code speeds up (since it understands the way the P4 "thinks" if you would)
skyconsulting
1 Message
0
May 15th, 2003 21:00
Depending upon how W2K displays/calculates processor utilisation, 25% utilization with 2 HyperThreaded processors (="4" processors) could be exactly the same as 50% utilization with 2 processors.
You application is still only using one "real" processor, just that it represents a smaller % of the apparently available processors as W2K now thinks you have 4.
Hope this makes sense
Cheers
Steve
BarbaryLtd
5 Posts
0
February 16th, 2004 13:00
Yes windows hasn't realy got a way of reporting it correctly.
If you run hyperthreading you should see a small drop in performance on single apps but when running manny together it will be faster over all.
I write solvers and I clearly see a 40% improvement when running 4 or more solvers at the same time but if I run one it takes slightly longer.
Yes windows reports it incorrectly. If I run one solver it runs almost as fast as when I run it on a machine with HT switched off. Yet it shows 25% usage not 50%.
There are other problems with HyperThreading too. If you run a 2 Processor machine and run 2 threads low priority it uses all left over cycles. Now run something important in normal priority and you see it get a processor all to itself the two low priority thread being kicked to share the same processor.
With hyperthreading switched on you will run 4 low priority process's. When you now run your app at normal priority it gets a whoel virtual processor to it's self. Which in reality means it's still sharing the processor with 1 (or more) of the low priorty peices of work.
RaymondY
3 Posts
0
February 23rd, 2004 03:00
We have a Precision 450 system with dual processor 3.2 GHz. We have been running our computing-extensive application program on this machine (without using any multiple threads). We tried both enabling and disabling hyper-threading. We did find the system performace wtih hyper-threading on degraded. It was measured by calculating our application program performace, not by looing at the percentage of CPU usage.
--Yiqing
mscjr
7 Posts
0
February 27th, 2004 14:00