Xeon X5670 (dual CPU) Very Slow

Andrew
Andrew
Joined: 25 Nov 17
Posts: 3
Credit: 100229260
RAC: 0
Topic 219273

Hi everyone, I hope this is the right forum. 

I recently bought an old workstation with dual Xeon x5670 CPUs. It has 60 gigs of RAM and an RX 570. They are a bit old, but that is 12 cores total. It is running Einstein at home much slower than I thought it would. For a 144,000 gigaflop CPU task it crawls at 2.1 percent per hour. My 7820hk in my laptop at the same clock rate goes at alittle over 14 percent per hour and even taking into account the cores the laptop CPU is faster overall. Hyperthreading is enabled on both. 

I have tried the steam VR test which came back fine with only 12 frames CPU bound and the CPUZ benchmark which showed multicore performance to be easily over twice the speed of the 7820hk. 

The RX570 is working very quickly. 

Anyone have any ideas on what might be wrong or is this a bad CPU for Einstein at home? 

mikey
mikey
Joined: 22 Jan 05
Posts: 11888
Credit: 1828057366
RAC: 206496

Andrew wrote:Hi everyone, I

Andrew wrote:

Hi everyone, I hope this is the right forum. 

I recently bought an old workstation with dual Xeon x5670 CPUs. It has 60 gigs of RAM and an RX 570. They are a bit old, but that is 12 cores total. It is running Einstein at home much slower than I thought it would. For a 144,000 gigaflop CPU task it crawls at 2.1 percent per hour. My 7820hk in my laptop at the same clock rate goes at alittle over 14 percent per hour and even taking into account the cores the laptop CPU is faster overall. Hyperthreading is enabled on both. 

I have tried the steam VR test which came back fine with only 12 frames CPU bound and the CPUZ benchmark which showed multicore performance to be easily over twice the speed of the 7820hk. 

The RX570 is working very quickly. 

Anyone have any ideas on what might be wrong or is this a bad CPU for Einstein at home? 

There is probably nothing wrong, as you said it's not a newer cpu and the gpu is running just fine. What I do is crunch using the gpu on projects that can use them and use the cpu's on projects that don't have gpu apps. That lets me get credits in 2 projects at once.

If you want to speed up the gpu a little bit try running 2 workunits at once and see if that works better. Each one may slow down a little bit but since you are crunching two at a time it should be faster overall.

Richie
Richie
Joined: 7 Mar 14
Posts: 656
Credit: 1702989778
RAC: 0

How many cpu tasks are you

Hi!

How many cpu tasks are you running at once? Which application?

What's the exact model of the motherboard?

Have you monitored cpu temps while crunching?

Andrew
Andrew
Joined: 25 Nov 17
Posts: 3
Credit: 100229260
RAC: 0

I have tried it with all 24,

I have tried it with all 24, and 23 with no difference. It's just whatever the standard BOINC is that you download. 

I'll look up the model when I get home. 

Yes, and they do get too warm. I brought it back to 50 percent CPU time and it is still slow compared to the 7820hk also at the same clock speed and percentage usage, but maybe this is the problem. At 50 percent they go from 50 to 70 celcius. 

Andrew
Andrew
Joined: 25 Nov 17
Posts: 3
Credit: 100229260
RAC: 0

I didn't know you could run

I didn't know you could run two. Does that increase your temps? If so I might wait until the end of summer. How do you do this? 

Richie
Richie
Joined: 7 Mar 14
Posts: 656
Credit: 1702989778
RAC: 0

Andrew wrote:I have tried it

Andrew wrote:
I have tried it with all 24, and 23 with no difference.

Einstein CPU tasks from these current apps are not the lightest to run for a system, compared with some other projects. My experience with LGA1366 systems with single processor (also X5670) suggests that running more than 6 of these tasks (HT enabled) the additional gain will quickly be small. Tasks will run slower of course plus there will be even more heat.

If I remember correct those dual processor LGA1366 systems in a way don't have double the 'internal bandwidth'. Because of that, maximum stress might cause them get bottlenecked relatively more quickly.

I would keep the CPU time at 100% but reduce the concurrent tasks count to something like 12-16. There's a good chance the total output might not be much different from current situation.

It could also be a good idea to use Process Lasso to see how the system really allocates the tasks to those processors. If the system for some reason tends to give them a very uneven load, then Process Lasso could be used to adjust how the load is spread... and test if that would make any difference.

Joseph Stateson
Joseph Stateson
Joined: 7 May 07
Posts: 173
Credit: 2921660969
RAC: 1448075

I have similar system and was

I have similar system and was running several RX-570 up until recently.  You cannot compare CPU task completions to GPU.  Always slower.  I suggest you run an assortment of projects that are CPU bound to take advantates of the 24 threads and run two GPU apps concurrently on your RX570.

My system is in the garage with ambient  temps about 95 (Fahrenheit)   Even an open mining rack with risers did not provide enough cooling and I converted to liquid with LCD showing 43c at the pump intake.  If temps a problem disable hyperthreading. If  your temps go above 83c then the CPU starts throttling down.

With that much RAM you can leave non-GPU tasks in memory while suspended.  I recommend World Community Grid (have 4 different CPU tasks) also NumberFields, LHC, and Rosetta.  If you are a member of gridcoin then these projects are whitelisted unless something has changed recently.  Set Einstein to use 1/2 gpu and same for Milkyway.  with Einstein at higher priority.  that way you get two tasks running unless both project go offline for maintenance at same time.

Mad_Max
Mad_Max
Joined: 2 Jan 10
Posts: 153
Credit: 2134779636
RAC: 446606

I think Xeon x5670 performs

I think Xeon x5670 performs relatively slow due to lack of AVX support (+ generally very old architecture - it 1st generation of iCore, while 7820hk is 7th).

AVX very actively used in E@H CPU apps if CPU supports it. I see similar low performance on CPUs without AVX.

For example 1 GW WU takes about 60-70к sec on 1 CPU core of AMD Phenom II X6.
It about on par with your x5670 taking about 110к sec per GW (considering it actually running up to 2 WUs per each  CPU core due to HT).

And 7820hk is working "only" about 2 times faster - it runs GW tasks about 60к sec each: https://einsteinathome.org/ru/host/12784947/tasks/0/49

So you probable have mistaken in original comparison (may be look for WUs of different types).
~100% (х2) speed up at same CPU clock up is pretty normal for 7th generation vs 1st and AVX on vs AVX off. 

 

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.