Computing comparison in BOINC for WCG between E5-2683 Intel and Ryzen 1800x

[Markfw]

So, my E5-2683 V3 Intel 14 core, 28 thread, 2.5 ghz systems do a WCG "unit" (what you see in simple view) in 5.5-6 hours. My Ryzen 1800x@4.0 8 core 16 thread box will do its first full unit (without rebooting) in 6 hours. WOW, thats a winner IMO. I will have to watch it in the future, and see how it averages.

Just for those of you thinking of getting a Ryzen for DC apps. Also, I have a post about how the Ryzen keep the CPU@100% where the Xeon go down to 80% usage all the time, every minute or so.

Here:

That is a E5-2683 v3 (14 cores and 28 threads)

That is my Ryzen7 1800x@3975. Notice that the CPU there is constant 100% where the Xeon has dips ?

 

[crashtech]

Nice to see good results from Ryzen, though I would not blame dips in utilization on the CPU without more info, seems like the platform or the process is having trouble keeping the CPU fed properly for whatever reason.

 

[Markfw]

OK, Now I feel really stupid. While looking for the interval I noticed that I had selected "use at most 90% of the cpu time". Once I set it to 100%, it now looks like the Ryzen.

 

[crashtech]

OOPS! I guess that might change the comparison a bit, though I think Ryzen is shaping up to be a pretty efficient performer for DC.

 

[StefanR5R]

Ah, I wouldn't have thought of this... I often use the "Use at most ... % of the CPUs" setting from the advanced view, but never had a need for the "... % of CPU time" setting.

Looking forward to device stats comparison. I guess it can take 2 or 3 days or so for task validations at WCG to catch up with your production.

 

[Markfw]

Ah, I wouldn't have thought of this... I often use the "Use at most ... % of the CPUs" setting from the advanced view, but never had a need for the "... % of CPU time" setting.

Looking forward to device stats comparison. I guess it can take 2 or 3 days or so for task validations at WCG to catch up with your production.

Well, the "average" number isn't corrct for weeks it seems. I can look at how much more I do total, but just have not even had a full day yet.

But the system has 23 hours uptime, and it has 17,000 points. I think I got a few thousand before I got it stable, but its most likely close to the E5-2683 systems.

 

[Markfw]

Looks like between 13 and 14 thousand. I will let you know at noon tomorrow. 12,000 is the most of any of my E5-2683 v3's

 

[Markfw]

13,000 give or take a couple of hundred, so it beats my best E5-2683 by at least one thousand !

 

[crashtech]

How about power at the wall?

 

[Markfw]

How about power at the wall?

5 case fans (2 of them on the H100i) and one 550ti (not doing anything) 225 watts. overclocked to 4.0

 

[StefanR5R]

13,000 give or take a couple of hundred, so it beats my best E5-2683 by at least one thousand !

Whow, from what I can tell, that's... a lot! Not to mention the bang for the buck.

My 4-core Haswell Xeon E3 is doing WCG Mapping Cancer Markers currently: 5,500 PPD

Back in January, my 2x14-core Broadwell Xeon E5s did OpenZika with 58,000 PPD. Deduct maybe 10 or 15 % from that for their yield in Mapping Cancer Markers, and they still come out a little bit ahead of Ryzen in PPD/core, but not by much --- and at what system costs!

 

[crashtech]

5 case fans (2 of them on the H100i) and one 550ti (not doing anything) 225 watts. overclocked to 4.0

How does that compare to the 2683?

 

[StefanR5R]

5 case fans (2 of them on the H100i) and one 550ti (not doing anything) 225 watts. overclocked to 4.0

I guess if you went back to stock (with auto boost and XFR), PPD will decrease just a little but power usage goes way down. At stock, the internal undervolting would kick in and work its powersaving magic, to some extent probably even under 100 % WCG load.

 

[Markfw]

Mine is setup as WCG default, which has Open Zika, Mapping cancer markers, childhood cancer, and several others all combined. Its the MIX that I say is beating my 14 core-28 thread Xeons (the V3, are those haswell ? or broadwell)

All 5 of my WCG boxes are configured the same, and now doing somewhere of 60k ppd for all 5.

 

[StefanR5R]

Xeon E3/5/7 v3 = Haswell, v4 = Broadwell.

 

[TennesseeTony]

I guess I'd better get back to WCG soon before you pass me.