PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
1) Message boards : Generalized Fermat Prime Search : GFN21 GFN22 Scoreboard, post your times/hardware (Message 120718)
Posted 300 days ago by shinozkProject donor
TITAN V: 4 h

Tesla V100-SXM2-16GB: 3 h 07


Titan V may be limited by memory-bandwidth for ocl in GFN21.
GPU-Z
GPU-Z shows that PrefCap-Reason is nothing and Memory-Controller-Load is maxed out.
Power-Consumption never goes above 60%(=150W).

Tesla(900GB/sec) vs Titan V(653GB/sec) = 3h vs 4h

2) Message boards : Generalized Fermat Prime Search : GFN21 GFN22 Scoreboard, post your times/hardware (Message 120536)
Posted 309 days ago by shinozkProject donor
TITAN V (OC: GPU+65MHz , MEM+175MHz)

GFN21: 13,420 seconds
3) Message boards : Number crunching : LLR 3.8.20 Going Live! (Message 106440)
Posted 846 days ago by shinozkProject donor
For benchmark of multi-threaded LLR, can we use Prime95.exe v28.10?
Both use LLR v3.8.20.

4) Message boards : Number crunching : ram, rank and AIDA64 (Message 95186)
Posted 1156 days ago by shinozkProject donor
How about third tier timings?

Asus says "the third tier of memory timings – read to read, read to write, write to write and write to read are larger factors in memory performance than the primary timing set."

http://rog.asus.com/forum/showthread.php?51063-The-Haswell-E-Overclocking-Thread-and-Overclocking-Guide

This is the biggest factor of DDR4: RTL
http://overclocking.guide/x99-ddr4-memory-overclocking-guide/
5) Message boards : Number crunching : A new machine (Message 73504)
Posted 1983 days ago by shinozkProject donor
I do not know a detail because I do not have a memory of GSkill.

A overclocking memory may have only a few profiles in XMP.

I use a 2400 memory of Team group.
My memory has only 1333,2133 and 2400 in XMP.

If your memory does not have 1866, you may not use 1866.
6) Message boards : Number crunching : Best CPU for LLR/GFN? (Message 73395)
Posted 1987 days ago by shinozkProject donor
For comparison, I disable 2 cores of 4960x in BIOS to make 4-core system because I don't have 4-core CPU.
The size of L3 cache is diffrent from that of a genuine 4-core sytem.

4960x(3.7GHz, Core 4, HT off)
dual-channel(1333) 2.1hours 65W (http://www.primegrid.com/result.php?resultid=523720913)
quad-channel(1333) 2.1hours 70W (http://www.primegrid.com/result.php?resultid=523762116)


In conclusion,

4960x(3.7GHz, Core 6, HT off)
dual-channel(1333) 3.5hours ; 6.9WU/day * 6Core = 41.4WU
quad-channel(1333) 2.3hours ; 10.4WU/day * 6Core = 62.4WU
quad-channel(2400) 2.2hours ; 10.9WU/day * 6Core = 65.4WU

4960x(3.7GHz, Core 4, HT off)
dual-channel(1333) 2.1hours ; 11.4WU/day * 4Core = 45.7WU
quad-channel(1333) 2.1hours ; 11.4WU/day * 4Core = 45.7WU

The bandwidth of dual-channel memory is exhausted by 4 core.
The productivity for 6 core with dual-channel memory is about same as that for 4 core.
Moreover, quad-channel memory for 4-core system is less important.

7) Message boards : Number crunching : Best CPU for LLR/GFN? (Message 73390)
Posted 1987 days ago by shinozkProject donor
I run SR5 with memory overclocking to 2400cl9.

4960x(3.7GHz, Core 6, HT off)
dual-channel(1333) 3.5hours 85W (http://www.primegrid.com/result.php?resultid=523564454)
quad-channel(1333) 2.3hours 95W (http://www.primegrid.com/result.php?resultid=523629619)
quad-channel(2400) 2.2hours 96W (http://www.primegrid.com/result.php?resultid=523689530)

For current SR5, I conclude that if we use 6 core with stock clock, we must use the quad-channel, and further memory OC is less important.

For a large FFT or CPU overclocking, the memory OC may become imporant.

8) Message boards : Number crunching : Best CPU for LLR/GFN? (Message 73364)
Posted 1987 days ago by shinozkProject donor
The effects of the quad-channel memory become more important for IB-E or IV-E.

I run SR5 with 4960x@stock(IV-E).
When all 6 cores are running, the turbo-boost technology sets the frequency of the 4960x to 3.7GHz.
And I select 1333cl9 for memory clock, because my memory module does not have 1600 in XMP.

The result is :

4960x(3.7GHz,1333cl9, Core 6, HT off)
dual-channel 3.5hours 85W (http://www.primegrid.com/result.php?resultid=523564454)
quad-channel 2.3hours 95W (http://www.primegrid.com/result.php?resultid=523629619)

The watt of the CPU is measured by HWINFO64. The lower watt means a memory starvation.
9) Message boards : Number crunching : Buliding a new PC (Message 69749)
Posted 2112 days ago by shinozkProject donor
CPU tasks(LLR) requires much memory bandwidth.
If you use the LGA 2011 platform, you should a quad-channel momory.

Not 2 * 8Gb but 4 * 4Gb.

10) Message boards : Generalized Fermat Prime Search : Source code of Genefer for OpenCL is available. (Message 69021)
Posted 2143 days ago by shinozkProject donor
I also confirmed that backgroud CPU tasks interfere with GeneferOCL.

For my Titans,
GPU load is 97% without CPU tasks.
GPU load is 65%-77% with CPU tasks.

To change priority of the CPU or GPU processes is no effect.
To run two GPU tasks in single GPU is no effect. GPU load is 85% still.

Maybe, the fastest cards complete a GPU kernel execution within a time slice of sleep(0).

you can enable HyperThread, free logical core 0 and set CPU affitiy of GPU process to core 0, then CPU load is half.


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2019 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 1.46, 1.32, 1.47
Generated 22 Jul 2019 | 10:04:50 UTC