• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

TPU's F@H Team

Joined
Oct 17, 2020
Messages
64 (0.35/day)
Location
United States
Yes, but can I squeeze even more points if I couple strongest GPU's to stronger CPU's?
On my testbench I am currently running a 1080 and 1080 Ti side by side on a Phenom II 955. I think that comes with a penalty in PPD.
I was thinking about swapping the 1080 Ti with another plain 1080, but then i have to be sure the 1080 Ti will fit the other PC case.

And look out for @mstenholm, the next couple of days he is going to make Top200 rank !
I've decided that single-thread execution speed is the key factor for Folding GPU/CPU and of course, sufficient cores available to feed & control the GPU.
I recently upgraded a dual Xeon folder CPUs and saw a definite increase in Folding efficiency.
I did a blog post about the transition:
 
Joined
Oct 24, 2020
Messages
121 (0.68/day)
Location
Belgium
System Name Asus 6700k
Processor i7-6700k
Motherboard Asus prime Z270-P
Cooling Bequiet Pure Rock
Memory 16GB Kingston HyperX 2133 C14 Fury Black
Video Card(s) ASUS ROG STRIX GTX 1080 Ti GAMING OC
Storage Samsung 960 Evo NVMe
Display(s) Medion 23'
Case Antec Three Hundred v.2
Audio Device(s) onboard
Mouse Logitech Trackman T-BB18
Software Windows 7 , FSX
I have a PC, basically an hp i5 bussiness machine with a GTX 1080, getting more than double PPD than my ex-gaming PC with a 1080 Ti (the one in my sig), doing the same type of work units? :kookoo:
The 1080 Ti core and memory clock are running at max frequency, but the card itself isn't hitting temperature or power limits.

Upon further inspection, I see the memory controller load on the GTX 1080 is relatively higher in GPU-Z (1080 memory controller load 60% vs 1080 Ti memory controller load 20%).
Overall the 1080 Ti should be better in absolute terms than a plain 1080 ? It looks as if it just sits there, wasting cycles at high clock speeds.

The 1080 is taking 3h to finish a WU , the 1080 Ti is going to take 5h.
After finishing the WU, I decided to reboot the machine, because the next WU was also going to take to much time.
After reboot, the memory controller load on the 1080 Ti went up to 50% and TPF dropped considerably.

It's worth to check on TPF once in a while and reboot if necessary when the GPU gets locked in an unfavourable state and is not performing as expected.
 
Last edited:
Joined
Aug 21, 2015
Messages
264 (0.13/day)
Location
North Dakota
System Name Oswald
Processor i7-9700K
Motherboard MSI Z370-A PRO
Cooling be quiet! Pure Rock 2
Memory 16GB Patriot DDR4-3200
Video Card(s) Sapphire RX 470 8GB
Storage Samsung SM961 256GB, Toshiba DT01ACA200
Display(s) Sharp 46" TV
Case Antec Sonata SOLO
Power Supply Seasonic SSR-450RM
Mouse Logitech M185
Keyboard Some wireless Microsoft thing with a trackpad
Software Windows 10 Home
Benchmark Scores Cinebench R15: 1398
Check it out, I'm about to join the 1% :D

1618255221345.png
 
Joined
Oct 24, 2020
Messages
121 (0.68/day)
Location
Belgium
System Name Asus 6700k
Processor i7-6700k
Motherboard Asus prime Z270-P
Cooling Bequiet Pure Rock
Memory 16GB Kingston HyperX 2133 C14 Fury Black
Video Card(s) ASUS ROG STRIX GTX 1080 Ti GAMING OC
Storage Samsung 960 Evo NVMe
Display(s) Medion 23'
Case Antec Three Hundred v.2
Audio Device(s) onboard
Mouse Logitech Trackman T-BB18
Software Windows 7 , FSX
With buying new or used GPU's now out of the picture, i am looking at replacing some systems that are still running on 2011's athlon/phenom processors.
I already have some lower end Skylake i5 systems running, with good results, so I am thinking about getting more of these.
 
Joined
Oct 17, 2020
Messages
64 (0.35/day)
Location
United States
That's a good approach. With CPU attached dual PCIe x16 slots running at x8 you can get good performance.
This should give your single-thread execution speed a generous boost over that vintage gear.
The other approach would be eBay server-class systems and mining frames, which is what I did. These systems can support several GPU per frame with PCIe x16 or x8 lanes.
I have surplus slot capacity now for Folding.
Edit:
I should add that I saw no advantage for PCIe4 over PCIe3 with the RTX3070 if anyone was wondering about an architecture update.
The recently announced Nvidia Grace architecture made me think that it was an effort to eliminate the bottlenecks in current architectures that exhibit results as I had seen.
Perhaps Nvlink and Grace-like GPU connections will be the next quantum leap in compute and gaming, certainly, PCIe4 wasn't it for Folding.
 
Last edited:
Top