Page 6 of 15
Re: Nvidia Pascal GPU cards???
Posted: Mon Jun 06, 2016 9:30 pm
by Torin3
foldy wrote:@Nebulae or Torin3: I'm interested in GTX 1080 pcie bus usage. Can you check the GPU pcie bus usage while folding with e.g. msi afterburner?
Running between 32% and 42%, fairly even spread between the two end.
This is on a 9704 WU, getting 745K ppd.
9160 WU is between 38% and 40%. Very little variation. 800K ppd.
Re: Nvidia Pascal GPU cards???
Posted: Mon Jun 06, 2016 10:53 pm
by bruce
Torin3: What speed is your PCIe bus?
Re: Nvidia Pascal GPU cards???
Posted: Mon Jun 06, 2016 11:02 pm
by Torin3
It is a x16 3.0 slot running at x16 3.0 according to GPU-Z
Re: Nvidia Pascal GPU cards???
Posted: Sat Jun 11, 2016 12:26 pm
by foldy
More FahBench 1.2 explicit results in ns/day
GTX 1080 OC _ 113 for 700$
GTX 1070 OC _ 103 for 450$
GTX Titan X _ _ 84
http://arstechnica.com/gadgets/2016/06/ ... 70-review/
I think GTX 1070 OC will be the new dollar per PPD King and all at 200 Watt.
Re: Nvidia Pascal GPU cards???
Posted: Sat Jun 11, 2016 2:33 pm
by mihapiha
http://abload.de/img/0044la5g.png
I'm testing right now on a EVGA GTX1080 FE with EK-Waterblock installed.
However the load is at only 89%. I don't know what I have to do to get it to 100%....
Mod edit: Changed overly large image link to an URL
Re: Nvidia Pascal GPU cards???
Posted: Sat Jun 11, 2016 2:45 pm
by foldy
Max GPU load shown on my PC was 94% and this is really heavy on GPU.
What CPU do you have? Maybe you can pause CPU folding and only fold on GPU to see if this makes a difference in GPU load % ?
If it does then you need to reduce your CPU slot max core count to an even number because GPU needs a CPU core to support it.
Re: Nvidia Pascal GPU cards???
Posted: Sat Jun 11, 2016 5:59 pm
by bruce
mihapiha wrote:However the load is at only 89%. I don't know what I have to do to get it to 100%....
You probably can't.
Data must be transferred between main RAM and the GPU before the shaders can work on it. A certain amount of time is spent transferring data, some of which may be overlapped with processing and some of which cannot. As you increase the speed of the GPU without increasing the speed of the PCIe bus, the percentage of non-overlapped time increases. and the GPU isn't as busy because it's finishing what it can to faster.
Re: Nvidia Pascal GPU cards???
Posted: Sat Jun 11, 2016 6:04 pm
by mihapiha
bruce wrote:mihapiha wrote:However the load is at only 89%. I don't know what I have to do to get it to 100%....
You probably can't.
Data must be transferred between main RAM and the GPU before the shaders can work on it. A certain amount of time is spent transferring data, some of which may be overlapped with processing and some of which cannot. As you increase the speed of the GPU without increasing the speed of the PCIe bus, the percentage of non-overlapped time increases. and the GPU isn't as busy because it's finishing what it can to faster.
It folded just one WU with the CPU, because it automatically launched one. So I didn't have a choice and let it finish. I have a Core i7-4930k @ 4.5 GHz supported by DDR3-2333 MHz, and now it's doing absolutely nothing while folding, yet my GPU load doesn't surpass 89%...
I have the BCLK on my motherboard set to 125 MHz as well. Too bad, I thought I had the ideal setup to fold even faster. Seems I just hit the limitations.
Re: Nvidia Pascal GPU cards???
Posted: Sat Jun 11, 2016 6:20 pm
by bruce
On modern systems, PCIe data transfers are not limited by a lack of CPU resources. The bus speed has it's own limitations.
With a system like that, you should be folding with your CPU, too.
Re: Nvidia Pascal GPU cards???
Posted: Sat Jun 11, 2016 6:59 pm
by foldy
mihapiha wrote: Too bad, I thought I had the ideal setup to fold even faster. Seems I just hit the limitations.
It also depends which FahCore and work unit you are currently running.
e.g. FahCore_21 with 13xxx work unit really has high usage like 95% while other cores or work units can have below 90%.
Re: Nvidia Pascal GPU cards???
Posted: Tue Jun 14, 2016 12:31 am
by _r2w_ben
mihapiha wrote:It folded just one WU with the CPU, because it automatically launched one. So I didn't have a choice and let it finish. I have a Core i7-4930k @ 4.5 GHz supported by DDR3-2333 MHz, and now it's doing absolutely nothing while folding, yet my GPU load doesn't surpass 89%...
I have the BCLK on my motherboard set to 125 MHz as well. Too bad, I thought I had the ideal setup to fold even faster. Seems I just hit the limitations.
p11430 has a relatively low number of atoms compared to other FahCore_21 projects so it may be harder to evenly allocate work across the entire GPU. You can see the atom counts on
psummary.
Re: Nvidia Pascal GPU cards???
Posted: Sun Jun 19, 2016 12:15 pm
by X1900AIW
Short feedback from a mixed running of a new 1070, FAH Client 7.4.4. and BOINC, Win8.1,
Eco mode by Gigabyte ´Etreme Gaming´ tool. [No overclocking, no default setting, only eco clock so far.]
Recognition was done fast, folding started at once, finished some x18/x21 workunits, everything fine.
No more FAHMon support, tried it, but sorry, no detailed PPD information (made screenshots with ~510-530k PPD, 415k with the current P10489), it can be calculated with the period time from the log.
Code: Select all
*********************** Log Started 2016-06-18T15:58:57Z ***********************
...
15:58:57: GPU 0: UNSUPPORTED: NV3 [PCI]
15:58:57: CUDA: 6.1
15:58:57: CUDA Driver: 8000
...
15:59:00:Updated GPUs.txt
15:59:00:Read GPUs.txt
15:59:00:Trying to access database...
15:59:00:Successfully acquired database lock
15:59:00:Enabled folding slot 01: READY gpu:1:GP104 [GeForce GTX 1070]
...
16:01:59:WU00:FS01:0x18:*********************** Log Started 2016-06-18T16:01:59Z ***********************
16:01:59:WU00:FS01:0x18:Project: 9152 (Run 8, Clone 15, Gen 152)
...
16:02:24:WU00:FS01:0x18:Completed 0 out of 2500000 steps (0%)
...
17:46:04:WU00:FS01:0x18:Completed 2500000 out of 2500000 steps (100%)
...
17:48:00:WU00:FS01:Final credit estimate, 37745.00 points
...
18:43:57:WU00:FS01:0x18:*********************** Log Started 2016-06-18T18:43:57Z ***********************
18:43:57:WU00:FS01:0x18:Project: 9161 (Run 268, Clone 0, Gen 38)
...
18:44:24:WU00:FS01:0x18:Completed 0 out of 2500000 steps (0%)
...
20:35:20:WU00:FS01:0x18:Completed 2500000 out of 2500000 steps (100%)
...
20:36:09:WU01:FS01:0x21:*********************** Log Started 2016-06-18T20:36:09Z ***********************
20:36:09:WU01:FS01:0x21:Project: 13131 (Run 68, Clone 9, Gen 0)
...
20:36:34:WU00:FS01:Final credit estimate, 34553.00 points
...
20:38:14:WU01:FS01:0x21:Completed 0 out of 1600000 steps (0%)
...
23:19:05:WU01:FS01:0x21:Completed 1600000 out of 1600000 steps (100%)
...
23:19:16:WU00:FS01:0x21:*********************** Log Started 2016-06-18T23:19:15Z ***********************
23:19:16:WU00:FS01:0x21:Project: 11407 (Run 1, Clone 0, Gen 198)
...
23:19:25:WU00:FS01:0x21:Completed 0 out of 5000000 steps (0%)
...
23:22:14:WU01:FS01:Final credit estimate, 49027.00 points
...
02:50:23:WU01:FS01:0x21:*********************** Log Started 2016-06-19T02:50:23Z ***********************
02:50:23:WU01:FS01:0x21:Project: 9211 (Run 19, Clone 1, Gen 20)
...
02:50:53:WU00:FS01:Final credit estimate, 84327.00 points
...
02:51:01:WU01:FS01:0x21:Completed 0 out of 2500000 steps (0%)
...
06:58:15:FS01:Paused
...
06:59:53:FS01:Unpaused
06:59:53:WU01:FS01:Starting
...
06:59:53:WU01:FS01:0x21:*********************** Log Started 2016-06-19T06:59:53Z ***********************
06:59:53:WU01:FS01:0x21:Project: 9211 (Run 19, Clone 1, Gen 20)
...
11:09:39:WU01:FS01:0x21:Completed 2500000 out of 2500000 steps (100%)
...
11:09:46:WU00:FS01:0x18:*********************** Log Started 2016-06-19T11:09:46Z ***********************
11:09:46:WU00:FS01:0x18:Project: 10489 (Run 0, Clone 104, Gen 500)
...
11:10:09:WU00:FS01:0x18:Completed 0 out of 5000000 steps (0%)
...
11:18:31:WU01:FS01:Final credit estimate, 160004.00 points
11:18:31:WU01:FS01:Cleaning up
11:19:58:WU00:FS01:0x18:Completed 200000 out of 5000000 steps (4%)
11:22:21:WU00:FS01:0x18:Completed 250000 out of 5000000 steps (5%)
11:24:52:WU00:FS01:0x18:Completed 300000 out of 5000000 steps (6%)
11:27:15:WU00:FS01:0x18:Completed 350000 out of 5000000 steps (7%)
11:29:47:WU00:FS01:0x18:Completed 400000 out of 5000000 steps (8%)
11:32:09:WU00:FS01:0x18:Completed 450000 out of 5000000 steps (9%)
11:34:32:WU00:FS01:0x18:Completed 500000 out of 5000000 steps (10%)
11:37:04:WU00:FS01:0x18:Completed 550000 out of 5000000 steps (11%)
11:39:28:WU00:FS01:0x18:Completed 600000 out of 5000000 steps (12%)
11:42:00:WU00:FS01:0x18:Completed 650000 out of 5000000 steps (13%)
11:44:24:WU00:FS01:0x18:Completed 700000 out of 5000000 steps (14%)
11:46:48:WU00:FS01:0x18:Completed 750000 out of 5000000 steps (15%)
11:49:20:WU00:FS01:0x18:Completed 800000 out of 5000000 steps (16%)
11:51:44:WU00:FS01:0x18:Completed 850000 out of 5000000 steps (17%)
11:54:16:WU00:FS01:0x18:Completed 900000 out of 5000000 steps (18%)
11:56:40:WU00:FS01:0x18:Completed 950000 out of 5000000 steps (19%)
11:59:04:WU00:FS01:0x18:Completed 1000000 out of 5000000 steps (20%)
12:01:37:WU00:FS01:0x18:Completed 1050000 out of 5000000 steps (21%)
12:04:01:WU00:FS01:0x18:Completed 1100000 out of 5000000 steps (22%)
Function and stability was tested with
FAHBench 2.2, there were very different gpu usages according to the workunit:
Code: Select all
ECO (PT -50)
76.6743 - dhfr - GPU clock result: 1974
397.426 - dhfr implicit - GPU clock result: 1734
12.7859 - nav - GPU clock result: 1721
OC (PT +11, TT +9, GPU +100)
82.1734 - dhfr - GPU clock result: 2034
483.528 - dhfr implicit - GPU clock result: 2025
14.5883 - nav - GPU clock result: 2038
Re: Nvidia Pascal GPU cards???
Posted: Sun Jun 19, 2016 1:55 pm
by toTOW
You're living in another century, FAHMon support stopped a long time ago (it never supported v7 client). Use HFM instead : viewtopic.php?p=285846#p285846
Re: Nvidia Pascal GPU cards???
Posted: Mon Jun 20, 2016 6:49 pm
by Sn1ken
This is my first WU using 1070.
http://i.imgur.com/KSlgTDA.jpg
Mod edit: Changed overly large image link to an URL
Re: Nvidia Pascal GPU cards???
Posted: Mon Jun 20, 2016 8:08 pm
by foldy
Which one is faster? What vendors and models?