GTX 1070ti Review

[Paragon]

Just finished a write-up on the GTX 1070 Ti.

https://greenfoldingathome.com/2019/02/ ... me-review/

One thing I'm wondering is if anyone has found a magical driver for this card for Win10 use. There doesn't seem to be an equivalent to the 372.90 driver for the Ti cards

[MeeLee]

I know sometimes drivers cut back on power consumption, to prevent the card to run hot.
Something that's easy to circumvent with the Linux drivers, or, through the use of overclocking.

Most GTX cards take 100Mhz on CPU overclock, and 500Mhz on vram overclock very well, and can go as high as 150Mhz / 700Mhz overclock with sufficient cooling.

The main factor for folding, aside from power consumption, is temperature management.
Making your system more efficient through the use of ducts or well placed fans.

[Paragon]

Yeah, I have a lot of case cooling to help deal with this (1 x 120 exhaust, 1 x 80 exhaust, PSU exhaust, 2 x 120 intakes for a positive pressure setup). I also cranked up the fan on the card and matched the clock rates to the 1070 I had tested previously (2050 MHz core). I could get the 1070 ti to slightly beat the 1070, but the difference wasn't much. There is a driver component here...the 372.90 drivers which are the best Nvidia folding drivers for the non Ti 10 series cards don't support the 1070ti. Am wondering if there is an equivalent good folding driver for the 1070ti

[MeeLee]

Strange. Usually nVidia drivers are universal,
It's a single driver that works on most of their cards. (unless the driver is too old, and the card didn't exist at the time they made the driver).
I usually stick with the most modern driver available.

[toTOW]

One thing I'm wondering is if anyone has found a magical driver for this card for Win10 use. There doesn't seem to be an equivalent to the 372.90 driver for the Ti cards

Did you try the latest ones : 418.91 ?

[Paragon]

I tried some old ones (388.13) and some fairly recent ones (417.17). Haven't tried 418.91, although I'll do that today.

[HaloJones]

what's so special about 372.90?

[Paragon]

For some reason, driver 372.90 and 373.06 seems to produce the highest PPD for the GTX 1060, 1070, and 1080. It's an old driver. The thought is that there are less optimizations for games and such, so it does a bit better for compute.

Threads on this (just search the threads for 372.90 and you'll see what people say about it)

viewtopic.php?f=38&t=30458
https://forums.evga.com/GeForce-Driver- ... 07955.aspx
https://forums.evga.com/GTX-970-PPD-m2687228.aspx
https://www.overclockers.com/forums/sho ... Base-Clock

I tested this out on the GTX 1070 vs. a newer driver. Results here:

https://greenfoldingathome.com/2019/01/ ... -gtx-1070/

Basically I found a 50k PPD boost for running that older driver vs. the new stuff on the 1070.

The problem is the 1070 Ti is a newer card, and that old driver doesn't support it.

[Theodore]

A lot of people don't really understand folding, and just let it run for a few minutes to an hour, and recording speed differences this way.
Quite often, they are assigned a WU that is working better with their cards, and when they update the driver, get assigned another WU that's slightly slower on their card.

Also, some of the posts you mention, were created when the older driver was the newest thing out there. So of course it would be faster than any drivers before that.
Generally speaking, diver updates fix issues, don't really lower performance.
It may be that older drivers didn't have enough control over the power consumption on those cards, and Nvidia fixed it by releasing newer drivers.
There may be a lot of reasons why, but I believe you should at least install the latest stable driver, or perhaps the newest might get you more stable performance.

If you want faster speed, there are other ways to get it (through overclocking, changing OS (Windows to Linux), adjusting cooling, making sure your fastest cards are installed in 16x, 8x, or 4x speed slots, and slower cards in 1x slots (with risers), and change settings of your PC to cause the least amount of interruptions possible (eg: like removing screensavers).

[artoar_11]

For some reason, driver 372.90 and 373.06 seems to produce the highest PPD for the GTX 1060, 1070, and 1080. It's an old driver. The thought is that there are less optimizations for games and such, so it does a bit better for compute.

Threads on this (just search the threads for 372.90 and you'll see what people say about it)

viewtopic.php?f=38&t=30458
https://forums.evga.com/GeForce-Driver- ... 07955.aspx
https://forums.evga.com/GTX-970-PPD-m2687228.aspx
https://www.overclockers.com/forums/sho ... Base-Clock

I tested this out on the GTX 1070 vs. a newer driver. Results here:

https://greenfoldingathome.com/2019/01/ ... -gtx-1070/

Basically I found a 50k PPD boost for running that older driver vs. the new stuff on the 1070.

The problem is the 1070 Ti is a newer card, and that old driver doesn't support it.

You should search by keyword "workaround" I do not know if this "workaround" was removed from the newer drivers?

viewtopic.php?f=80&t=29276&p=293524&hilit=workaround#p293524

Why should we assume that nV intends to "remove" the hot-fix from future versions.?

While the FAH developers seem to have been in close communications with nV deveopment folks, I have not found a detailed description of the problem, itself, or the nature of the changes done by either of those groups. (If somebody else has, point me to that information.)

Why should we assume that all of the people who have encountered this issue are part of the FAH community? I heard an (unconfirmed) rumor that BOINC has encountered the same problem. Even if we assume that the changes that FAH has made to Core_21 will make it possible to run pre-376.48 crivers without error AND we assume that the Core_21 .18 provides better performance than Core_21 .17 (Both nice assumptions, but as yet, unproven) nVidia still has a responsibility to the rest of the world.

I disagree a little and my understanding is:
Except folding@home the OpenCL world had no problem with 375 and later nvidia drivers. (Boinc GPUGRID had a totally different bug with driver 378.49/66 on GTX 980 Ti only)

1) Core_21 v17 had a bug which only showed up when Nvidia made driver changes in 375.
2) Nvidia made a workaround in driver 376.48 and later which also decreases PPD.
3) Core_21 v18 fixed the bug so any driver works again but 376.48 and later still decrease PPD
4) Future: Nvidia should remove the workaround to increase PPD again

=> For best PPD currently stay on nvidia driver 372.xx but any driver will work

[gordonbb]

I’m running my EVGA GTX 1070 Ti SC Gaming Black under Ubuntu 18.04.1 LTS Server with the 410.27 ppa Driver (I thought I’d upgraded all my systems to 417 but missed this one). I haven’t seen much variation in PPD under Linux running the 390 or 410 or 415 drivers on any of my cards and have been upgrading drivers for support for the newer RTX models as I acquire them.

I too am attempting to find the efficiency “sweet spot” and with that in mind am running all my slots this week at their “Stock” settings using the Default Power Limit as reported by nvidia-smi-q (180W), Auto Fan mode, Persistence Mode disabled and no Graphics (Shader) or Memory Offsets (overclocks). The FAHClient is not configured for “Advanced” units.

Since 00:00 EST 24 Feb 2019 I’m seeing:

Code: Select all

Pwr     GPU T    Fan  Util.   Yield Efficiency GPU Clk.
157.27W 64.9C 48.78% 97.83% 891kPPD 5.72kPPD/W 1.99GHz

While your method of measuring system power consumption is useful I am running dual-GPU rigs so my system consumption is 341.57W averaged over the same interval with the second GPU averaging 121.01W so my calculated “System overhead” is 63.29W.

[Theodore]

I’m running my EVGA GTX 1070 Ti SC Gaming Black under Ubuntu 18.04.1 LTS Server with the 410.27 ppa Driver (I thought I’d upgraded all my systems to 417 but missed this one). I haven’t seen much variation in PPD under Linux running the 390 or 410 or 415 drivers on any of my cards and have been upgrading drivers for support for the newer RTX models as I acquire them.

I too am attempting to find the efficiency “sweet spot” and with that in mind am running all my slots this week at their “Stock” settings using the Default Power Limit as reported by nvidia-smi-q (180W), Auto Fan mode, Persistence Mode disabled and no Graphics (Shader) or Memory Offsets (overclocks). The FAHClient is not configured for “Advanced” units.

Since 00:00 EST 24 Feb 2019 I’m seeing:

Code: Select all

Pwr     GPU T    Fan  Util.   Yield Efficiency GPU Clk.
157.27W 64.9C 48.78% 97.83% 891kPPD 5.72kPPD/W 1.99GHz

While your method of measuring system power consumption is useful I am running dual-GPU rigs so my system consumption is 341.57W averaged over the same interval with the second GPU averaging 121.01W so my calculated “System overhead” is 63.29W.

On my system, I have found that running more cards that are power limited to the lowest setting nets higher efficiency, than running fewer cards at stock settings.
Running 4 cards, at near to minimum power settings netted me about 15% more PPD than running 3 cards, fully stock, at the same system power draw.
Not sure if this holds true in every case, as there could be cases where adding another card will drop the PCIE lane speed, and might reduce efficiency instead of increasing it.

I have 2x 2060 OC cards, so it might be that running them at continuous overclock ratios, is taking something away of the folding efficiency, and perhaps the effect is less observed when running non OC'ed cards.

[QuintLeo]

The first driver version from NVidia that supported the 1070 ti at all seems to be the highest performance, in the testing I've done.
Also seems to be the most reliable so far, though I've seen a couple of folks recommend the 391.35 drivers I've not had the chance (YET) to test with those on the 1070 ti.