9401 fails on GM107 but not GK106 {Hopefully fixed}

[7im]

Let's not forget this is Maxwell hardware. There is no guarantee the current fahcores will run on a new GPU architecture.

[7im]

Okay, wasn't KB2909921 (IE 10 Cumulative update). Just installed that one and rebooted. Still folds.

You sure they didn't add a driver when you weren't looking?

I wouldn't put trying past them, but I didn't see anything in the list of offerings which looked like a GPU driver update and I didn't do anything manually to them. I assume then by one mechanism or another that batch of updates broke something and removing them fixed it. Considering how much software it takes to run Windows alone never mind accomplish something I'm surprised this is unusual.

Jim

ETA - The three other Win7 boxes I have exhibited no issues in the same timeframe, and I never ran any of those updates on them.

Somewhat bad news Jim.

Over the last day, I installed each of the pending Windows update files on my Windows 7 32 bit system with a GTX 760. I rebooted after each update, and verified that fahcore_17 was still folding okay.

Now I may not have all the same hardware drivers, or software apps, so not quite apples to apples, but maybe this is something you could try yourself. Nail down which one did you in.

[Rel25917]

i lowered the clock speed to 990mhz which limits the boost to the card near it's default clock speed. it still fails on a 8900.

What about the memory speeds? Core17 is much more sensitive to memory speeds than the core clock speed.

[soa_rru]

i lowered the clock speed to 990mhz which limits the boost to the card near it's default clock speed. it still fails on a 8900.

What about the memory speeds? Core17 is much more sensitive to memory speeds than the core clock speed.

My 750Ti doesn't even start to fold, it just sits idle for several minutes before the core decides to dump the wu so it's nothing to do with clock rates.

But like 7im said

Let's not forget this is Maxwell hardware. There is no guarantee the current fahcores will run on a new GPU architecture.

[heikosch]

But uddarts wrote viewtopic.php?f=80&t=25850#p259669 that his 750ti works.

Heiko

[soa_rru]

But uddarts wrote viewtopic.php?f=80&t=25850#p259669 that his 750ti works.

Heiko

That post mentions he's added a 750Ti to the mix but the post was mainly telling us what ppd his 560Ti got using the latest drivers.

Re: 334.89 WHQL drivers

Postby uddarts » Tue Feb 25, 2014 6:07 am
i added a 750ti today so am using the 334.89 drivers.

9401: 560ti - 16' 44" - 33,750/ppd

ud

Both his & my 750Ti will fold core15 quite happily but core17 is a no no

[rwh202]

I find it surprising that Stanford (or nVidia) haven't tested this already.
The Maxwell release isn't exactly a shock and I would have thought at the bare minimum they would have had access to cards when reviewers did and show stoppers found and communicated.
We've already seen a post from Prof. Pande saying they have high hopes for Maxwell, so it was on their roadmap, but why are users being expected to find and test these issues at their own expense?

[bruce]

I find it surprising that Stanford (or nVidia) haven't tested this already.

I don't. Stanford is NOT a hardware reviewer, nor do they design drivers. Their expertise is producing scientific code that works on "standard" hardware.

Some of the hardware reviewers use FAH in their test suite and you can read what they say, just like the rest of us.

Depending on the changes involved in any new generation of hardware, driver changes are often required and any bugs discovered must go through the normal driver release cycle before you or I get to see them. If nVidia included Core17 in their driver pre-certification process, they know about the bugs found when tested at stock speeds, not necessarily anything that might be encountered with elevated clock rates. Stanford does not recommend overclocking, nor do they prohibit it. Once you decide to overclock, you're exploring untested conditions which are entirely your own responsibility.

Here on the support forum, we strongly advise you to reset all clock rates to the factory recommendations and test that case BEFORE complaining. From time to time, FAH has had a history of (unintentionally) providing the world with a more rigorous stability test than traditional benchmarks. Perhaps this has happened with your GPU, perhaps not. Please let us know what you find.

[uddarts]

But uddarts wrote viewtopic.php?f=80&t=25850#p259669 that his 750ti works.

Heiko

i glanced through that thread yesterday and at first read wondered too, and i wrote it.



ud

[PantherX]

...The Maxwell release isn't exactly a shock and I would have thought at the bare minimum they would have had access to cards when reviewers did and show stoppers found and communicated...

In addition to what bruce stated, there can be times when the press samples are limited in number so each reviewer has a set amount of time before they have to pass it on. Other times, the press gets engineering samples.

I am unsure of Nvidia's Maxwell situation but when it comes to code development, you don't really want to spend hours debugging on an engineering sample since it may (or may not) vary slightly with the retail version which may (or may not) break your code. Moreover, code development takes a significantly long time (specially if you want to optimize it) as opposed to running a standard benchmarks (it might be time consuming but you aren't debugging anything).

[bfromcolo]

So all that being said, is anything being done to address this or are we just waiting for a new driver from NVIDIA that addresses whatever the problem is? I was interested in buying one, but until it folds and I have some idea of likely performance I will just wait.

I would be using this in Ubuntu, so has anyone reported any success with Linux?

[PantherX]

Please note that the development team has been informed of this.

Moreover, Dr. Vijay has stated that they plan to optimize OpenMM for Maxwell GPUs (viewtopic.php?p=259238#p259238).

[soa_rru]

Just thought i'd try the new 335.23's but still the same result

[Freightanimal]

I have the same issue with my new GTX 750 ti. So disappointed!! I was hoping to do some decent folding. I too tried again after the driver release without any success. I would be curious to know if this is a problem on other OS's. I might try the card in an i3 box I have. I could install XP on a spare hdd and give it a go. It looks like it's a different driver version. No time in the near term though.

[7im]

Let's not forget this is Maxwell hardware. There is no guarantee the current fahcores will run on a new GPU architecture.

As stated, we likely need to wait for a new version of fahcore_17.