Titan V not suitable for scientific simulations

[spartaman64]

Quote

One engineer told The Register that when he tried to run identical simulations of an interaction between a protein and enzyme on Nvidia’s Titan V cards, the results varied. After repeated tests on four of the top-of-the-line GPUs, he found two gave numerical errors about 10 per cent of the time. These tests should produce the same output values each time again and again. On previous generations of Nvidia hardware, that generally was the case. On the Titan V, not so, we're told.

Quote

Nvidia said the best solution to these problems is to avoid using Titan V altogether until a software patch has been released to address the mathematical oddities.

There are issues with the Titan V card that causes it to make errors doing calculations. This doesn't affect gaming but it gives inconsistent results to researchers and engineers so until Nvidia fixes the issue the gpu shouldn't be used for those purposes. Not many gamers can afford or think it's worth it to spend 3000 on the Titan V so Nvidia should remedy this quickly or else they will probably lose a big part of the Titan v's market. 

 

Link: https://www.theregister.co.uk/2018/03/21/nvidia_titan_v_reproducibility/

[Guest The Viking]

then people blame AMD for selling bad cards... look at this, a 3000$ card that is pretty much the equivalent to... oh, right, "crap". 

[porina]

Buy cheap, this is what you get. Yes, I'm using cheap relatively here... the real pro cards get ECC don't they? The Titan V, even if it does have most of the features, is kinda in an odd space between gaming and the true pro cards. The price was cut, this is not unexpected. From what I've seen in general, ram clock is generally close to the limit on gaming cards, where infrequent errors generally go unnoticed. When using consumer GPUs for compute, I've had to un-overclock them before to get stable results, so this really isn't anything new.

[mr moose]

 

I'd like to know just how much Nvidia is actually making out of the titan line up,  I haven't really seen anyone recommend them for anything for a rather long time.

[leadeater]

Um, LOL?

 

This only for Titan V or for all of Volta?

[GoldenLag]

Does the quadro variant of the card have the same issues? Cause as far as i know its the same GPU die but with different HBM stacks. I believe the Quadro variant is the V100. Would be intetesting to see if they have identical issues, because then it would be a big issue. 

[tikker]

Well that's unfortunate...

[Brooksie359]

16 minutes ago, GoldenLag said:

Does the quadro variant of the card have the same issues? Cause as far as i know its the same GPU die but with different HBM stacks. I believe the Quadro variant is the V100. Would be intetesting to see if they have identical issues, because then it would be a big issue. 

I thought the v100 was a tesla not Quadro. I do believe they are different. 

[AlTech]

1 hour ago, leadeater said:

Um, LOL?

 

This only for Titan V or for all of Volta?

Sounds like only for Titan V. Even though the Tesla V100 is based off of the same chip.

[Coaxialgamer]

1 hour ago, leadeater said:

Um, LOL?

 

This only for Titan V or for all of Volta?

that's the real question here . And if it is for all of volta , then it requires a chip respin. Micro-code can't fix logic problems with any sort of reasonable performance

[Misanthrope]

I've been playing GTA a bit and I think that given the last remaining use for the Titan V the V should stand for 'Vapid'

[2FA]

I honestly forgot the Titan V existed.

 

I wonder if the issue was with specific software or just happened in general.

[AnonymousGuy]

2 hours ago, mr moose said:

 

I'd like to know just how much Nvidia is actually making out of the titan line up,  I haven't really seen anyone recommend them for anything for a rather long time.

There is a delay between the  Titan  and the Ti version of the Titan, so if you want absolute max gaming performance without waiting you buy the Titan.

[DoctorWho1975]

*waiting for a Viceroy hit piece as to why the NVIDIA stock should be zero*

[Sauron]

This is why quadros and teslas come with ECC vram and USED to cost 3 times as much, but the titan brand is just nvidia trying to market stuff to both gamers and professionals and making a mess

[Coaxialgamer]

4 minutes ago, DoctorWho1975 said:

*waiting for a Viceroy hit piece as to why the NVIDIA stock should be zero*

bankruptcy intensifies 

[Notional]

In other words, these cards are great for pubg? 

 

It's always odd when hardware encounters such bizarre behaviour. I wonder if the software fix will decrease performance? If not, it's "only" an issue for everyone who used these GPU's for any type of research. If perf. will decrease, then it sucks for everyone owning them.

[Valentyn]

So what you're saying is....it's not worth buying to play World of Warcraft and their upcoming expansion that uses DX12?

I don't want to lose 10% of my DPS due to calculation errors!

[SpaceGhostC2C]

3 hours ago, mr moose said:

 

I'd like to know just how much Nvidia is actually making out of the titan line up,  I haven't really seen anyone recommend them for anything for a rather long time.

 

1 hour ago, Sauron said:

This is why quadros and teslas come with ECC vram and USED to cost 3 times as much, but the titan brand is just nvidia trying to market stuff to both gamers and professionals and making a mess

 

I don't think it's marketed to anyone anymore, but it rather became a marketing stunt. It will top some benchmarks most outlets can run, and it will be in every flagship build not meant for daily used and/or cost-effectiveness (like, if you make a PC for Terry Crews, no one, including yourself, cares about how much it costed). But when it comes to actually build a PC, it will be too expensive for gaming, and too gaming for workstations.

I think it used to be different (more towards the pro side) with the first Titans, but now it seems to be mostly a brand placement device.

[AresKrieger]

They probably are just going to tighten the memory timings or something along those lines if they can actually fix it, ECC exists for a reason granted the Titan V would still be useful for machine learning ventures but without ECC it will not be suitable for things like molecular physics, but the only reason people thought it would be was poor marketing decisions by nivida (though the marketing was likely only to produce hype for Volta variant gaming cards while also not needing to stock any currently)

[leadeater]

1 hour ago, Sauron said:

This is why quadros and teslas come with ECC vram and USED to cost 3 times as much, but the titan brand is just nvidia trying to market stuff to both gamers and professionals and making a mess

Doesn't sound like memory errors though, nothing would work if it was that. Sounds firmware ish.

[GilmourD]

I've always been on Team Green, mostly due to performance and driver suckiness issues with ATI/AMD... But... I dunno... nVidia's kinda pushing me away these days.

 

[descendency]

So... it's a $3000 gaming card without RGB?

[rawrdaysgoby]

Hey scientist I like to take that card off your hand for 2k it is junk to you now!!!!

[Trixanity]

6 hours ago, GilmourD said:

I've always been on Team Green, mostly due to performance and driver suckiness issues with ATI/AMD... But... I dunno... nVidia's kinda pushing me away these days.

 

7 hours ago, leadeater said:

Doesn't sound like memory errors though, nothing would work if it was that. Sounds firmware ish.

If it was due to lack of ECC memory, I'm sure Nvidia could spin it in a marketing angle: so fast the memory can't keep up.