NVIDIA GeForce GTX 1080 Ti Shows Off Strong OpenCL Performance

[coder]

Manufacturer stated numbers are always at least a tad unreliable and you're generally never going to reach those figures except in benchmarks that really don't represent performance in real applications.

To paraphrase what I said (and not by much): real benchmarks are better, but the specs tell you the upper limit of what you can hope to achieve.

In case you didn't notice, I said "I'm a fan of including a couple FP64 benchmarks", a few posts above that. So, we're not arguing over FP64 benchmarks. I was just trying to inform you of what's known about these GPUs, but if you're intent on having the last word, I'll leave you to it.

Hopefully, at least you'll stop claiming that Titans (and Quadros, writ large) have non-crippled FP64 performance. It's simply not true (and hasn't been, for a while).

[L_A_G]

Hopefully, at least you'll stop claiming that Titans (and Quadros, writ large) have non-crippled FP64 performance. It's simply not true (and hasn't been, for a while).

I wouldn't say that you can quite yet use the phrasing "for a while" when Nvidia has only released one Titan card with crippled FP64 performance.

[coder]

I wouldn't say that you can quite yet use the phrasing "for a while" when Nvidia has only released one Titan card with crippled FP64 performance.

Titan X is the name they used for both the GM200 (Maxwell 2nd gen)-based card and the GP102 (Pascal). Don't complain to me about their poor choice of names, though they already took Titan Z and Titan Y sounds too much like an uncomfortable question (Titan, why?).

So, that means 3 years have elapsed since the triumvirate of Kepler-based, FP64-enabled Titans got released. That's 2-3 generations of GPU architectures (GK1xx, then GM1xx, GM2xx, and GP1xx) ago, depending on how you count. Yes, in GPU terms, it was a while. I can't know what you deem "a while", but now you have the specifics.

[L_A_G]

..

As I said, it's still just one card in the Titan line that's come with hobbled FP64 performance. You usually wait for at least two versions of something before you declare the change to be permanent.

[coder]

As I said, it's still just one card in the Titan line that's come with hobbled FP64 performance. You usually wait for at least two versions of something before you declare the change to be permanent.

Wow, it's like you didn't read what I wrote. Or is this some kind of alternative fact?

There are two products called Titan X, each based on different GPU architectures. That brings the total of GPUs in Titan-branded products to three: GK110, GM200, and GP102 (in that order). Only the first had decent FP64 performance. So, you have one generation that had it, followed by two that don't. That really calls out the first as being the exception.

[coder]

Maybe this will help:

GPU     Introduced      Products
------------------------------------
GK110   Feb 2013        Titan, Titan Black, Titan Z, GTX 780, GTX 780 Ti
GM200   Mar 2015        Titan X (Maxwell), GTX 980 Ti
GP102   Aug 2016        Titan X (Pascal), GTX 1080 Ti

[L_A_G]

Wow, it's like you didn't read what I wrote. Or is this some kind of alternative fact?

There are two products called Titan X, each based on different GPU architectures. That brings the total of GPUs in Titan-branded products to three: GK110, GM200, and GP102 (in that order). Only the first had decent FP64 performance. So, you have one generation that had it, followed by two that don't. That really calls out the first as being the exception.

Well someone sure dispenses flavor-of-the-month insults willy-nilly...

Seems like I didn't read your post properly, but my original point you brought up about the FP64 performance stands. It would be nice to get some kind of actual performance FP64 figures compared to non-hobbled cards as theoretical figures given out by vendors really don't tell anything close to the whole picture. You going on about them already having a previous card sharing the same name as an un-hobbled card doesn't really take away from that, specially when people don't upgrade their hardware every single time a new generation of hardware comes out.

[coder]

Well someone sure dispenses flavor-of-the-month insults willy-nilly...

Seems like I didn't read your post properly,

How can you misread that entire post? You directly contradicted the whole point of it, which I had also clearly stated in an earlier post in this thread. In light of such actions, it's certainly appropriate to question whether the counter-party is acting in good faith. At least it seems to have gotten your attention.

but my original point you brought up about the FP64 performance stands. It would be nice to get some kind of actual performance FP64 figures

For the third time, we're not arguing about that. I'll just go ahead and quote myself, so you can not read it, again.

BTW, I'm a fan of including a couple FP64 benchmarks. Maybe it will nudge Nvidia in the direction of including more FP64 performance, in future high-end GPUs. I doubt it, but we can hope. Same goes for FP16, actually.

It will also highlight any differences between Nvidia and AMD GPUs, in this regard.

[L_A_G]

....

So all you had to add to the conversation was to point out that I missed Nvidia hobbling FP64 in the second card sold under the Titan X moniker?

[coder]

So all you had to add to the conversation was to point out that I missed Nvidia hobbling FP64 in the second card sold under the Titan X moniker?

No. First of all, you made broad generalizations about Titan FP64 performance from which even this generation wasn't excluded.

the reason why people generally buy Titan and Quadro series cards over regular old GeForce cards is that Nvidia hasn't hobbled their FP64 performance (FP64 performance doesn't matter all that much for most people who buy GeForce GPUs as they use them for games which use FP32 and FP16 and not FP64).

Here, you also inaccurately state that games depend on strong FP16 performance, which I pointed out.

I then provided this helpful link to more details, and a deeper dive on the 1080 Ti and GP102 GPU, including very topical analysis of its compute performance potential relative to the Titan XP.

http://www.anandtech.com/show/11172/...-next-week-699

Then, I voiced my own support for FP64 benchmarks. And offered this helpful link to far more information about Nvidia GPUs:

https://en.wikipedia.org/wiki/List_o...orce_10_series

Next, I offered supportive comments to adakite, as well as advice (and links) on cost-effective avenues for adding FP64 compute performance to one's system (which that poster claimed they needed).

It wasn't until this point (first post on page 3), that I finally got dragged down into the impasse of your stubborn ignorance. You got me repeating myself, in a fashion that'd make any good troll proud. I did manage to provide a helpful table of Titan GPUs and related products. And that brings us up to the latest gem of your contributions to this thread.

I actually don't mind the retrospective, as it makes me feel like my time in this thread mightn't have been quite such a waste of time as I thought.

I like how instead of talking about any actual issue or technical matter, I'm somehow being indicted. I hope I've played along to your satisfaction.