To nip [something] in the bud: http://www.phrases.org.uk/meanings/256600.html
And yeah, that could very well be the goal.
The NVIDIA tweet, the only actual source WCCFtech has, says nothing that would even remotely hint any new graphics cards
The tweet has link to http://www.nvidia.com.tw/content/apac/event/en/computex2017/ which lists NVIDIA AI Forum and NVIDIA Deep Learning Institute Workshop as the only two events NVIDIA is having there, Huangs keynote is part of the AI Forum
The tweet has link to http://www.nvidia.com.tw/content/apac/event/en/computex2017/ which lists NVIDIA AI Forum and NVIDIA Deep Learning Institute Workshop as the only two events NVIDIA is having there, Huangs keynote is part of the AI Forum
Yeah, wouldn't expect Volta here.
I think the rumor about possibly a new card first came from the May 9 Q1 earnings call Q&A, though it could be for any of Nvidia's markets.
http://wccftech.com/nvidia-graphics-announcement-gtc-2017/
Well, they dont have to forcibly launch a new graphics card, just present it with later availability, We know that AMD will have Vega there, so it is not first time, that Nvidia come at the same time, day with something just for cover the competition presentation.
At the same time, maybe they could launch a new Titan at 2K $ price tages with low availability at least for the 4 next months. With no real review, just peoples know the card exist.
But this is a little bit too early for that in the year. I will be surprised to see a Volta gaming part so early . ( but who know )
At the same time, maybe they could launch a new Titan at 2K $ price tages with low availability at least for the 4 next months. With no real review, just peoples know the card exist.
But this is a little bit too early for that in the year. I will be surprised to see a Volta gaming part so early . ( but who know )
Huangs keynote is actually today, while AMD shows their stuff tomorrow.
But that statement refers to GV100/V100 and nothing else, it even says "product" instead of plural
NVIDIA’S GPU 2017: A.I. – ARTIFICIAL INTELLIGENCE IS HERE AND THE ROBOTS ARE MAKING GOALS…
https://mlsentertainment.wordpress....ence-is-here-and-the-robots-are-making-goals/
https://mlsentertainment.wordpress....ence-is-here-and-the-robots-are-making-goals/
Infinisearch
Veteran
With Vega releasing so late do you think Nvidia will accelerate or slowdown the release of Volta? I wonder if NV will try to go for a killing blow by releasing early. Also any rumors on whether consumer Volta will retain FP16 support?
Nvidia will do whatever they believe will make them the most money. But I don't think any attempt at a "killing blow" makes much sense. Nvidia has already "won", and the majority of people still buying AMD probably won't be influenced by whatever Nvidia does. Of course this doesn't mean Vega/AMD can't be competitive, but competitive doesn't shift marketshare overnight. Vega would have to be dramatically better than whatever Nvidia is offering to achieve significant short-term change, and that is just not going to happen....
As for FP16, who knows? If Nvidia genuinely cared about creating a cohesive development environment, of course it would (and the question wouldn't even exist in the first place). But these days, they never seem to miss an opportunity to go full scumbag....
With the introduction of 'tensor cores' I'm not sure where that leaves double rate FP16 and DP4A on the GP100 and GP102/4/6 chips. As far as segmentation goes they could opt to make the tensor cores exclusive to GV100, whilst introducing 2xFP16 to the 2/4/6 chips. But what about DP4A? Even when compared to the 120 'tensor flops' DP4A on GV100s shader array would bring roughly 60 uh, tera 8bit ops (there must be a better way to say this but I ahve forgotten it) which is still significant as for as inference goes. One could also envision a future tensor core oriented chip with fewer traditional SMs and a focus on those matrix operations
The real question is can they do it .. Nvidia have market and production timeline ... and Volta dont change from what we have seen since many year in their timeline.. a big show, showing the next architecture, professional architecture shown.. Then middle grade GPU's ( 980-1080 etc ), then Titan etc ..
Maybe they could release Titan further .. but well. i dont think any consumer part is set to use GV100.,, surely something more like GP102 ).. But if they plan to use GDDR6 for it, can they allready have them ?
Last edited:
Since the "killing blow" would have to be dealt by gaming oriented chips, another question worth pondering is: How much do Volta's changes net for Geforce-/Radeon-type of applications (i.e. games). From what I've seen with GV100, there is no real improvement in transistor density nor in clock speeds from the "new"(ly labelled) process tech. Tensor Cores are not made for gaming. HBM2 won't be in line with Nvidias production strategy of cheap(er) components, IMO. Smaller SM blocks... well, we haven't seen the GP100-style SMs in GP10x-chips either. The unified Shared-Memory/L1-cache maybe, but AFA sebbi characterized it, this is more an enabler than a performance thing. If that's not total marketing BS and only true for VERY select tasks, then maybe the higher energy efficiency of the SMs might help a Volta-based gaming card. Will 2×FP16-rate be there and enabled? If so, when will games take advantage of it?
You forgot to mention that efficiency went to the roof. in fact, Nvidia claims that Volta SM is 50% more efficient than Pascal one, a clear indication that Volta SM is not a tweaked update but a totally new design. Only with this advantage, Nvidia will keep a large lead in power efficiency vs Vega, thus in performance too, even if it implies a bigger die
Likely in specific types of workloads though.
Maybe, but it's not using Tensor cores for sure, otherwise they will claim more than 5 times efficiency gain on Volta!
Similar threads
