It’s been silent with new graphics card releases from Nvidia. We’ve mentioned it a couple of times already, we really do not expect Volta with HBM2 to reach consumer grade graphics anytime soon. The chatter is about to get big as the first indications are here that Nvidia will be launching a GA104, a GPU called Ampere, on April 12th.
You are going to see a multitude of disclaimers in this news-item, however, the rumor that will flood the web after today is derived and traced back towards 3dcenter, and is nothing other than speculation based on a few observations. Nvidia would be readying a GPU series called Ampere, and not Volta. The name “Ampere” has been mentioned in the past a couple of times already, in fact, we wrote a few items about that GPU name popping up, constantly. It’s now said and written that Nvidia halted production of the GP102 (e.g. 1080 Ti / Titan X) and likely GP104 GPU (e.g. 1070/1080), all originally released back in 2016. So, the question then arises, what is happening with Volta GPUs? Where are they?
Why not Volta?
Currently, Volta is based on HBM2 memory which is not available in large enough mass-volume quantities, and it is expensive to purchase and implement as well. It also is an architecture with Tensor cores, that’s expensive and not likely benefiting gaming. So why produce high-end graphics cards, or better yet, GPUs with Tensor cores? Then there are the recent announcements on GDDR6 graphics memory. Earlier on we reported about GDDR6 closing in, really fast and actually already announced as available. Currently, it is already possible to fab graphics cards with the blazingly new fast GDDR6 memory. So my thesis here is that Volta will remain on track for the HPC / enterprise side of things including data-centers with its Tensor cores, and Ampere GPU series would be used for the consumer slash gaming parts. And hey, think about it, the specs of the Titan V and the Nvidia Volta architecture just don’t that much sense for gaming with its 640 tensor cores that games are never going to use.
Ampere; a new architecture or Pascal refresh?
In short, the “Ampere” GPUs series could be better suited for consumers, and not the “deep learning” market, and we think they are based on GDDR6 graphics memory. Ampere as an architecture never made it into the long-term roadmaps from Nvidia, never ever … hence logic dictates here that Ampere would be based on refresh Pascal architecture, perhaps fabbed on a smaller node and optimized fabrication process. Basically, Nvidia would release the GA104 and then we can speculate onwards, in a speculated chart, based on just assumptions(!), that would look something like this (courtesy of 3dcenter):
GPU | market segment | Release | Names | Performance aim |
---|---|---|---|---|
GA104 | High end | April 2018 | GeForce GTX 2070 & 2080 | 2070 ~ GeForce GTX 1080 Ti |
GA106 | Mid range | Late summer / fall 2018 | GeForce GTX 2060 | 2060 ~ GeForce GTX 1080 |
GA102 | Enthusiast | End of 2018 to spring 2019 | GeForce GTX 2080 Ti | 2080 Ti ~ GeForce GTX 1080 Ti + 70-80% |
GA107 | Mainstream | Spring / Summer 2019 | GeForce GTX 2050 & 2050 Ti | 2050 Ti ~ GeForce GTX 1060 |
GA108 | Entry level | Spring / Summer 2019 | GeForce GT 2030 | 2030 ~ GeForce GTX 1050 |
Ampere in April 2018?
If it is Ampere that will be released, the new first released GPU would thus be the GA104, and albeit retail naming remains speculative, these would/could/should first be released named as GeForce GTX 2070 and 2080, and the rumor is indicating and pointing towards April 2018, the 12th for new high-end consumer graphics card announcements. That date seems … so incredibly short, so here, I have to raise concerns. I’ll add something to that though, NVIDIA could even announce (not launch) GeForce 2000 at GTC, which takes place at the end of March. Let’s call this what it is though, a spicy rumor and nothing more. But good gosh, it’s really time for something new eh?