AMD aren’t the only ones who’re planning on releasing a new architecture – Nvidia are working on Pascal PK100, the follow up to their rather successful Maxwell series (the GTX 9xx series). During their recent earnings call, the companies CEO, Jen-Hsun Huang took the opportunity to tease a few details of the GPU’s – most of which is already known, but he did reveal that they would be “coming soon” and the performance gains over the current generation would be substantial.
Nvidia boast that Pascal will provide 4x the mixed precision performance and 2x the performance per watt over their current Maxwell architecture; and are going to be jumping ship from GDDR5 to HBM2 (unlike AMD, who will have Fiji running on HBM1, then it’s successor on HBM2). This will allow them to supposedly provide 3x the bandwidth of the current Maxwell parts, and also 2.7x the amount of VRAM on board.
This equates to roughly 1TB/s of memory bandwidth (to put this into perspective, most modern high end GPU’s, such as the R9 200 series or the GTX 900 series put out around 300GB/s) and the GPU can support up to 32GB of VRAM… not too shabby, right?
“I cannot wait to tell you about the products that we have in the pipeline,” said Jen-Hsun Huang, Nvidia’s CEO. “There are more engineers at Nvidia building the future of GPUs than just about anywhere else in the world. We are singularly focused on visual computing, as you guys know.”
He continued:“We have found over the years to be able to focus on just one thing, which is visual computing, and be able to leverage that one thing across PC, cloud, and mobile, and be able to address four very large markets with that one thing: gaming, enterprise, cloud, and automotive,” said Mr. Huang. “We can do this one thing and now be able to enjoy all and deliver the capabilities to the market in all three major computing platforms, and gain four vertical markets that are quite frankly very exciting.” …“I cannot wait to tell you all about it, you are just going to have to wait just a little longer.”
So, what else do we know about Pascal? Well, not as much as we’d like – we do know of two Pascal parts in development, the first being PK104 – which is the follow up to GM204 (the Maxwell flagship). The second GPU will be PK100 – but less details are known about this GPU. Supposedly they will be shifting to either 14 or 16nm (from their current 28nm process) and will ‘mostly’ be working with TSMC (despite some rumors they would be working with Samsung).
During the earnings conference Jen-Hsun said: “We are constantly evaluating foundry suppliers, ….[but] we largely purchase from TSMC, the vast majority of our wafers we buy from TSMC. We are in 20nm, we are expecting to ramp 16nm. We are deeply engaged with TSMC for many, many nodes to come, including 10nm.” CEO, Nvidia
“There are just so many ways for us to deliver energy efficiency and performance, I would not get too obsessed about the process technology all by itself.”….“But we are always looking at new foundry suppliers, and competition keeps everybody sharp……But for all intents and purposes, TSMC is our primary partner.”
It’s far too early to know how this will stack up against Fiji, or indeed – even IF it will be facing off against Fiji, as it’s possible AMD will have pushing out Arctic Islands (the 400 series) around the same time. What we do know is that both companies are addressing the problems of both performance and Performance Per Watt (PPW).
There are certainly PCI-E bandwidth constraints (which new standards will eventually help resolve) but large amounts of Frame Buffer memory help alleviate data swapping from system to VRAM (over the PCIE bus) and, of course, with ExecuteIndirect and other techniques in DX12 offloading some draw instructions (or compute instructions) from the CPU to the GPU, further bandwidth will be saved.
If you’re going to take anything from this news and article, take it as a great sign for the industry. While I’ve been a little critical of Maxwell (not because it’s bad, but just because in terms of raw performance, it wasn’t super exciting) and indeed even AMD for not releasing anything that exciting over the past year, things are changing – quickly. Hopefully we’ll all be in a very different situation in a years time – and with any luck, we’ll also have Zen to kick some competition in to the CPU industry too.