NVIDIA Rumored To Release Pascal Refresh With GDDR5X and Faster Clocks – Volta To Feature HBM2 and GDDR6 Support, 16 GB Standard Capacity
A hand full of information regarding NVIDIA's Pascal and Volta GPUs has been shared by Baidu Forums. A forum member known as USG Ishimura has detailed the plans for NVIDIA's next generation GPUs. There's no reason to doubt the credibility of this leak however, this is coming from the same guy who talked about NVIDIA Volta GPUs a few months back.
NVIDIA Pascal Refresh and Volta GPU Details Leaked - Rumored For Launch in 2017
NVIDIA launched their Pascal GPU architecture in Q2 2016 for consumers. Since launch, NVIDIA has released several cards for consumer, server and HPC segments. With the new FinFET process, NVIDIA managed to up the already great efficiency of Pascal and take it to new heights. The result is an absolutely stellar architecture that covers the entire stack from top to bottom.
But the story is not yet over with Pascal. Pascal is based on an infant node which has just seen the light of day for a few months. TSMC is always refining their process to offer better yields on Pascal GPUs, as those get better, so will the final product. NVIDIA plans to offer new graphics cards in the lineup as the process matures and these ones will be faster and better.
NVIDIA Pascal Refresh To Get Faster Clocks, GDDR5X - GTX 1060 and GTX 1070 Successors in 2017
The rumor is that NVIDIA plans to offer more chips based on the GP102 architecture. They already have a partially enabled model known as Titan X available to consumers but they would try to offer more variants in the stack. We are looking at cheaper variant of the Titan X (Pascal) and the GTX 1080 which will be faster in several ways.
This sounds like a repetition of what NVIDIA did with their GeForce 700 series which was based on refreshed Kepler cores. The GP102 core is relatively new and hence doesn't need to be updated much. GP104 on the other hand was one of the first consumer chips and that could be updated with the refined 16nm process.
The refined 16nm process is meant to offer even higher clock speeds and further stability on the Pascal GPUs. We can see the new chips clocking close to 2 GHz and maintaining those speeds under maximum load. Furthermore, GDDR5X has been a main issue for NVIDIA. Micron is under pressure to offer better yields of GDDR5X chips and that has gotten better over the while.
Micron's GDDR5X would be featured across the new Pascal lineup, except the entry level, GP107 core. GDDR5X is currently being used on the GeForce GTX 1080, GeForce GTX Titan X (Pascal) and Tesla P40 (ECC). With the coming refresh, Pascal based cards would have GDDR5X on board. Most of the high-end products launching under the refresh would be based on GP102 GPU. With GK110, NVIDIA offered several variants that included the GTX 780, GTX Titan, GTX 780 Ti and GTX Titan Black. We could see a similar trend with Pascal refresh.
NVIDIA Volta Architecture Introduction at GTC 2017 - New NVIDIA Roadmap Unveiling Expected
Moving on from Pascal, we have Volta. Volta is the last GPU of NVIDIA's current roadmap and is said to be introduced at GTC 2017. This marks a year after the introduction of Pascal to the world. NVIDIA's CEO, Jen-Hsun Huang will take the stage and introduce the latest Volta GPU architecture to the world. The chip will be aimed at HPC first, just like its Pascal based predecessor and would feature faster and higher capacity HBM2.
Jen-Hsun is also expected to showcase an updated NVIDIA GPU roadmap which will include new codenames and tech details for future chips. NVIDIA is expected to drop 10nm process and go straight for 7nm with their post-Volta GPUs, supporting HBM3 and GDDR6 memory standards.
NVIDIA Volta For Consumers Could Launch in 2018 - Rumored To Support Micron/Samsung GDDR6 Memory, 16 GB Standard Capacity
For those excited about Volta in the consumer section, you might be a bit disappointed. Volta, as mentioned, will launch first for the professional market and followed by a consumer launch. That launch is expected around 2018, not 2017. Pascal refresh will span the majority of 2017 and that architecture will be fully tuned for next generation AAA gaming titles.
Now here are some interesting details regarding the memory technologies being featured on Volta. Currently, GDDR5X offers 10 GB/s bandwidth but GDDR6 is expected to go beyond that, up to 16 GB/s. Now you might say that GDDR5X and GDDR6 are the same thing since theoretically, GDDR5X can also reach the same speeds. But, GDDR6 offers improved efficiency that in return offers lower power consumption on high capacity GPU boards.
We are talking about standard 16 GB VRAM capacities on 256-bit cards, a 2x increase over GTX 1080 and GTX 1070. The GV102 GPU is expected to feature a 384-bit bus with 24 GB of VRAM, 48 GB in updated variants. GDDR6 is slated for 2018 launch so it inclines well with consumer launch of Volta products. NVIDIA's partnership with Micron and Samsung has been going really well in the memory industry and we can expect them to be the first to utilize their new tech as soon as it's available.
NVIDIA Volta To Come in 16 GB Standard Capacity Variants
As for HBM2, Volta will be fully compliant with the technology but it will only be available for special products. Currently, the only chip to support HBM2 in the Pascal lineup is GP100, a HPC focused chip. NVIDIA could follow the same trend with Volta and feature HBM2 (32-48 GB @1~1.5 TB/s) on a HPC focused chip, say GV100.
These are some really interesting details and most of them do sound very possible. We still recommend to wait for more official information from NVIDIA themselves. NVIDIA is hosting a major keynote at CES 2017 but we don't expect to hear anything about Volta until GTC 2017 in May.
|GPU Family||AMD Vega||AMD Navi||NVIDIA Pascal||NVIDIA Volta|
|Flagship GPU||Vega 10||Navi 10||NVIDIA GP100||NVIDIA GV100|
|GPU Process||14nm FinFET||7nm FinFET||TSMC 16nm FinFET||TSMC 12nm FinFET|
|GPU Transistors||15-18 Billion||TBC||15.3 Billion||21.1 Billion|
|GPU Cores (Max)||4096 SPs||TBC||3840 CUDA Cores||5376 CUDA Cores|
|Peak FP32 Compute||13.0 TFLOPs||TBC||12.0 TFLOPs||>15.0 TFLOPs (Full Die)|
|Peak FP16 Compute||25.0 TFLOPs||TBC||24.0 TFLOPs||120 Tensor TFLOPs|
|VRAM||16 GB HBM2||TBC||16 GB HBM2||16 GB HBM2|
|Memory (Consumer Cards)||HBM2||HBM3||GDDR5X||GDDR6|
|Memory (Dual-Chip Professional/ HPC)||HBM2||HBM3||HBM2||HBM2|
|HBM2 Bandwidth||484 GB/s (Frontier Edition)||>1 TB/s?||732 GB/s (Peak)||900 GB/s|
|Graphics Architecture||Next Compute Unit (Vega)||Next Compute Unit (Navi)||5th Gen Pascal CUDA||6th Gen Volta CUDA|
|Successor of (GPU)||Radeon RX 500 Series||Radeon RX 600 Series||GM200 (Maxwell)||GP100 (Pascal)|