Report: Nvidia’s GeForce Volta GPUs will Not use HBM2

A new report suggests that Nvidia is not going to utilize HBM2 to power its consumer-level GeForce Volta cards. Instead, the next-gen GeForce will use GDDR5X(6) memory as the best solution around.

No HBM2 for GeForce Volta, It’s Too Early & Expensive

Nvidia unveiled the next-generation Volta architecture at their annual GPU Technology Conference last month. Like the Pascal GPU architecture before it, Volta’s reveal didn’t include GeForce graphics cards but a monstrous Tesla V100 GPU designed for supercomputers.

READ ALSO: Nvidia accused of Rigging The SDR vs HDR Monitor Comparison at Computex

Built using the industry leading TSMC 12nm FinFET process, the Tesla V100 packs 21 billion transistors and 5,120 CUDA cores running at 1,455MHz. It will be interesting to see how many CUDA cores fit into consumer GeForce Volta cards, especially since the Tesla V100 is said to deliver 1.5 times the compute performance of its Pascal predecessor.

While we don’t know about the CUDA core count of Volta yet, Fudzilla claims to have information on the type of memory the GPU will be making use of — or to be precise, the type of memory it won’t be making use of.

According to the site’s well-informed sources, the next GeForce will not use the second generation High Bandwidth Memory (HBM2). The reason behind the move is cited as being too early for that, and HBM2 is still quite expensive – perhaps prohibitively so for consumer-level parts.

That should mean the consumer GeForce Volta will use GDDR5X, or even GDDR6 memory. But since GDDR6 will not arrive until next year at the earliest, the next GeForce cards are most likely to feature R5X given they launch later this year or even in the early part of the next year.

Expect High-end Volta GPUs to feature GDDR6 memory

Nvidia may, however, use GDDR6 memory on its future Volta based cards, say a GTX 2080 Ti or whatever it ends up being called. SK Hynix recently confirmed it’s prepping cutting-edge GDDR6 RAM to ship in high volume on an all new graphics processor in early 2018.

The memory supplier didn’t say whether the graphics product is from AMD or Nvidia. But, we know it would possibly be from Team Green because AMD’s next-gen enthusiast Radeon GPU, called Vega, will ship with HBM2 technology.

READ MORE: AMD’s RX Vega Price Rumor: HBM2 4GB memory stack costs $80

The consumer RX Vega chip will be equipped with two stacks of HBM2, incorporating up to 8GB of memory. While HBM2 upstages DDR in bandwidth and power, the technology is super expensive to produce. There are a lot of design complexities associated with its production.

There have been rumors about the low production of HBM2 that could limit the supply of Radeon RX Vega. Reportedly, AMD would release only 16,000 units of RX Vega for consumers in the first few months of its launch.

That said, the above information on Nvidia Volta’s memory comes from only one source. There is nothing official yet, so better take it with the proverbial pinch of salt.