NVIDIA’s next-generation GPU Ampere is probably the best in recent years, so far the hammer has been so bad, even the release time is uncertain, even at the March GTC conference, even if officially announced, may be just the core of the GA100 for the data center.
According to reports, the GA100 core area of up to 826mm2, it is known that up to 7680 CUDA core, memory space width 320bit, the scale is much higher than the current GT102 core.
The ampere GPU for the data center should be a 7nm process, because to ensure performance and energy efficiency, but the game-oriented GPU is not necessarily, Videocardz website received anonymous reports that a certain ampere game GPU will use the 8nm LPP process, has been ready to stream.
Although not explicitly stated, but the 8nm LPP process now only Samsung has, this should be Samsung contract.
Is this going to pop out of an 8nm ampere GPU, which seems a little unexpected? It makes sense to think about it, first of all, NVIDIA began to look for Samsung from The Pascal generation, 14nm contracted the GTX 1050 series GPU chips.
NVIDIA has previously stated that the new generation GPU will have Samsung and TSMC two contract workers, TSMC should still account for the highest proportion, but Samsung is also indispensable, although the Turing generation does not have their share.
Second, NVIDIA released DRIVE AGX Orin at the GTC Conference in China late last year, the next-generation self-driving platform, featuring the new NVIDIA GPU and 12-core ARM CPU, with 200TFLOPS performing seven times as much as Xavier did.
NVIDIA did not announce the specific specifications of the GPU for DRIVE AGX Orin, but said it was a new generation GPU, not surprisingly the Ampere GPU.
Now, combined with the revelations, NVIDIA will use Samsung’s 8nm LPP process in the less-than-state-of-the-art Ampere GPU, and it looks like the mobile, desktop, and autonomous driving products will have Samsung’s 8nm LPP process.
Samsung’s 8nm LPP sounds 1nm worse than 7nm, but in fact it’s two generations of craft, 7nm is a brand new node, 8nm LPP is a second-degree modification of the 10nm process, although compared to 7nm is not so fragrant, but the benefits of the mature process is to save money.
Considering the NVIDIA over the past few years on the 7nm process cost scare, now look at the Ampere GPU family multi-process route should be nailed on the board.