One risk for a novel characteristic was that AMD would depart the GPU’s double-precision mathematical capability intact. The Radeon MI50 that the Radeon VII is predicated on gives 6.7TFLOPS of double-precision mathematical number-crunching capability and THIRTEEN.4TFLOPS of unmarried-precision efficiency. The latter will come to Radeon VII — it even choices up a bit of of a lift, at 13.8TFLOPs. However AMD won’t be bringing complete FP64 efficiency to its up to date 7nm GPU.
THG experiences that the Radeon VII will be offering 862GFLOPS of efficiency or roughly 13 p.c of what the MI50 is in a position to. that is a little of an enchanting determination for AMD, for a few purposes. With the hot focus on AI and ML, either one of which are inclined to prioritize decrease kinds of mathematical precision, now not upper, one may expect to see firms artificially proscribing efficiency while the use of FP16 or INT8 information varieties in place of continuing to lock down on FP64. Limiting FP64 has always been some way for AMD and Nvidia to make sure that researchers had just right reason to shop for their top-finish cards and it’ll be attention-grabbing to peer if manufacturers start cracking down in other ways to restrict the go-pollination among structures. (Nvidia has already all started limiting combined-mode efficiency in positive instances on client hardware.)
Another a laugh be aware: To the best of my wisdom, the AMD consumer GPU that still gives the best FP64 efficiency, assuming you’ll be able to scale or run your app throughout each GPUs on the related time, is the Radeon HD 7990. That 2013-generation GPU presented complete FP64 performance throughout both cores, delivering 1.894TFLOPS of FP64 performance. Granted, it used to be throughout GPUs and used a much older model of GCN — nevertheless it also drives house the purpose that a top-finish twin-card from April 2013 still technically provides extra FP64 performance than the 509 GFLOPS of the RTX Titan X, the 420GFLOPS at the RTX 2080 Ti, or sure — the 862GFLOPS at the Radeon VII. actually, just one of the two GPUs aboard the HD 7990 would still theoretically outperform the Radeon VII.
We say theoretically, in fact, as a result of variations in GCN, OCL support, and RAM bandwidth might smartly create a scenario the place the older card got trounced, in spite of on-paper advantages. However neither AMD nor Nvidia have proven so much passion in creating an FP64 monster within the consumer area earlier the HD 7990. This used to be one house where AMD may have tried to make a differentiating play with its recent high-end GPU — we will have to see what the company has in mind a little closer to release day.