This site may earn affiliate commissions from the links on this folio. Terms of employ.

Vega-Feature

At that place'due south a new set of rumors cropping upward virtually AMD'south Vega. Final month, we covered an AMD press slide that stated the GPU would only make it sometime in the commencement half of 2022. Now, reports bespeak that the fleck could drib in the first quarter — still quite late compared with Nvidia'due south Pascal, which will have been in market place for nearly a year at that point, merely better than the late June timeframe typically hit past "H1" launch windows.

The upcoming Vega 10 scrap will feature 24 TFLOPS of 16-bit precision performance, 16GB of HBM2, and 512GB/due south of retention bandwidth, with 64 compute units in total, according to VideoCardz. If Vega keeps GCN'due south 64 compute cores per compute unit, that puts a total-size Vega cadre at 4,096 cores in total. The 16GB of HBM2 is expected based on the specs we've seen for that standard, but the RAM bandwidth is significantly lower than what we've expected to date. HBM2 is designed to back up up to 256GB/s of bandwidth per stack, and first-generation HBM2 solutions are expected to ship with 4GB stacks. That works out to up to 1TB/due south of memory bandwidth for a full implementation — and the first-generation HBM implementations that AMD shipped were full implementations.

sk_hynix_hbm2_implementations

If these stats are authentic, it implies that AMD has kept bandwidth flat between Fiji and Vega rather than improving information technology further past pushing HBM2. Nvidia's GP100, which is fix to debut in 2022, offers 720GB/due south of memory bandwidth. 512GB/s of bandwidth would compete well against Nvidia'south GTX 1080 and Titan X, however, and AMD may take scaled its bandwidth accordingly. TDP on the new loftier-stop card is supposedly 225W, which would put information technology near Nvidia's Titan X.

The study suggests AMD volition launch Vega 10 first, with Vega 11 supposedly replacing Polaris ten later in the year. This would imply AMD volition waterfall HBM2 down to cheaper SKUs relatively quickly, just we'd nevertheless be surprised to see Vega 11 pushing into the $200 – $240 marketplace where Polaris is supposed to live. Meanwhile, there's supposedly a Vega 20 on the roadmap as well, with 32GB of HBM2 retentivity and a 7nm architecture. Current ship dates for 10nm put that process technology as available in 2022, which means we're looking at a 2022 or 2022 delivery date for a 7nm replacement.

This is less an consequence than it might seem. GlobalFoundries has said it wants to evangelize 7nm and skip over 10nm entirely, but even if it manages to pull this off, we've already seen evidence that GPU developers are capable of delivering operation improvements even while stuck on the same procedure node. Nvidia may have done more than of this than AMD did on 28nm, but there were real performance and functioning-per-watt improvements for GCN over the same time flow.

There'south no word on the specifics of Vega'due south architecture, and that'south what we'd need to make any predictions about overall performance. Our sources take indicated Vega is a footing-up new architecture rather than a GCN derivative — but what that means for AMD and its relative performance position to Nvidia is still unknown. AMD'south Polaris architecture delivered some real improvements over their 28nm production, but Nvidia'south Pascal family however holds the upper hand in price/functioning comparisons, thanks partly to the ongoing trouble AMD has had in delivering hardware that meets its actual MSRPs. NV has had bug of its own as well, just equally nosotros've discussed, not equally much as AMD has had.