AMD has spent the previous couple of years difficult Intel throughout the desktop, server, and cellular markets, however the hole between the 2 firms is arguably largest in server. At current, AMD ships as much as 64 cores in a single socket, the place Intel has solely stepped as much as delivery 40 cores this week with the launch of Ice Lake SP. Earlier Intel Cascade Lake CPUs topped out at 28 cores. A brand new leak suggests Intel’s next-generation CPU platform, codenamed Sapphire Rapids, will lastly search to scale back a few of the gaps between itself and AMD’s Epyc.
As at all times, take this leak together with your day by day dose of salt. This slide comes from VideoCardz and it builds on some information we’ve beforehand seen.
Sapphire Rapids, when it launches, will (supposedly) specify one other TDP enhance — as much as 350W, this time. AMD’s present “Milan” CPUs high out at 280W, identical to Rome. Reminiscence assist strikes to DDR5, as anticipated, and the slide claims Sapphire Lake provides 1TB of bandwidth on 64GB of HBM2E. We knew Sapphire Lake was going to supply HBM2E as an possibility, however 64GB of on-die reminiscence with 1TB/s of bandwidth is big. It’d be actually attention-grabbing to see how system efficiency scaling modifications with this configuration in contrast with fashions with out HBM2.
A top-end Sapphire Rapids, if these rumors are correct, would supply a small pool of ultra-high bandwidth reminiscence, backed by a far bigger pool of lower-bandwidth reminiscence. An eight-channel DDR5 system utilizing DDR5-4800 would supply 307.2GB/s of reminiscence bandwidth to as much as 4TB of RAM (assuming Intel retains present Ice Lake SP limits).
Sapphire Rapids is alleged to characteristic as much as 80 PCIe 5.zero lanes on some SKUs, with others restricted to simply 64 lanes. It’s a four-tile design. This meshes with what we’ve realized about Intel’s plan for tiles, that are roughly analogous to AMD’s chiplets, however with totally different methods for I/O, package deal routing, and interconnects.
As for when these chips can be in-market, that’s a little bit onerous to learn proper now. Intel has made noise about delivery Sapphire Rapids in 2021, however we’ve additionally heard that the chip wasn’t more likely to launch earlier than 2022. Prior to now, there was once a big distinction between TSMC and Intel when it got here to the query of “quantity manufacturing.” That distinction is shrinking.
Intel would use the time period just a few months earlier than a chip went on sale, whereas TSMC may announce quantity manufacturing so long as a yr earlier than chips grew to become out there to shoppers. Intel claimed to be in quantity manufacturing for Ice Lake SP in January 2021 and launched in April, however reports from Dell recommend servers with the CPU received’t be out there till Might, and that that is “in sync with Intel’s timelines.” A January quantity announcement with Might availability is a four-month delay. That’s longer than is typical for Intel.
As of this writing, we’re guessing Sapphire Rapids will pattern in 2021 however not launch till 2022. It’ll compete in-market towards a mix of Milan and Genoa components. Genoa is anticipated to be constructed on 5nm and to make use of AMD’s Zen four structure. There are rumors of an extra core rely enhance, as much as 96 cores, however which will or is probably not true.
With Zen three, AMD targeted on bettering Infinity Cloth efficiency and clock speeds, nevertheless it wound up spending considerably extra energy on “uncore” actions than Rome did. The corporate might select to give attention to bettering IF and CPU effectivity with Zen four and maintain core counts equal, or it might choose to make the most of 5nm’s density enhancements and push core counts as soon as once more. 96 cores with out HBM2 and 12 reminiscence channels tackling 56 cores with HBM2 and eight reminiscence channels? Sounds fascinating to us.
This slide additionally mentions third-generation Optane, aka Crow Move, and claims bandwidth could possibly be improved by as much as 2.6x in blended learn/write situations. Not one of the information concerning Optane has been good currently, to the purpose that we’re watching to see if Crow Move even involves market. Assuming that it does, nevertheless, it appears to be like just like the reminiscence normal will lastly get an actual efficiency kick. No phrase on whether or not Crow Move helps PCIe four.zero or PCIe 5.zero, however Intel is clearly pushing to get Xeon again on a aggressive footing. Ice Lake SP is a stable effort for Chipzilla, nevertheless it doesn’t totally shut the hole with AMD. Sapphire Rapids provides Intel one other shot at doing so.