AMD Slims Down Compute With Radeon Skilled W7900 Twin Slot For AI Inference

Whereas the vast majority of AMD’s Computex presentation was on CPUs and their Instinct lineup of devoted AI accelerators, the company moreover has a small product refresh for the expert graphics and workstation AI crowd. AMD is releasing a dual-slot mannequin of their high-end Radeon Skilled W7900 card – aptly named the W7900 Twin Slot – with the intent being to reinforce compute density in workstations by making it doable to place in 4 of the enjoying playing cards inside a single chassis.

The discharge of a dual-slot mannequin of the cardboard comes after the distinctive Radeon Skilled W7900 was the first time AMD went with an even bigger, triple-slot kind difficulty for his or her flagship workstation card. With the W7000 period bringing an all-around enhance in power consumption, pushing the W7900 to 295 Watts, AMD initially opted to launch an even bigger card for improved acoustics. Nonetheless this acquired right here on the worth of compute density, as most applications would possibly solely match 2 of the thicker enjoying playing cards. In consequence, AMD is opting to launch a dual-slot mannequin of the {{hardware}} as successfully, to provide a further aggressive product for high-density workstation applications – considerably these doing native AI inference.

Aside from the narrower cooler, the Radeon Skilled W7900DS is for all intents and features equal to the distinctive W7900, with the an identical Navi 31 GPU being pushed to the an identical clockspeeds, and the final board being run to the an identical 295 Full Board Vitality (TBP) prohibit. That’s paired with the an identical 18Gbps GDDR6 as sooner than, giving the cardboard 48GB of VRAM.

Formally, AMD doesn’t have a noise specification for these enjoying playing cards. Nonetheless you might anticipate that the W7900DS will doubtless be louder than its triple-slot senior. By all appearances, AMD is just using the cooler from the W7800, which was a dual-slot card from the start, so that cooler is being tasked with coping with one different 35W of heat dissipation.

As a result of the W7800 was moreover AMD’s quickest dual-slot card up until now, it’s an apt stage of comparability for compute density. With its full-fat Navi 31 GPU, the W7900DS will present about 36% further compute/pixel throughput than its sibling/predecessor. So it’s a not-insubstantial enchancment for the very specific space of curiosity AMD has in ideas for the cardboard.

And like so many alternative points being launched at Computex this 12 months, that space of curiosity is AI. Whereas AMD affords PCIe variations of their Instinct MI210 accelerators, these enjoying playing cards are geared at servers, with fully-passive coolers to match. So workstation-level compute is mainly picked up by AMD’s Radeon Skilled workstation enjoying playing cards, which might be presupposed to enter a traditional PC chassis and use energetic cooling (blowers). On this case, AMD is especially going after native inference workloads, as that’s what the Radeon {{hardware}} and its essential VRAM pool are biggest fitted to.

The Radeon Skilled W7900 Twin Slot will drop on June 19^th. Notably, AMD is introducing the cardboard at a barely decrease value tag than they launched the distinctive W7900 in the end 12 months, with the W7900DS hitting retail cupboards at $3499, down from the W7900’s genuine $3999 price tag.

ROCm 6.1 For Radeons Coming as Successfully

Alongside the discharge of the W7900DS, AMD could be promoting the upcoming Radeon launch of ROCm 6.1, their software program program stack for GPU computing. Whereas baseline ROCm 6.1 was launched once more in April, the Dwelling home windows mannequin of AMD’s software program program stack continues to be a trailing (and have restricted) launch. In order that’s slated to lastly get bumped as a lot as a ROCm 6.1 launch on June 19^ththe an identical day the W7900DS launches.

ROCm 6.1 for Radeons is slated to ship a couple of primary changes/enhancements to the stack, considerably within the case of accelerating the scope of accessible choices. Notably, AMD will lastly be transport Dwelling home windows Subsystem for Linux 2 (WSL2) help, albeit at a beta stage, allowing Dwelling home windows clients to entry the quite a bit richer perform set and software program program ecosystem of ROCm beneath Linux. This launch may even incorporate improved help for multi-GPU configurations, wonderful timing for the launch of the Radeon Skilled W7900DS.

Lastly, ROCm 6.1 sees TensorFlow built-in into the ROCm software program program stack as a first-class citizen. Whereas this matter contains further complexities than is likely to be summarized in a simple data story, native TensorFlow help beneath Dwelling home windows was beforehand blocked by a shortage of a Dwelling home windows mannequin of AMD’s MIOpen machine finding out library. Blended with WSL2 help, builders may have two strategies to entry TensorFlow on Dwelling home windows applications going forward.