AMD has unveiled the Radeon Pro V340, the first in its new Radeon Pro V Series of server graphics cards, designed for use in data centers and render farms.
A dual-GPU card based on AMD’s new Vega architecture, the Radeon Pro V340 features 32GB of HBM2 memory, divisible into up to 32 individual 1GB virtual machines.
Aimed at multi-user data centers and server farms
Like Nvidia’s Tesla GPUs and DGX GPU systems, the Radeon Pro V340 is designed for use in multi-user data centers and server farms rather than individual artists’ workstations.
Accordingly, the spec that AMD is promoting hardest is the option to divide its 32GB of on-board memory into 32 virtual machines – according to its press release, “up to 33% more than the competitive solution”.
That presumbably means the nearest-priced Nvidia solution, rather than the newest, since the comparison is to the old Tesla P40 rather than current-gen Nvidia cards like the Tesla V100.
The V340 uses AMD’s MxGPU GPU virtualisation technology, and features two virtualised encode engines, compressing video output streams into H.264 or H.265 formats.
Built on two Radeon Pro Vega 56 GPUs, with some serious passive cooling
The Radeon Pro V340 is built from two Radeon Pro Vega 56 GPUs, so the other specs are pretty much what you’d expect: 7,168 Stream processors, as opposed to the 3,584 in the Vega 56; and 112 compute units.
It’s a dual-slot card with a fairly hefty 300W TDP, although to judge from the exploded view above, it features some pretty serious passive cooling architecture.
Pricing and availability
The Radeon Pro V340 is expected to begin shipping in Q4 2018. AMD hasn’t announced pricing yet.
Find more specs for the Radeon Pro V340 on AMD’s product website