AMD has recently added two Vega device IDs to Linux patches, one of them was 6864:00.
AMD Vega: 16GB HBM2 and 1600 MHz clock
The new ID just popped up on CompuBench database. It is undoubtedly Vega: it was detected as gfx900 architecture, has 64 Compute Units and has new IDs confirmed by Linux patch to be Vega.
The reason why I didn’t name this post “Radeon RX Vega” is because we don’t know if this chip is Radeon RX, Radeon Pro or Radeon Instinct. It would work as a clickbait, but it might be misleading. AMD only confirmed two stack HBM2 for Radeon RX Vega and showcased a sample with 8GB on board. We simply don’t know if the 16GB HBM2 chip is planned for Radeon RX series.
Either way, the most interesting part of this leak is maximum clock frequency (CL_DEVICE_MAX_CLOCK_FREQUENCY). For comparison, Radeon RX 480 in CompuBench database has the same entry with 1266 MHz and Radeon RX 580 up to 1430 MHz (because most cards are overclocked). This could be an indicator that Vega has 1600 MHz maximum GPU clock. Whether that’s a frequency limit or actual boost clock, we don’t know yet.
Lastly, we have a confirmation that this particular ID has 64 Compute Units enabled. If AMD kept the same principle for GFX9 architecture (64 cores per cluster), then we should expect 4096 Stream Processors (64*64).
|CL_DEVICE_GLOBAL_MEM_SIZE||16978542592||bytes =~ 16 GB|
|CL_DEVICE_MAX_CLOCK_FREQUENCY||1600||Maximum Boost Clock (MHz)|
|CL_DEVICE_MAX_COMPUTE_UNITS||64||64 * 64 = 4096 Stream Processors|
Many thanks to Miklos for the tip!