AMD Radeon RX Vega Graphics Card Spotted On Compubench and Gets Compute Benchmarked

Author Photo
Aug 9, 2017

AMD’s upcoming Radeon RX Vega graphics card has been spotted in the Compubench database. It is unclear which particular variant (RX Vega 64 or 56) of Vega GPU this graphics card is but it appears to be an early test run by some engineer or someone from the driver team. The Radeon RX Vega family of GPUs will be launching on August 12 and will mark the first high end Radeon update since the Fury lineup.

AMD’s Vega GPU Compubench performance gets leaked a few days before launch

The compute performance metrics used during the testing are sparse and this makes comparison and interpretation of this leak hard if not downright impossible. We do have one or two reference points which were found in the GTX 1080 entry as well so we have some inkling of the kind of performance to expect. The original benchmark entry can be found over here:

articleRelatedAMD Radeon RX Vega 64 Liquid Cooled Power Profiles Exploration

The RX Vega scores 4381 mVoxels/s in the Level Set Segmentation 256. The GTX 1080 entry available on the site did not have the test listed. The reference points I was talking about however, are listed and show inconsistent performance which would indicate that the test was conducted with half baked drivers. In one test, the Catmull-Clark Subdivision Level 3, the RX Vega scores approximately half the score of its green counterpart. In another, the Vertex Connection and Merging, they can be be seen trading blows and finally, in TV-L1 Optical Flow, we see the Vega GPU outperforming NVIDIA’s GTX 1080.

Out of all the benchmarks out there, Compubench is probably the least indicative of real life performance, so should be taken with a grain of salt (both for green campers and red campers). It is also worth pointing out that this was quite obviously a partial test run and one that was done on imperfect drivers. RX Vega’s launch is just a few days away and we should be getting pretty of unpartial reviews soon!

AMD’s high end RX Vega 64 graphics card will contain 64 Compute Units, which (assuming the same ratio of CU to SPs as the current iteration of GCN) translates to exactly 4096 Stream Processors. The internal codename of this GPU is GFX9 (in this case the device ID is 687F:C3). Remember all our internal nomenclature analysis? Well, it’s the same thing, only in a more appealing format. Hawaii was GFX7, Polaris is GFX8 and RX Vega is GFX9.

The RX Vega 56, as the name suggests, has 56 CUs with a stream processor count of 3,584. The Radeon RX Vega 56 will also feature 8 GB of HBM2 VRAM which will deliver a rated bandwidth of 410 GB/s along a 2048-bit bus interface. It will be using HBM2 just like its big brother and will retail for $$399. The chip will feature a total compute horsepower of 10.5 TFLOPs (FP32) and will feature a TDP of 210W which will be provided through an 8 and 6 pin power connector configuration.

android-loves-appleRelatedApple iPhone 8 Plus Vs Samsung’s Galaxy Note 8, Galaxy S8+ Vs LG’s G6 Vs The Nokia 8; A Much Needed Benchmark Comparison

This means that it is competitively priced to go against the GTX 1070 which originally retailed for $449. If the leaked RX Vega 56 performance numbers are true than we are easily looking at one of the performance per $ leaders of this generation. Just like the R9 290 and the Fury, it looks like AMD has reserved the value sweet spot for the second most powerful GPU in its lineup.

AMD Radeon RX Vega 64 and Vega 56 Graphics Card Lineup:

Graphics CardAMD Radeon R9 Fury XAMD Radeon RX Vega NanoAMD Radeon RX Vega 56 ReferenceAMD Radeon RX Vega 64 ReferenceAMD Radeon RX Vega 64 LimitedAMD Radeon RX Vega 64 Liquid
GPUFiji XTVega 10Vega 10Vega 10Vega 10Vega 10
Process Node28nm14nm FinFET14nm FinFET14nm FinFET14nm FinFET14nm FinFET
Compute Units64TBD56646464
Stream Processors4096TBD3584409640964096
Raster Operators646464646464
Texture Mapping Units256TBD224256256256
Clock Speed (Base)1000 MHzTBD1156 MHz1247 MHz1247 MHz1406 MHz
Clock Speed (Max)1050 MHzTBD1471 MHz1546 MHz1546 MHz1677 MHz
FP32 Compute8.6 TFLOPsTBD10.5 TFLOPs12.6 TFLOPs12.6 TFLOPs13.7 TFLOPs
FP16 Compute8.6 TFLOPsTBD21.0 TFLOPs25.2 TFLOPs25.2 TFLOPs27.4 TFLOPs
Memory Bus4096 bit2048 bit2048 bit2048 bit2048 bit2048 bit
Bandwidth512 GB/sTBD410 GB/s484 GB/s484 GB/s484 GB/s
($499 US Actual)
($599 US Actual)