A Look At AMD’s Radeon RX Vega 64 Workstation & Compute Performance

AMD Radeon RX Vega 64 - Angled Shot
Print
by Rob Williams on August 14, 2017 in Graphics & Displays

After months and months of anticipation, AMD’s RX Vega series has arrived. The first model out-of-the-gate is the RX Vega 64, going up against the GTX 1080 in gaming. In lieu of a look at gaming to start our Vega coverage, we decided to go the workstation route – and we’re glad we did. Prepare yourself to be decently surprised.

Sandra: Cryptography, Science, Finance & Bandwidth

On the previous page, I mentioned that SPEC is an organization that crafts some of the best, most comprehensive benchmarks going, and in a similar vein, I can compliment SiSoftware. This is a company that thrives on offering support for certain technologies before those technologies are even available to the consumer. In that regard, its Sandra benchmark might seem a little bleeding-edge, but at the same time, its tests are established, refined, and accurate across multiple runs.

While Sandra offers a huge number of benchmarks, just four of the GPU ones are focused on: Cryptography, Financial Analysis, Scientific Analysis, and also memory bandwidth. Some of the results are a bit too complex for a graph, so a handful of tables are coming your way.

SiSoftware Sandra 2017

Cryptography

Sandra 2017 - Cryptography (High)
Sandra 2017 - Cryptography (Higher)

Hot damn. I feel like these kinds of gains are those that AMD should promote out the wazoo. So many reviews posted today are likely to paint a rough picture of this card’s gaming performance, but on the other side of the fence, compute performance on Vega quite simply kicks ass. The results here may be able to give an impression of Vega 64’s future mining performance. Mining benchmarks you’ll see around the web in other launch reviews will show an edge over a top-end NVIDIA card. I would not be surprised if AMD optimizes its driver sometime in the future to vastly improve mining performance. Vega should technically be better than the 30MH/s you’ll see reported today, based on all of the compute performance seen here.

Financial Analysis

Sandra 2017 – Financial Analysis (FP32)
Black-ScholesBinomialMonte Carlo
NVIDIA TITAN Xp14 G/s2.5 M/s6.7 M/s
NVIDIA Quadro P600011.6 G/s2.3 M/s6.5 M/s
NVIDIA GeForce GTX 1080 Ti11.6 G/s2.2 M/s6 M/s
AMD Radeon RX Vega 649.4 G/s3 M/s4.4 M/s
NVIDIA Quadro P40006.5 G/s1.1 M/s2.9 M/s
AMD Radeon Pro WX 71005.2 G/s1.3 M/s1.9 M/s
NVIDIA Quadro P20003.8 G/s656 k/s1.8 M/s
AMD Radeon Pro WX 51003.4 G/s478 k/s672 k/s
AMD Radeon Pro WX 41002.2 G/s531 k/s773 k/s
AMD Radeon Pro WX 31002.5 G/s321 k/s467 k/s
Results in options-per-second. 1 GOPS = 1,000 MOPS; 1 MOPS = 1,000 kOPS.
Sandra 2017 – Financial Analysis (FP64)
Black-ScholesBinomialMonte Carlo
AMD Radeon RX Vega 642.2 G/s186 k/s542 k/s
NVIDIA TITAN Xp1.44 G/s142 k/s297 k/s
NVIDIA Quadro P60001.3 G/s131 k/s271 k/s
NVIDIA GeForce GTX 1080 Ti1.3 G/s134 k/s272 k/s
NVIDIA Quadro P4000622 M/s63 k/s129 k/s
AMD Radeon Pro WX 7100958 M/s81 k/s239 k/s
NVIDIA Quadro P2000360 M/s36 k/s75 k/s
AMD Radeon Pro WX 5100406 M/s49 k/s97 k/s
AMD Radeon Pro WX 4100395 M/s35 k/s98 k/s
AMD Radeon Pro WX 3100219 M/s18 k/s55 k/s
Results in options-per-second. 1 GOPS = 1,000 MOPS; 1 MOPS = 1,000 kOPS.

The RX Vega 64 continues to perform extremely well in compute tests, slotting in just behind the 1080 Ti in single-precision, and leading the pack in a very significant way in the double-precision test. That’s thanks to the fact that Vega’s architecture delivers 1:16 the performance of single-precision, in double-precision, vs. NVIDIA’s 1:32.

Scientific Analysis

Sandra 2017 – Scientific Analysis (FP32)
GEMMFFTN-Body
NVIDIA Quadro P60006.4 TFLOPS495 GFLOPS5.9 TFLOPS
NVIDIA TITAN Xp7 TFLOPS258 GFLOPS5.5 TFLOPS
AMD Radeon RX Vega 646 TFLOPS344 GFLOPS5.3 TFLOPS
NVIDIA GeForce GTX 1080 Ti6 TFLOPS217 GFLOPS5.12 TFLOPS
NVIDIA Quadro P40003.1 TFLOPS128 GFLOPS2.7 TFLOPS
AMD Radeon Pro WX 71002.5 TFLOPS210 GFLOPS2.2 TFLOPS
NVIDIA Quadro P20001.8 TFLOPS87 GFLOPS1.7 TFLOPS
AMD Radeon Pro WX 5100945 GFLOPS138 GFLOPS755 GFLOPS
AMD Radeon Pro WX 41001 TFLOPS85 GFLOPS917 GFLOPS
AMD Radeon Pro WX 3100670 GFLOPS70 GFLOPS647 GFLOPS
GEMM = General Matrix Multiply; FFT = Fast Fourier Transform; N-Body = N-Body Simulation.
Sandra 2017 – Scientific Analysis (FP64)
GEMMFFTN-Body
AMD Radeon RX Vega 64611 GFLOPS164 GFLOPS475 GFLOPS
NVIDIA TITAN Xp352 GFLOPS199 GFLOPS277 GFLOPS
NVIDIA GeForce GTX 1080 Ti332 GFLOPS163 GFLOPS267 GFLOPS
NVIDIA Quadro P6000323 GFLOPS128 GFLOPS253 GFLOPS
NVIDIA Quadro P4000156 GFLOPS95 GFLOPS127 GFLOPS
AMD Radeon Pro WX 7100276 GFLOPS80 GFLOPS194 GFLOPS
NVIDIA Quadro P200090 GFLOPS53 GFLOPS84 GFLOPS
AMD Radeon Pro WX 5100123 GFLOPS56 GFLOPS103 GFLOPS
AMD Radeon Pro WX 4100109 GFLOPS33 GFLOPS84 GFLOPS
AMD Radeon Pro WX 310063 GFLOPS33 GFLOPS49 GFLOPS
GEMM = General Matrix Multiply; FFT = Fast Fourier Transform; N-Body = N-Body Simulation.

With its beefy double-precision performance (as far as gaming cards go, at least), Vega 64 soars to the top of that respective chart, and manages to best the 1080 Ti in single-precision.

Bandwidth

Sandra 2017 - Memory Bandwidth

With its HBM2 memory in tow, the RX Vega 64 places right behind NVIDIA’s GTX 1080 Ti, by about 27GB/s. Its interface transfer, however, manages to best everything else in the lineup. How this correlates to real-world performance is hard to gauge, especially since it’s up to the rest of a GPU’s architecture to properly complement the bandwidth it’s given.

Rob Williams

Rob founded Techgage in 2005 to be an 'Advocate of the consumer', focusing on fair reviews and keeping people apprised of news in the tech world. Catering to both enthusiasts and businesses alike; from desktop gaming to professional workstations, and all the supporting software.

twitter icon facebook icon googleplus icon instagram icon