NVIDIA’s RTX A6000 is the flagship Ampere GPU-based graphics card, and AdoredTV has leaked its first performance benchmarks. The benchmarks tell us what to expect from a card that will replace the Quadro RTX 6000 from 2018.
The RTX A6000 was announced earlier this month. NVIDIA has dropped the Quadro branding soon after dropping the Tesla branding; hence this card will be called RTX A6000 and will target the content creation and workstation marketing segment.
As far as the specifications are concerned, the full GA102 GPU core with 84 SMs or 10752 CUDA cores powers the RTX A6000. Although exact clock speeds are unknown as NVIDIA hasn’t revealed TFLOPs or Tensor numbers for the card yet, though it will offer higher clocks than the Quadro RTX A6000 as the RTX A6000’s blower design offers higher cooling performance.
The RTX A6000 features 48 GB of GDDR6 memory, and its 16 Gbps memory dies offer 768 GB/s speeds. It also supports vGPU with various graphics configurations that start from 1GB and go all the way to 48GB VRAM buffer. The new EPS 12V 8-pin connector(located on the back) delivers up to 300W of power to the GPU. NVILINK is the interconnect that offers speeds of 112.5 GB/s (bi-directional) as compared to the 16 GB/s speed provided by the native PCIe Gen 4. DisplayPort 1.4 is the display option for A6000.
The benchmarks of both the Quadro RTX 6000 and RTX A6000 achieved in SPECviewperf 2020 at a resolution of 3180×2120 were compared, and RTX A6000 was ahead in all of the benchmarks by 25%. It means that on average, RTX A6000 is 11% faster than the Quadro RTX 6000.
There’s no explanation as to why the RTX A6000 is only 11% faster than its predecessor but it may be due to many factors that include poor optimization of graphics in SPECviewperf or the graphic card drivers. NVIDIA will release RTX A6000 in mid of December through its partners, and wider availability is planned for the next year.