NVIDIA’s second GTC of 2020 is going down this week, and as has rapidly turn into a practice, considered one of CEO Jensen Huang’s “kitchenside chats” kicks off the occasion. As the de facto alternative for GTC Europe, this fall digital GTC is a little bit of a lower-key occasion relative to the Spring version, however it’s nonetheless one that’s seeing some NVIDIA {hardware} launched to the world.
Starting issues off, we’ve got a pair of latest video playing cards from NVIDIA – and a launch that seemingly signifies that NVIDIA is on the brink of overhaul its skilled visualization branding. Being introduced right this moment and set to ship on the finish of the yr is the NVIDIA RTX A6000, NVIDIA’s next-generation, Ampere-based skilled visualization card. The successor to the Turing-based Quadro RTX 8000/6000, the A6000 shall be NVIDIA’s flagship skilled graphics card, providing every thing underneath the solar so far as NVIDIA’s graphics options go, and chart-topping efficiency to again it up. The A6000 shall be a Quadro card in every thing however title; actually.
NVIDIA Professional Visualization Card Specification Comparison |
||||||
A6000 | A40 | RTX 8000 | GV100 | |||
CUDA Cores | 10752 | 10752 | 4608 | 5120 | ||
Tensor Cores | 336 | 336 | 576 | 640 | ||
Boost Clock | ? | ? | 1770MHz | ~1450MHz | ||
Memory Clock | 16Gbps GDDR6 | 14.5Gbps GDDR6 | 14Gbps GDDR6 | 1.7Gbps HBM2 | ||
Memory Bus Width | 384-bit | 384-bit | 384-bit | 4096-bit | ||
VRAM | 48GB | 48GB | 48GB | 32GB | ||
ECC | Partial (DRAM) |
Partial (DRAM) |
Partial (DRAM) |
Full | ||
Half Precision | ? | ? | 32.6 TFLOPS | 29.6 TFLOPS | ||
Single Precision | ? | ? | 16.three TFLOPS | 14.eight TFLOPS | ||
Tensor Performance | ? | ? | 130.5 TFLOPS | 118.5 TFLOPs (FP16) |
||
TDP | 300W | 300W | 295W | 250W | ||
Cooling | Active | Passive | Active | Active | ||
NVLink | 1x NVLink3 112.5GB/sec |
1x NVLink3 112.5GB/sec |
1x NVLInk2 50GB/sec |
2x NVLInk2 100GB/sec |
||
GPU | GA102 | GA102 | TU102 | GV100 | ||
Architecture | Ampere | Ampere | Turing | Volta | ||
Manufacturing Process | Samsung 8nm | Samsung 8nm | TSMC 12nm FFN | TSMC 12nm FFN | ||
Launch Price | ? | ? | $10,000 | $9,000 | ||
Launch Date | 12/2020 | Q1 2021 | This fall 2018 | March 2018 |
The first skilled visualization card to be launched primarily based on NVIDIA’s new Ampere structure, the A6000 can have NVIDIA hitting the market with its finest foot ahead. The card makes use of a fully-enabled GA102 GPU – the identical chip used within the GeForce RTX 3080 & 3090 – and with 48GB of reminiscence, is full of as a lot reminiscence as NVIDIA can placed on a single GA102 card right this moment. Notably, the A6000 is utilizing GDDR6 right here and never the quicker GDDR6X used within the GeForce playing cards, as 16Gb density RAM chips should not out there for the latter reminiscence presently. As a outcome, regardless of being primarily based on the identical GPU, there are going to be some fascinating efficiency variations between the A6000 and its GeForce siblings, because it has traded reminiscence bandwidth for total reminiscence capability.
In phrases of efficiency, NVIDIA is selling the A6000 as providing almost twice the efficiency (or extra) of the Quadro RTX 8000 in sure conditions, significantly duties profiting from the numerous improve in FP32 CUDA cores or the same efficiency improve in RT core throughput. Unfortunately NVIDIA has both but to lock down the specs for the cardboard or is opting in opposition to saying them presently, so we don’t know what the clockspeeds and ensuing efficiency in FLOPS shall be. Notably, the A6000 solely has a TDP of 300W, 50W decrease than the GeForce RTX 3090, so I might count on this card to be clocked decrease than the 3090.
Otherwise,…