For Tesla K80 NVIDIA has produced a new GPU – GK210 – and then put two of them into a single card. Given the costs in bringing a new GPU revision to market – just the masks alone are increasingly expensive – the situation implies that NVIDIA expects to more than make back their money on additional sales enabled by GK210, which in turn indicates that they have quite a bit of faith in the state of the GPU compute market since it alone would be where the additional revenue would come from.Fitting a pair of GPUs on a single card is not easy, and that is especially the case when those GPUs are GK210. That means fewer steps for anyone looking to maximize application performance and get the results fast.NVIDIA websites use cookies to deliver and improve the website experience. For Tesla K80 NVIDIA has produced a … Try a Tesla K80 GPU Today in the Cloud. A 300W TDP presents its own challenges, but in surmounting that it’s now possible to get 8 GK210 GPUs in a 1U form factor, which would put the FP64 compute throughput of such a setup at over 10 TFLOPS in 1U.Specifications aside, Tesla K80 represents an unexpected evolution in Tesla designs. In the Tesla space NVIDIA introduced this on Tesla K40 in a far more limited implementation than on their consumer GPUs. But with GM204 clearly ahead of GK110/GK210 in graphics, GK210 seems destined to Tesla cards and at most a Titan card for the budget compute market. Dubbed the Tesla K80, NVIDIA’s latest Tesla card is an unusual and unexpected entry into the Tesla lineup. Update your graphics card drivers today. That said, this also reflects on the state of the GPU market, and how Kepler will still be with us for some time to come.Meanwhile Tesla K80 will also be pushing the power envelope, again to get 2 GPUs on a single card.
The NVIDIA ® Tesla ® K80 Accelerator dramatically lowers data center costs by delivering exceptional performance with fewer, more powerful servers. That said, with K40 NVIDIA made clockspeeds deterministic for GPU workload sync issues, so it’s not entirely clear why non-deterministic clockspeeds are now okay just a year later. The net result is a card with no peers; NVIDIA has done dual GPU Tesla cards before (Tesla K10) and there have been dual GPU GK110 cards before (GeForce Titan Z), but nothing quite like Tesla K80.From both a performance and power standpoint, NVIDIA is expecting to once again raise the bar.
It’s a combination that shrinks the time to discovery. Whereas a GK110(B) SMX has a 256KB register file and 64KB of shared memory, GK210 doubles that to a 512KB register file and 128KB of shared memory. So far we have only seen passive cards, and given the need to move 300W of heat we expect that these cards will need to be passive in order to be paired up with appropriately powerful external fans.Wrapping things up, Tesla K80 will be a hard launch from NVIDIA and their partners, with individual cards and OEM systems equipped with them expected to be available today. Just talk to users like Wolfgang Nagel, director of the Center for Information Services and High Performance Computing, TU Dresden, and Yann LeCun, director of AI Research at Facebook and professor at New York University.Tesla K80 is a high-performance, cost-effective way to increase GPU density and ease of use. It's engineered to boost throughput in real-world applications by 5-10x, while also saving customers up to 50% for an accelerated data center compared to a CPU-only system. In the oil and gas industry, K10 made great strides with its single precision performance and high-memory bandwidth. Consequently energy efficiency gains are almost entirely reliant on what kind of performance Tesla K80 can sustain at 300W; the worst case scenario is that it’s only 2% more energy efficient than K40 while the best case is 59%, with the realistic case being somewhere in the middle.The change in implementation is no doubt driven by the more complex thermal environment of a multi-GPU card, not to mention the need to squeeze out yet more efficiency. Meanwhile GK210 will be in an odd place as it will likely be the first NVIDIA GPU not to end up in a consumer card; prior to this generation every GPU has pulled double duty as both a compute powerhouse and a graphics king.