Home / Component / Graphics / Nvidia unveils Tesla K80: dual-GPU accelerator with 24GB of GDDR5 memory

Nvidia unveils Tesla K80: dual-GPU accelerator with 24GB of GDDR5 memory

Nvidia Corp. on Monday unveiled what it claims to be the most powerful compute accelerator on the planet. The Tesla K80 dual-GPU solution sports 4992 stream processors as well as 24GB GDDR5 memory in a bid to offer unprecedented performance for difficult computational challenges, ranging from astrophysics, genomics and quantum chemistry to data analytics.

Nvidia Tesla K80 accelerator carries two brand-new GK210 processors developed specifically for high-performance computing market. Just like the GK110 graphics chips, the GK210 features up to 2880 stream processors (and up to 15 streaming multiprocessor clusters) as well as 384-bit memory bus. However, in a bid to offer better performance in compute applications, Nvidia increased the size of L1 cache of every SM cluster to 128KB (up from 64KB) and enlarged register file of every SM cluster to 512KB (up from 256KB). The new caching sub-system should significantly improve performance in numerous memory-bound applications.


The Tesla K80 accelerator is based on two GK210 processors with 2496 stream processors enabled (4992 stream processors in total). The GPUs operate at clock-rates between 562MHz and 875MHz thanks to GPU Boost technology. 24GB of GDDR5 memory function at 5.0GHz. effective frequency. The Tesla K80 delivers up to 8.74TFLOPS single-precision and up to 2.91TFLOPS double precision peak floating point performance.


“Nvidia GPUs have become the de facto computing platform for the deep learning community,” said Yann LeCun, director of AI research at Facebook, and silver professor of computer science and neural science at New York University. “Because the accuracy of deep learning systems improves as the models and datasets get larger, we always look for the fastest hardware we can find. The Tesla K80 accelerator, with its dual-GPU architecture and large memory, gives us more teraflops and more GB than ever before from a single server, allowing us to make faster progress in deep learning.”


The Tesla K80 compute accelerators are available inside HPC servers and as compute cards. Each card costs $5000.

Discuss on our Facebook page, HERE.

KitGuru Says: It looks like the code-named GM200 graphics processing unit is not really ready, which is why the company uses the GK210 chips on its premium accelerator. Still, it seems that GK210 will not be used on consumer graphics cards, its enlarged caches will not give any advantages to graphics applications, whereas additional complexity will likely limit clock-rate potential.

Check Also

Advent Calendar Day 4: Win a Palit GTX 1070Ti!

It's getting close to Christmas, so we are getting in the holiday spirit! For day four, we are teaming up with PALIT, giving readers the chance to win a Palit GTX 1070Ti Super JetStream graphics card!

  • Imran

    Do scientists use these cards? NASA?

  • Valdyr

    It is designed mainly for specific types of servers, “designed for high-end applications that require high performance computing clusters” Aka for HPC servers and rigs with high end applications. I can see this primarily used by different types of scientists, server rigs with opterons and xeons, developers, and high end multimedia editing…possibly gaming server rigs/systems too.

  • Mezinis Gugalis

    but can it run dayz?

  • TT


  • Vivek

    Jesus that’s an enormous pile of ram. Are they trying to compete with quantum computers or something?

  • WiNG

    But can it run AC Unity?

  • Guest

    Interesting faster then TitanZ while having lower powerdraw on the same architecture. It still drawns slightly more power then the W9100 but it has the performance edge in double precession.

    295X2 remains king of single precession though…