Announces Tesla M40 and Tesla M4
After introducing the Maxwell GPU to its Tesla M60 and M6 cards for VDI And video encoding, Nvidia is now bringing Maxwell GPUs to Tesla server cards with the Tesla M40 and Tesla M4 cards.
Calling them the "hyperscale accelerators", the new Nvidia Tesla M40 and Tesla M4 are aimed at data center machine learning market. While the Maxwell GPU is generally not suitable for high-precision HPC work, Nvidia can use it for every other server use, including the deep center machine learning, exactly what the new Tesla M40 and Tesla M4 are made for.
Based on the GM200 GPU, the Tesla M40 packs 3072 CUDA cores and packs 12GB of GDDR5 VRAM. While Maxwell GPU does not support on-chip ECC for memory and cache, Nvidia might have implemented a software ECC on these cards. The Tesla M40 will offer up to 7 TFLOPS of single-precision compute performance and up to 0.21 TFLOPS of double-precision compute performance as well as have a passive cooling solution, due to its "lower" 250W TDP.
The Tesla M4 on the other hand is a PCIe half-height low-profile form factor server card aimed at dense cluster servers. Based on Nvidia's GM206 Maxwell GPU, it packs 1024 CUDA cores and 4GB of GDDR5 memory, offering up to 2.2 TFLOPS of single- and up to 0.07 TFLOPS of double-precision compute performance.
Both new Maxwell Tesla cards are meant of machine learning marketing, which has been in Nvidia focus for the last year. Nvidia actually believes that machine learning is the next big market for GPUs and with the recently introduced Drive PX system for automotive sector, it is obvious that Nvidia will keep focusing on that market.
Nvidia's new Tesla M40 and M4 cards will offer greater performance and combined with the new Nvidia Hyperscale Suite, offer much easier execution of various machine learning tasks. The Telsa M40 will deal with training tasks, for neural nets and other systems, the Tesla M4, on the other hand, will focus on dense cluster of systems executing various machine learning tasks.
Nvidia expects that the Tesla M40 and the Hyperscale Software Suite will be available later this year while the Tesla M4 should come in Q1 2016.
Source: Anandtech.com.
Calling them the "hyperscale accelerators", the new Nvidia Tesla M40 and Tesla M4 are aimed at data center machine learning market. While the Maxwell GPU is generally not suitable for high-precision HPC work, Nvidia can use it for every other server use, including the deep center machine learning, exactly what the new Tesla M40 and Tesla M4 are made for.
Based on the GM200 GPU, the Tesla M40 packs 3072 CUDA cores and packs 12GB of GDDR5 VRAM. While Maxwell GPU does not support on-chip ECC for memory and cache, Nvidia might have implemented a software ECC on these cards. The Tesla M40 will offer up to 7 TFLOPS of single-precision compute performance and up to 0.21 TFLOPS of double-precision compute performance as well as have a passive cooling solution, due to its "lower" 250W TDP.
The Tesla M4 on the other hand is a PCIe half-height low-profile form factor server card aimed at dense cluster servers. Based on Nvidia's GM206 Maxwell GPU, it packs 1024 CUDA cores and 4GB of GDDR5 memory, offering up to 2.2 TFLOPS of single- and up to 0.07 TFLOPS of double-precision compute performance.
Both new Maxwell Tesla cards are meant of machine learning marketing, which has been in Nvidia focus for the last year. Nvidia actually believes that machine learning is the next big market for GPUs and with the recently introduced Drive PX system for automotive sector, it is obvious that Nvidia will keep focusing on that market.
Nvidia's new Tesla M40 and M4 cards will offer greater performance and combined with the new Nvidia Hyperscale Suite, offer much easier execution of various machine learning tasks. The Telsa M40 will deal with training tasks, for neural nets and other systems, the Tesla M4, on the other hand, will focus on dense cluster of systems executing various machine learning tasks.
Nvidia expects that the Tesla M40 and the Hyperscale Software Suite will be available later this year while the Tesla M4 should come in Q1 2016.
Source: Anandtech.com.
News by Luca Rocchi and Marc Büchel - German Translation by Paul Görnhardt - Italian Translation by Francesco Daghini
Previous article - Next article
comments powered by Disqus
comments powered by Disqus