NVIDIA independently compared the capabilities of Google TPU with P40 accelerators

	xtreview is your : Video card - cpu - memory - Hard drive - power supply unit source

Recommended : Free unlimited image hosting with image editor

POSTER: computer news   ||    NVIDIA INDEPENDENTLY COMPARED THE CAPABILITIES OF GOOGLE TPU WITH P40 ACCELERATORS	DATE:2017-04-11
Last week, Google published the results of testing the Tensor Processing Unit (TPU) branded solutions based on customized LSIs and compared them with the NVIDIA K80 accelerators and Intel Xeon processors.  According to the results obtained in Google, to solve problems with deep machine learning TPU Google are 30-80 times better than the serial products of Intel and NVIDIA.  Yesterday, NVIDIA questioned the validity of Google's findings.  More precisely , it suggested comparing the capabilities of Google's TPU with the current P40 accelerators, not the K80 adapters that were released five years ago. According to NVIDIA, the P40 accelerators double the Google TPU by the ratio of decisions taken per second with a delay level of less than 10 ms.  To be more precise, the operation for making a decision (inferences) already includes operations on machine training.  At the same time, NVIDIA is confident that a combination of high performance in single-precision calculations at the level of 12 teraflops and a 10-fold advantage of the P40 when accessing memory allows the NVIDIA solution to demonstrate superiority over the Google TPU. At the same time, when working with 8-bit operations typical of tensor calculus, Google's TPU promises the most optimal execution of the computational load.  The TPU consists of 98 executive blocks INT8, and in the GPU P40 there are only 48. Due to this TPU, Google shows a good result when processing tensor arrays with a consumption of 75 watts compared to the 250-watt NVIDIA P40 accelerator.  It is worth noting that NVIDIA supported the decision of Google to create an accelerator for deep machine learning using tensors.  According to experts, this is the most correct way for processing large data sets for a wide range of tasks related to machine learning.  Intel, by the way, also creates an accelerator for processing tensors (Lake Crest).   Related Products :

	xtreview is your : Video card - cpu - memory - Hard drive - power supply unit source

Xtreview Support N-Post:xxxx Xtreview Support	NVIDIA INDEPENDENTLY COMPARED THE CAPABILITIES OF GOOGLE TPU WITH P40 ACCELERATORS
	Please Feel Free to write any Comment; Thanks

Related news:

We are looking for copper coolers on NVIDIA Volta computing accelerators (2017-09-08)
NVIDIA told about the progress of Volta in comparison with Pascal (2017-08-23)
AMD continues to demonstrate the capabilities of its processors using NVIDIA graphics cards (2017-08-23)
NVIDIA is among the top three out-of-the-box developers of microchips in terms of revenue (2017-08-17)
TSMC will increase the output of 12-nm NVIDIA products by the end of the year (2017-08-15)
THE SHARES OF NVIDIA CONTINUE TO FALL IN PRICE (2017-08-14)
THE HEAD OF NVIDIA ONCE AGAIN ANNOUNCED THE SUPERIORITY OF THE GPU OVER THE TPU (2017-08-14)
The head of NVIDIA admitted that Volta costs a lot in production (2017-08-11)
The financial optimism of NVIDIA is due to a profitable strategic position (2017-08-11)
In the second quarter, NVIDIA revenue grew 1.5 times (2017-08-11)
Over the past three months, NVIDIA has earned at least 150 million (2017-08-11)
NVIDIA promises to release a game Volta, but while calls to buy Pascal (2017-08-11)
The share price of NVIDIA shares rose in anticipation of quarter results announcement (2017-08-08)
Analysts believe in the ability of AMD Radeon RX Vega to squeeze NVIDIA products (2017-08-02)
NVIDIA TITAN Xp is now more popular with professionals (2017-08-01)
Miners could raise the revenue of AMD and NVIDIA by up to 875 million (2017-07-18)
NVIDIA GeForce GTX G-Assist an April Fools joke that revived (2017-07-18)
Experts believe in the artificial-intellectual potential of NVIDIA (2017-07-13)
The characteristics of NVIDIA Volta prove that Moores law is more likely to live (2017-07-10)
Miner solutions based on NVIDIA GP104 have already been patented in GPU-Z (2017-07-07)

Post Quick Reply:

                                        
Verif code:

                 

computer news computer parts review Old Forum Downloads New Forum Login Join Articles terms Hardware blog Sitemap Get Freebies