facebook rss twitter

Nvidia details Tegra 5 and 64-bit Tegra 6 SoC plans

by Mark Tyson on 20 March 2013, 11:22

Tags: NVIDIA (NASDAQ:NVDA), PC

Quick Link: HEXUS.net/qabt4z

Add to My Vault: x

At the GPU Technology Conference (GTC) in San Jose, California yesterday, Nvidia CEO Jen-Hsun Huang fleshed out details of the Tegra SoC roadmap. He told attendees at his keynote speech that the Tegra 5 “Logan” is to include a Kepler-class GPU and will debut this year before full scale production in 2014. The Tegra 6 SoC “Parker” is going to be based upon a 64-bit ARM CPU and Nvidia’s next generation Maxwell GPU. Huang said that the journey from Tegra 2 to Tegra 6 will have brought a 100 times improvement in performance.

Tegra 5 “Logan”

Huang told attendees at GTC that the Tegra 5 is going to be a processor that will allow mobile devices to do “everything a modern computer should do”. Tegra 5 will be the first mobile processor to incorporate a Kepler-class GPU which will be fully CUDA 5 and OpenGL 4.3 compliant “out of the box”. CUDA allows the GPU to be used for many computational tasks other than graphics and image processing.

The Nvidia CEO said that Logan will debut this year before full scale production in early 2014.

Tegra 6 “Parker”

Parker will be the first Tegra SoC to be based upon ARM’s 64-bit Project Denver CPU. This processor is destined to be produced using the 3D finFET process which will have beneficial power consumption and performance effects. The Tegra 6 will also feature Nvidia’s Maxwell GPU which is due to be seen in desktop add-in-cards next year. The Maxwell GPU has a unified virtual memory architecture “All memory will be visible to all the processors” including the compute cores, explained Huang.

When the Tegra 6 Parker SoC is released in 2015 it will achieve 100x the performance of the Tegra 2 according to Huang. If Tegra progress had followed Moore’s Law the processor would have only been about eight times faster but “disruptive technology” had brought about this “quite staggering” rate of innovation.



HEXUS Forums :: 6 Comments

Login with Forum Account

Don't have an account? Register today!
To be fair, Tegra 2 was a complete dog, so 100x faster than that isn't as special as it sounds. OTOH Tegra 3 was alright and if we take the marketing slide above as being even slightly accurate then it will be 20x faster than that. But is that CPU, or combined CPU+GPU - Tegra 3's GPU is quite weak.
What ever Nvida can do, Samsung / Qualcomm seem to do better.
“The Maxwell GPU has a unified virtual memory architecture”

So, like AMD is introducing this year, they mean? ;)
The live demo of it running CUDA and OpenGL programs was pretty impressive I have to say.
More competition, more choice…

Always great for consumers.