4 ms·

Nvidia Tegra X1 Preview and Architecture Analysis

55 points by ismavis 12y ago

higherpurpose 12y ago
> X1’s GPU is looking very good out of the gate – at least when tuned for power over performance. Isn't it such a shame Nvidia never seems to do that in mobile, then? Tune it for power, that is. Yes, the GPU and the chip overall are "impressive", but all of Nvidia's mobile chips have been at the time of launch (perhaps with the exception of Tegra 4). I really hate to say this, but I want Nvidia to succeed and it really bothers me that after so many Tegra generations they still don't get it. The problem is Nvidia always over-optimizes for performance, and then the whole chips turns into a dud that nobody wants, whether OEMs or consumers. Instead of doing "2x" performance increase, I would've been happier with a 1.5x performance increase, at a much more manageable and sustainable power envelope for a mobile device. I'm not sure why they keep doing this despite this strategy not working. Doing the same thing over and over again and seeing no results is the definition of insanity after all. Perhaps it's the whole "PC-centric" DNA they have, which makes them think in terms of PCs that have plenty of ways to cool down. I guess this is why "incumbents" keep failing when they get disrupted, even if they adopt the disruption technology relatively early. The old ways of doing things are still ingrained in every decision they make. And what happened to that Icera modem technology they bought? It's been years now but still not seeing it integrated into their chips. Have they abandoned all hope of entering the smartphone market already? They seem to already be refocusing quite aggressively towards the automotive market.
- modeless 12y ago
  They need 2x performance increases every year to keep up with Apple/Imgtec and Qualcomm. The only reason their performance looks impressive is they announce long before shipping and compare their unobtanium to everyone else's shipping devices. If you compare shipping devices to shipping devices (e.g. the TK1 vs. A8X benchmarks they have in the article) they're not ahead on performance.
  - listic 12y ago
    As far as I can see, current gen NVidia tablet chips (K1) are still on top, rivaling even some previous-gen (1.5 years old) x86's in graphics performance. http://www.futuremark.com/hardware/mobile http://www.futuremark.com/hardware/mobile Aren't they?
    - Grazester 12y ago
      The K1 I think was the first chip Nvidia really impressed me with.
- xorcist 12y ago
  Perhaps there are still places where you should optimize for performance where they still have a fighting chance to make it in the marketplace, such as the self driving car use case they hyped during the press conference. I also think that Android TVs (or Ouya, if you want) could make use of some hefty power, to finally get that console gaming revolution to topple the big three that have been in the cards for quite some time now.
  - slantyyz 12y ago
    >> I also think that Android TVs (or Ouya, if you want) could make use of some hefty power, to finally get that console gaming revolution to topple the big three that have been in the cards for quite some time now. Having been a console guy since the 80s, the PS3 is very likely the last console I'll ever own. The thing that toppled the consoles for me? Steam. All of the games I play on consoles (in particular, 2D/3D fighting games) are now on or are coming to Steam. If I want to play on my TV, all I need to do is hook the TV up to my laptop.
annnnd 12y ago
How sad to see all this work go into performance while it is impossible to get a graphic card that had open source Linux drivers (at least ones that work 100% of the time). And I'm talking about suspend function, not about some advanced games. I mean, how difficult can it be?
- valarauca1 12y ago
  Actually very. Lets go back in time to the modern days of the early to mid 90's. GPU's were very new. Back in that day if you wanted to interface with a GPU you had to know what language it spoke. Every company had their own language, architecture, etc. So if you made software you had to make a different version for each hardware GPU. Case and point 5 version of MechWarrior, 1 version for each GPU company on the market. :.:.: Eventually we got Glide, Mantle, DirectX, and OpenGL. These aimed to standardize our interface to the GPU so that the GPU hardware guys can focus on building blindingly fast vector processors with no regard to architecture backwards compatibility, etc. Want to completely change register layout every 2 years? No problem. Want to redo caching every 18 months? No problem. This has lend to blindingly fast GPU scaling. And the market exploding as GPU's currently lead in GFLOPS/watt. Also allows for average customers to have access to TFLOPS level of computing which in turn have us the currently growing open science and data initiatives. :.:.: Pandora was let out of their box, its going to be very hard to put them back in. DirectX/OpenGL is more or less a Compiler for the byte code that LLVM/GCC generates for 3D acceleration. This byte code is then compiled into native code for the GPU based on the GPU driver for DirectX/OpenGL. This is the problem. Your GPU isn't a device, so much a fully feature complete computer that is programmed on the fly by your CPU.
  - singlow 12y ago
    But what does that have to do with making the suspend features work? I get it that keeping the 3d performance up with open source drivers is hard. But why can't nvidia help out the open source drivers with the basics like sleep/suspend/refresh/resolution etc.
    - deleted 12y ago
zdw 12y ago
The really freaking cool parts of this star on page 4: http://www.anandtech.com/show/8811/nvidia-tegra-x1-preview/4 http://www.anandtech.com/show/8811/nvidia-tegra-x1-preview/4 12 input camera board with 2 GPUs as a platform for automatic driving cars? Sounds awesome, especially with the crazy full tower + multi GPU rigs I've seen in the trunk of cars for this sort of stuff.

55 points by ismavis 12y ago

> X1’s GPU is looking very good out of the gate – at least when tuned for power over performance. Isn't it such a shame Nvidia never seems to do that in mobile, then? Tune it for power, that is. Yes, the GPU and the chip overall are "impressive", but all of Nvidia's mobile chips have been at the time of launch (perhaps with the exception of Tegra 4). I really hate to say this, but I want Nvidia to succeed and it really bothers me that after so many Tegra generations they still don't get it. The problem is Nvidia always over-optimizes for performance, and then the whole chips turns into a dud that nobody wants, whether OEMs or consumers. Instead of doing "2x" performance increase, I would've been happier with a 1.5x performance increase, at a much more manageable and sustainable power envelope for a mobile device. I'm not sure why they keep doing this despite this strategy not working. Doing the same thing over and over again and seeing no results is the definition of insanity after all. Perhaps it's the whole "PC-centric" DNA they have, which makes them think in terms of PCs that have plenty of ways to cool down. I guess this is why "incumbents" keep failing when they get disrupted, even if they adopt the disruption technology relatively early. The old ways of doing things are still ingrained in every decision they make. And what happened to that Icera modem technology they bought? It's been years now but still not seeing it integrated into their chips. Have they abandoned all hope of entering the smartphone market already? They seem to already be refocusing quite aggressively towards the automotive market.

modeless 12y ago
They need 2x performance increases every year to keep up with Apple/Imgtec and Qualcomm. The only reason their performance looks impressive is they announce long before shipping and compare their unobtanium to everyone else's shipping devices. If you compare shipping devices to shipping devices (e.g. the TK1 vs. A8X benchmarks they have in the article) they're not ahead on performance.
- listic 12y ago
  As far as I can see, current gen NVidia tablet chips (K1) are still on top, rivaling even some previous-gen (1.5 years old) x86's in graphics performance. http://www.futuremark.com/hardware/mobile http://www.futuremark.com/hardware/mobile Aren't they?
  - Grazester 12y ago
    The K1 I think was the first chip Nvidia really impressed me with.
xorcist 12y ago
Perhaps there are still places where you should optimize for performance where they still have a fighting chance to make it in the marketplace, such as the self driving car use case they hyped during the press conference. I also think that Android TVs (or Ouya, if you want) could make use of some hefty power, to finally get that console gaming revolution to topple the big three that have been in the cards for quite some time now.
- slantyyz 12y ago
  >> I also think that Android TVs (or Ouya, if you want) could make use of some hefty power, to finally get that console gaming revolution to topple the big three that have been in the cards for quite some time now. Having been a console guy since the 80s, the PS3 is very likely the last console I'll ever own. The thing that toppled the consoles for me? Steam. All of the games I play on consoles (in particular, 2D/3D fighting games) are now on or are coming to Steam. If I want to play on my TV, all I need to do is hook the TV up to my laptop.

annnnd 12y ago

How sad to see all this work go into performance while it is impossible to get a graphic card that had open source Linux drivers (at least ones that work 100% of the time). And I'm talking about suspend function, not about some advanced games. I mean, how difficult can it be?

valarauca1 12y ago
Actually very. Lets go back in time to the modern days of the early to mid 90's. GPU's were very new. Back in that day if you wanted to interface with a GPU you had to know what language it spoke. Every company had their own language, architecture, etc. So if you made software you had to make a different version for each hardware GPU. Case and point 5 version of MechWarrior, 1 version for each GPU company on the market. :.:.: Eventually we got Glide, Mantle, DirectX, and OpenGL. These aimed to standardize our interface to the GPU so that the GPU hardware guys can focus on building blindingly fast vector processors with no regard to architecture backwards compatibility, etc. Want to completely change register layout every 2 years? No problem. Want to redo caching every 18 months? No problem. This has lend to blindingly fast GPU scaling. And the market exploding as GPU's currently lead in GFLOPS/watt. Also allows for average customers to have access to TFLOPS level of computing which in turn have us the currently growing open science and data initiatives. :.:.: Pandora was let out of their box, its going to be very hard to put them back in. DirectX/OpenGL is more or less a Compiler for the byte code that LLVM/GCC generates for 3D acceleration. This byte code is then compiled into native code for the GPU based on the GPU driver for DirectX/OpenGL. This is the problem. Your GPU isn't a device, so much a fully feature complete computer that is programmed on the fly by your CPU.
- singlow 12y ago
  But what does that have to do with making the suspend features work? I get it that keeping the 3d performance up with open source drivers is hard. But why can't nvidia help out the open source drivers with the basics like sleep/suspend/refresh/resolution etc.
  - deleted 12y ago

zdw 12y ago

The really freaking cool parts of this star on page 4: http://www.anandtech.com/show/8811/nvidia-tegra-x1-preview/4 http://www.anandtech.com/show/8811/nvidia-tegra-x1-preview/4 12 input camera board with 2 GPUs as a platform for automatic driving cars? Sounds awesome, especially with the crazy full tower + multi GPU rigs I've seen in the trunk of cars for this sort of stuff.