they're not a userspace construct in any meaningful way. libutils/Thread (capital-T Thread) is a thin wrapper around pthread_t, and most Android code uses pthread_t directly. what this hit is probably a bug related to…
GPUs don't have a page fault handler; when there's a page fault, it's an unrecoverable crash. Accordingly, zero-on-allocate (or potentially zero-on-free, but that makes assumptions about startup and teardown that may…
X1 is actually the 64-bit ARM CPU configuration (Cortex A53 + Cortex A57), not Denver. K1, the predecessor of X1, comes in two flavors: 32-bit 4xCortex A15 and 64-bit 2xDenver. TK1 is 32-bit, Nexus 9 is 64-bit (and is…
property taxes are capped in CA, so the landlord's costs are essentially fixed: http://en.wikipedia.org/wiki/California_Proposition_13_%2819...
I actually read IJ on a Kindle and found it significantly easier for the most part than reading it in print because of links to endnotes, which removed the requirement to keep two sets of bookmarks (if you haven't read…
GPUs don't support precise exceptions. For example, you can't take a GPU program that contains a segfault, run it as a standard program (as in, not in a debug mode), and be presented with the exact instruction that…
https://gcc.gnu.org/onlinedocs/gcc-4.7.4/gcc/Function-Attrib... and http://pasky.or.cz/dev/glibc/ifunc.c look up the target attribute and the ifunc attribute--it's basically a way to compile multiple versions of a…
but that's GLES 2.0, which is significantly less flexible than the kinds of GPUs we're discussing here and is not even in the same ballpark as a CPU (and almost certainly significantly less strict in terms of floating…
disclaimer: I work in this space and have done so for a while, including previously on CUDA and on Titan. GPUs for general purpose computation were never 100x faster than CPUs like people claimed in 2008 or so. They're…
Kayvon teaches at Carnegie Mellon now and his class slides are definitely worth reading: http://graphics.cs.cmu.edu/courses/15869/fall2014/ http://15418.courses.cs.cmu.edu/spring2014/
sorry, the desktop Intel SIMD ISAs are generally not something I use on a day-to-day basis, so I get them mixed up a lot. (LRBni was the last one I looked at for any serious length of time.)
I spent a lot of the past month improving the RenderScript (Android data-parallel compute, using C99 plus vectors) codegen to better work with the LLVM vectorizers, so I have a fair amount of experience with this exact…
>> 20 years of legacy, needs a reboot and major simplification pass This was attempted. It failed: http://en.wikipedia.org/wiki/OpenGL#Longs_Peak_and_OpenGL_3....
B3D PM is fine.
I implemented scatter back in October, but it just barely missed Android 4.2. It's in the next release.
Desktop: high-end consumer GPUs have about 10-15x the single-precision FLOPs and 4-6x the bandwidth of a single Intel CPU socket. At this point, usually connected via PCIe Gen3. There are two real vendors (NVIDIA and…
Pretty much exactly this. Performance portability is our main concern, and we are willing to trade off some peak performance to get it because of how badly you will hurt yourself on different architectures. We are…
I worked on CUDA at NVIDIA for over four years and was the primary API designer for a large part of that time. I started on RS at Google in September. Basically, he gives us too little credit for the execution model…
I think it's a political move more than a technical one. WinRT lets MS say to Intel that Intel has to care about the tablet market and low-power processors or otherwise ARM will get a foothold in Windows (giving them…
They're blade systems, there's not really room for disks. Easier to keep it centralized and easily serviced.
(full disclosure: used to work for NV on CUDA and did very extensive work on Titan, so I am probably biased) If you think your existing MPI app is going to automatically scale to a heterogeneous architecture (high-power…
Tesla boards haven't shipped in a separate 1U form factor for a few years; they're all passively-cooled PCIe boards inside a x86 server chassis now.
I don't think that is true. If you're writing C, it's probably because you want to be close to the metal and want many of the guarantees that such proximity provides. If you're focused on that sort of thing, then it's…
The lack of any sort of physical connection to the NVIDIA GPU's display outputs is the fundamental feature of Optimus. Switchable graphics existed for years before Optimus introduced (and was usually usable under Linux…
You basically missed the entire point of it or why it's interesting. Current Intel CPUs have a very small GPU built on to the die of the CPU. By buying the CPU, you're paying for an Intel GPU anyway. At the same time,…
they're not a userspace construct in any meaningful way. libutils/Thread (capital-T Thread) is a thin wrapper around pthread_t, and most Android code uses pthread_t directly. what this hit is probably a bug related to…
GPUs don't have a page fault handler; when there's a page fault, it's an unrecoverable crash. Accordingly, zero-on-allocate (or potentially zero-on-free, but that makes assumptions about startup and teardown that may…
X1 is actually the 64-bit ARM CPU configuration (Cortex A53 + Cortex A57), not Denver. K1, the predecessor of X1, comes in two flavors: 32-bit 4xCortex A15 and 64-bit 2xDenver. TK1 is 32-bit, Nexus 9 is 64-bit (and is…
property taxes are capped in CA, so the landlord's costs are essentially fixed: http://en.wikipedia.org/wiki/California_Proposition_13_%2819...
I actually read IJ on a Kindle and found it significantly easier for the most part than reading it in print because of links to endnotes, which removed the requirement to keep two sets of bookmarks (if you haven't read…
GPUs don't support precise exceptions. For example, you can't take a GPU program that contains a segfault, run it as a standard program (as in, not in a debug mode), and be presented with the exact instruction that…
https://gcc.gnu.org/onlinedocs/gcc-4.7.4/gcc/Function-Attrib... and http://pasky.or.cz/dev/glibc/ifunc.c look up the target attribute and the ifunc attribute--it's basically a way to compile multiple versions of a…
but that's GLES 2.0, which is significantly less flexible than the kinds of GPUs we're discussing here and is not even in the same ballpark as a CPU (and almost certainly significantly less strict in terms of floating…
disclaimer: I work in this space and have done so for a while, including previously on CUDA and on Titan. GPUs for general purpose computation were never 100x faster than CPUs like people claimed in 2008 or so. They're…
Kayvon teaches at Carnegie Mellon now and his class slides are definitely worth reading: http://graphics.cs.cmu.edu/courses/15869/fall2014/ http://15418.courses.cs.cmu.edu/spring2014/
sorry, the desktop Intel SIMD ISAs are generally not something I use on a day-to-day basis, so I get them mixed up a lot. (LRBni was the last one I looked at for any serious length of time.)
I spent a lot of the past month improving the RenderScript (Android data-parallel compute, using C99 plus vectors) codegen to better work with the LLVM vectorizers, so I have a fair amount of experience with this exact…
>> 20 years of legacy, needs a reboot and major simplification pass This was attempted. It failed: http://en.wikipedia.org/wiki/OpenGL#Longs_Peak_and_OpenGL_3....
B3D PM is fine.
I implemented scatter back in October, but it just barely missed Android 4.2. It's in the next release.
Desktop: high-end consumer GPUs have about 10-15x the single-precision FLOPs and 4-6x the bandwidth of a single Intel CPU socket. At this point, usually connected via PCIe Gen3. There are two real vendors (NVIDIA and…
Pretty much exactly this. Performance portability is our main concern, and we are willing to trade off some peak performance to get it because of how badly you will hurt yourself on different architectures. We are…
I worked on CUDA at NVIDIA for over four years and was the primary API designer for a large part of that time. I started on RS at Google in September. Basically, he gives us too little credit for the execution model…
I think it's a political move more than a technical one. WinRT lets MS say to Intel that Intel has to care about the tablet market and low-power processors or otherwise ARM will get a foothold in Windows (giving them…
They're blade systems, there's not really room for disks. Easier to keep it centralized and easily serviced.
(full disclosure: used to work for NV on CUDA and did very extensive work on Titan, so I am probably biased) If you think your existing MPI app is going to automatically scale to a heterogeneous architecture (high-power…
Tesla boards haven't shipped in a separate 1U form factor for a few years; they're all passively-cooled PCIe boards inside a x86 server chassis now.
I don't think that is true. If you're writing C, it's probably because you want to be close to the metal and want many of the guarantees that such proximity provides. If you're focused on that sort of thing, then it's…
The lack of any sort of physical connection to the NVIDIA GPU's display outputs is the fundamental feature of Optimus. Switchable graphics existed for years before Optimus introduced (and was usually usable under Linux…
You basically missed the entire point of it or why it's interesting. Current Intel CPUs have a very small GPU built on to the die of the CPU. By buying the CPU, you're paying for an Intel GPU anyway. At the same time,…