my123

74 ms·

901.
by my123 6y ago
cudaStreamCreateWithPriority is what you'd like to use for creating a stream with a specifically defined priority. The GPU is able to run up to 128 kernels in parallel on modern NVIDIA GPUs, concurrently.
902.
by my123 6y ago
AMD doesn't even have the performance crown, by a long shot. NVIDIA did hold that pretty much non-stop since... Maxwell 2? As of no difference in compute workloads, I can put Blender as an example: https://techgage.com/
903.
by my123 6y ago
You have async compute on NVIDIA GPUs. See this slide deck about it: https://developer.download.nvidia.com/CUDA/training/StreamsA... (for the implementation in Fermi, it's even more flexible since then)
904.
by my123 6y ago
It actually swaps much more than the older machine that I have, at 3% wear already. And the indicator isn't linear. (at 30TB written currently, 250TB and the machine risks failure when estimating from a machine which actually failed fr
905.
by my123 6y ago
Nowadays, modern Nvidia hardware (Volta onwards) handles compute workloads better, notably through independent thread scheduling. RDNA2 with its cache and low main memory bandwidth didn't help either, because while the cache is there,
906.
by my123 6y ago
Lol? ROCm is a mess. It's awful. And it is not even supported on RDNA and RDNA2. They even dropped Polaris support last month so that it's only supported on Vega. And that's very much not a high quality GPU compute stack, s
907.
by my123 6y ago
Were. AMD totally squandered their advantage that they had with launching GCN early through an absolutely awful software stack.
908.
by my123 6y ago
Which is something that CUDA provided since the very beginning. (with PTX)
909.
by my123 6y ago
At least some years ago, you could build GCC on MSVC just fine. (Did it for GCC 3.3 and 4.2 on Interix way back then) It should probably be still functional, or easily fixed if not.
910.
mrustc: In-progress alternative Rust compiler (to C)(github.com)
80 points by my123 6y ago|62 comments
911.
by my123 6y ago
Yes, the benefit of securely implementing hibernation with Secure Boot on for Linux is a direct benefit for Linux users, no impact one way or another for those using Windows.
912.
by my123 6y ago
The HoneyComb is at PCIe 3.0. Sadly it's at ARMv8.0 (w/ Cortex-A72) so that removes it from consideration totally for me. And the Xavier has an official-but-still-beta EDK2 firmware available too, which includes ACPI support and c
913.
by my123 6y ago
Quite some recent controllers outright don't implement it, and newer devices don't require it...
914.
by my123 6y ago
> there an ARM motherboard out there with a 16x lane bus? For servers at sky-high prices only. For $699, your closest choice might be the NVIDIA Jetson AGX Xavier... PCIe 4.0 but with a x8 slot, and an NVMe M.2 x4 PCIe 4.0 slot too.
915.
by my123 6y ago
No. BitLocker when it uses the TPM uses verified boot. PCRs are filled during the boot process, recording the boot state. If the boot state does not match, the TPM will not release the key.
916.
by my123 6y ago
PCRs of the TPM. Combined with full drive encryption, the bootloader not matching will make the TPM not release the drive encryption key.
917.
by my123 6y ago
Without drive encryption, you cannot ensure that you have a secure system.
918.
by my123 6y ago
Because HIP/ROCm is awful. Instead of targeting an IR, it directly targets a given GPU's ISA, so that your existing binary will not run on future hardware. That's a total no-go for basically every non-HPC use case. Intel is m
919.
by my123 6y ago
My opinion about the subject: AMD actually does not care about GPU compute on customer hardware. If they did they would not have: - Have had support for Linux only, with no ROCm on Windows - Have made ROCm as a GPU-specific targeted proce
920.
by my123 6y ago
> IYO, is there any value to using legacy nvidia-driver-390 over X.Org? nouveau doesn't support compute at all on those... (as in OpenCL). Also, as far as I remember, rechecking work on Fermi didn't seem to be implemented the l
921.
by my123 6y ago
Your card is based on Fermi, an architecture that's no longer supported by NVIDIA since a long while now (last drivers in 2018, you're on a legacy branch). That means that apps using newer versions of CUDA will not work, and your
922.
by my123 6y ago
Apple M1 big core cache sizes: 256KB L1I/128KB L1D Little cores: 128KB L1I/64KB L1D
923.
by my123 6y ago
ProRender render quality was sadly sometimes pretty close to garbage. (I heard that ProRender 2.0 changed that, but didn't get the chance to test it yet) https://www.phoronix.com/scan.php?page=article&item=blender-.
924.
by my123 6y ago
It’s actually a bit lower level than KVM.
925.
by my123 6y ago
At zero interest at that.
926.
by my123 6y ago
The mining GPUs that they announced aren’t based on Ampere but on Turing AFAIK from the hash rate numbers that they gave. Not even the same manufacturing process...
927.
by my123 6y ago
Same thing for ECC being disabled on customer GPUs. And ECC disabled on customer Intel CPUs and AMD Ryzen APUs. (AMD is quite sneaky there, non-Pro CPUs have ECC but non-Pro APUs don’t). Market segmentation is a usual thing on the computer
928.
by my123 6y ago
It does not reduce the security level to at or below the one of any regular PC. The bootrom bug requires DFU and physical access to be triggered, which is already game over on most systems. Apple also doesn’t solely rely on measured boot fo
929.
by my123 6y ago
I can tell you that devices with Apple CPUs _do_ have multiple NVMe namespaces. The recovery OS and 1TR are not data needed to be saved permanently (as in serial number and such, without which you cannot restore the device), you can recover
930.
by my123 6y ago
As long as you don't shoot yourself in the foot by touching other NVMe namespaces (and even that should be recoverable) and the SPI flash onboard, things can be restorable through DFU.

More ›