I’ve wanted to buy an upgrade to my RX580 for years now, but I’d really like AV1 encoding support. With OBS finally supporting AV1 on all platforms (?), this actually makes sense. But I’m once again reminded how bad the used market for GPUs is in my country atm, so I’ll wait for a while longer.
Got a 6700XT second hand about a year ago when the price finally came down from astronomical ridiculous crypto bubble crazy, to almost reasonable. Just looked and they’re still going for the same price. Thought this would have dropped a bit by now, but I guess not.
Yes, I’ve also had an eye on the 6700XT, but I made the bad decision to wait for the new gen and hopefully a price drop for older GPUs. The stable used prices are probably because of people who bought at exorbitant prices who don’t want to sell their GPU for nothing, combined with the new gen having the same price to performance ratio.
Now with the 7600XT having 16GB VRAM, I’ve thought about buying until I noticed it only supports PCIe 4.0 x8, which is half the bandwidth on my PCIe 3.0 x16 slot. It’s a B350 board I want to upgrade to a 5800X3D and use for years to come. This means I’m basically forced to either go with a 7700XT, or go with an older 6700XT.
Anyway, waiting years for a new gen isn’t an option either, so I’ll stay frustrated for a while longer.
Fun coincidence, 16 lanes was one of my concerns as well when I got mine. I’m also on an old AM4 motherboard. Currently have a 3900X CPU which is plenty for my needs for now, but it’s good knowing I still have an upgrade path to an X3D. AM4 has been an awesome platform in terms of upgradability :)
That’s because H.265 is patent encumbered. Firefox doesn’t support H.265 at all and Chrome only supports it if the hardware does. In order to support accepting H.265 input from streamers, Twitch would basically have to pony up the compute resources for full-res realtime transcoding for every H.265 stream to H.264 – either that or put up with a lot of bad press surrounding people not being able to stream at full res anymore.
AV1 would introduce a similar hardware requirement because not everyone even has AV1 Decode, and even fewer have AV1 encode. AV1 encode would only be available on people on gpus using the latest generation, blocking anyone buying previous generation stuff (so no AMD 6000 or older, or Nvidia 3000 or older, and non Intel Arc products).
All (recent) major browsers I’m aware of have software AV1 decode as standard, so the receiving end wouldn’t be a problem apart from higher CPU usage. As for encode, obviously this wouldn’t be universal – just streamers who had the computing power (hardware or software) for realtime AV1 encode would be able to take advantage of that on Twitch.
the browsers have the software, but not the hardware decode step.
software decode, especially for mobile, would be battery draining and no streaming service would realistically would use it without the userbase having hardware decode support.
for pcs, av1 hardware decode is amd 6000 or newer, amd phoenix apus, nvidia 3000 or newer gpus, 11th Gen intel cpus or newer.
for mobile, its only like a small portion of the phones released in the past year and a half or so.
for iphone, the list is the iphone 15 pro max. and for the other devices, things using the M3.
as long as the world is a mobile first mindset, theres no way theyre going to ask evwryone on mobile to take a significant battery loss just for a higher resolution stream.
The application will stream the selected monitor if the mutter screencast portal is available. If it is unavailable, a fallback to X11 based frame grabbing will happen. As such, it should work fine in almost all setups.
Your GPU has a dedicated ASIC that can do the encoding simultaneously. On NVIDIA (not relevant in this case) that would be your NVENC encoder.
AMD and Intel have their own ASIC IP blocks that do encode/decode that’s part of the GPU “SoC” but wouldn’t consume GPU compute resources (eg CUs). That’s how you see people already using GPU encode with obs (non-AV1 codecs) while gaming, and really that’s how people like me using Sunshine/Parsec for the host PC for “remote” gaming (mostly for remoting into a Windows machine for the 1 game that cannot be run on Linux nor a VM due to anti-cheat). The only GPU resources you’re using are PCIe bandwidth and perhaps some VRAM usage? But I wouldn’t call it just dumping it from the CPU to the GPU, you have an ASIC that mitigates the brunt of the workload and AV1 with Sunshine has been amazing, can’t imagine now using it for recording my gameplay vids will hopefully be better than H264 (due to lower bitrates and hence smaller file sizes).
No, this kernel patch will be different to what’s in Windows code. It implements what’s necessary for wine to be more performant, not the actual Windows API itself.
Wine implements those Windows API/ABIs, which is legal because it’s done by reverse-engineering. I believe in some countries (US?) it’s also necessary for the devs to never have seen Windows code.
PS: Google v. Oracle is a US supreme court decision where Oracle lost at trying to patent Java API’s.
I’m glad to see that he is learning in prison, talking and working through things. This really is the point of prisons: not just a place to keep people but a place to reform them.
Anyone of us could become a criminal given the right pressures and circumstances. I wish all prisons would reform and educate their inmates and that they come out as better people who can live a peaceful and productive life.
I don’t live in the US. But I would hope that eventually prisons would adopt the mindset to reform inmates rather than just keep them locked up for nothing.
I haven’t used it yet personally, but I would bet as soon as Debian/Ubuntu LTS/CentOS/openSuse/other stable Linux distros get kernels new enough to it will be not just a btrfs killer, but a ZFS killer too.
The tiered write and read layers and SMR support put ZFS caching to shame.
Nobody uses SMR for live data anyway unless it’s in very particular circumstances.
Bcachefs is still at least a couple of years away from serious use. But sure, if it’s available and you have a good backup strategy you can use it today.
As for “years away” I agree. As my first post said people should wait till you can use bcachefs in the stable distros. Debian isn’t getting kernel 6.7 any time soon 😆. So years away is right in any case.
I think bcachefs addresses the reason why people don’t use SMR HDDs. (Aka changes resulting in cascading writes)
You could have a data pool with the following tiers.
Tier 1: SSDs
Tier 2: HDDs
Tier 3: SMR HDDs
With bcachefs you would only ever write to your tier 1 storage. In the background, as able, bcachefs would offload the data from the faster lower tiers to the slower higher tiers based on frequency of data access.
You would only ever read from the SMR HDDs and would never write to them. They act as a sort of async backing to your data.
Personally I would love a data pool with a few SSDs, backed by a few HDDs, backed by many SMR HDDs. You would save so much money just with good architecting.
Bcachefs should be a ZFS killer. All the features of ZFS with storage tiers being a superior version of ZFS’s L2arc with none of the DKIM kernel license incompatibility nonsense.
Damn, I didn’t think to include SMR drives when it comes to bcachefs. Your whole comment made me appreciate the whole concept under a whole new light actually, thanks!
Yes, which is why it is a little odd for the article author to include it without context, because we all immediately think of one social mistake that has nothing to do with Linux.
He did mention the murder of his wife and said he would detail his regret to anyone who asked. The rest of the letter describes the “social mistakes” in dealing with co-workers and the Linux community. He even asks that those co-workers’ names be added to the credits and his negative comments about them be deleted. There’s no forgiving what he did to his wife but there’s at least some evidence he’s changed since that happened.
He did mention the murder of his wife and said he would detail his regret to anyone who asked.
This is true - I’m reacting more to the title than the content. It’s a very peculiar choice of words.
There’s no forgiving what he did to his wife but there’s at least some evidence he’s changed since that happened.
Perhaps - it’s hard to tell. It still reads a lot like one of his standard narcissistic rants even as he’s complimenting others. It’s still all about his “dream”.
I’m not a doctor but he certainly seems neurodivergent based on his writing. It’s hard to imagine him ever changing in some significant way and being “rehabilitated” enough to be allowed back into society, hence the “some evidence”. It’s might be best he remains in jail rather than be paroled.
Yeah - I mean - I don’t want to get into the business of analyzing somebody’s metal state but he definitely seems to have issues with fixation. But I also don’t want to cross the line into saying that he’s necessarily dangerous because of that. He’s dangerous for other reasons though. I agree with your “some evidence” line in that he does seem to be focusing on the part of his personality that does seem to be the most dangerous - inability to manage conflict. Prison does provide for that conflict - but it also provides many rules and structures that he wouldn’t have on the outside. Dunno. I have a difficult time saying that anybody who has murdered their wife should ever see freedom again at all - “reformed” or not.
HIP is amazing. For everyone saying “nah it can’t be the same, CUDA rulez”, just try it, it works on NVidia GPUs too (there are basically macros and stuff that remap everything to CUDA API calls) so if you code for HIP you’re basically targetting at least two GPU vendors. ROCm is the only framework that allows me to do GPGPU programming in CUDA style on a thin laptop sporting an AMD APU while still enjoying 6 to 8 hours of battery life when I don’t do GPU stuff. With CUDA, in terms of mobility, the only choices you get are a beefy and expensive gaming laptop with a pathetic battery life and heating issues, or a light laptop + SSHing into a server with an NVidia GPU.
The problem with ROCm is that its very unstable and a ton of applications break on it. Darktable only renders half an image on my Radeon 680M laptop. HIP in Blender is also much slower than Optix. We’re still waiting on HIP-RT.
That’s true, but ROCm does get better very quickly. Before last summer it was impossible for me to compile and run HIP code on my laptop, and then after one magic update everything worked. I can’t speak for rendering as that’s not my field, but I’ve done plenty of computational code with HIP and the performance was really good.
But my point was more about coding in HIP, not really about using stuff other people made with HIP. If you write your code with HIP in mind from the start, the results are usually good and you get good intuition about the hardware differences (warps for instance are of size 32 on NVidia but can be 32 or 64 on AMD and that makes a difference if your code makes use of warp intrinsics). If however you just use AMD’s CUDA-to-HIP porting tool, then yeah chances are things won’t work on the first run and you need to refine by hand, starting with all the implicit assumptions you made about how the NVidia hardware works.
How is the situation with ROCm using consumer GPUs for AI/DL and pytorch? Is it usable or should I stick to NVIDIA? I am planning to buy a GPU in the next 2-3 months and so far I am thinking of getting either 7900XTX or the 4070 Ti Super, and wait to see how the reviews and the AMD pricing will progress.
Works out of the box on my laptop (the export below is to force ROCm to accept my APU since it’s not officially supported yet, but the 7900XTX should have official support):
Anything that is still broken or works better on CUDA? It is really hard to get the whole picture on how things are on ROCm as the majority of people are not using it and in the past I did some tests and it wasn’t working well.
Hard to tell as it’s really dependent on your use. I’m mostly writing my own kernels (so, as if you’re doing CUDA basically), and doing “scientific ML” (SciML) stuff that doesn’t need anything beyond doing backprop on stuff with matrix multiplications and elementwise nonlinearities and some convolutions, and so far everything works. If you want some specific simple examples from computer vision: ResNet18 and VGG19 work fine.
I think end-to-end refers to the “open source”, not the GPU acceleration. I know GPUs have always been a black magic to get working and so you often have to use proprietary, closed-source blobs from the manufacturer to get them to work.
The revolution that this is bringing seems to be that all that black magic has been able to be implemented in open-source software.
Could be wrong though, that’s just how I interpreted the article.
Yup, it’s definitely about the “open-source” part. That’s in contrast with Nvidia’s ecosystem: CUDA and the drivers are proprietary, and the drivers’ EULA prohibit you from using your gaming GPU for datacenter uses.
If it means I won’t have to do a ritual dance under the full moon, facing towards finland, just to get it installed correctly, I welcome my new gentleman overlords.
I never understood why AMD themselves don’t work in integration in Debian and Fedora. That way Ubuntu and RHEL would automatically inherit it. At worst it would be in Universe/EPEL.
phoronix.com
Active