Sorry guys, but it seems I made a mistake and ROCm is coming to LINUX, I hope it does come to windows soon as AMD needs some ground in this department! Cheers!
ROCm is already on Linux, but it doesn't officially support consumer GPUs (but it is already possible to 'trick' that and get it working, even on RDNA2 GPUs like RX 6600).
Formal support for RDNA 3-based GPUs on Linux is planned to begin rolling out this fall, starting with the 48GB Radeon PRO W7900 and the 24GB Radeon RX 7900 XTX, with additional cards and expanded capabilities to be released over time." - well, it's "consumer" xD.
Been wanting to learn Linux for a while. So now I can mind as well learn to install dual OS now on my laptop before deciding on a new workstation rig when they release these new RocM drivers
We will see in the future. In the present day AMD makes great hardware but it lacks in the software department, not so much these days, but in the past were terrible. I hope AMD will embrace for collaboration with soft devs and make the drivers more efficient like Nvidia has.
At the moment no. I have my doubts that ROCm will be as efficient as CUDA. It'll simply just be "good enough" for the gamers that like to dabble in content creation, but people that are serious with content creation and other professional applications will choose nVidia. Why would you choose AMD to emulate CUDA when you can just get nVidia and run them natively? We'll see how this develops in 5 to 10 years from now.
Depends but whatever AMD does some people won't migrate due to Brand Loyalty. I know some peopel who are adamant they will only use NVIDIA GPUs because they feel like AMD is slow (even though it's not even true in the slightest).
This will never happen unfortunately. These 2 companies are partnering together this year hand in hand to extract every dollar out of the consumer. If they both become equals, prices will become equals or barely different.
@@IslamGhonaym I doubt it. Mainly because, the purpose of both companies is to make profits... they will need to make their options financially attractive to consumers (especially with inflation, etc.). AMD is already charging less money than NV, but yes, right now, you can't exactly use their GPU's for accellerating workloads via the GPU on Windows, however, once this changes, AMD can easily continue charging less money than NV and people will flock to them because they will have a lot more options.
I think they know it. It seems it is more about having a complete AI suite. The advantage is that I'll be able to train light language processing models (rather than using the CPU) on my main rig and that's bloody cool. The convenience for developpement is nice to have for a data scientist using a laptop to process large datasets (they all do lol). AMD wants to gain shares in this market
Awesome news! I really love AMD now. Ever since I got my 6700xt I've been team AMD. I just discovered something that has increased my productivity in Ableton 11 (music software). AMD virtual super resolution! I can set my 1080p monitor to 4k! It's only good for this purpose though, the picture quality badly suffers but it doesn't matter if I'm just working on music.
Sadly, I have little faith in ROCm. I would hope they will fix it, but the documentation is chaos. Coding with even the AMD workstation GPUs that support it, is nightmarish. Difficult compatibility, bugs, the support is not there. And we really need a competitor in the GPGPU and computational market
AMD's marketing department should pay attention to this kind of coments. It's not the first time I read people complaining about lack of support and documentatión for devs
if they implement this appropriately this could be a huge win for AMD. I'm sure it still won't be quite as fast as true cuda cores, but if they could hit a solid 80 to 90 percent of cuda speeds with less expensive cards they could pull many professionals over. I worked in the enterprise side of things and nvidia has been pissing their partners off like crazy basically since Volta and Pascal
Amd hip is not translation layer. Developer needs to convert their cuda codes into hip codes then recompile. However the conversion is said to be easy as replacing cuda... to hip...
@lolthekidison It's why nvidia cards sometimes run better in some games and worse in some games to amd competitors. Nvidia has a vram bottleneck, and amd has a software bottleneck. It's that tradeoff sometimes people will make but I prefer better hardware so I switched from rtx to AMD. Amd better start impressing me though, nvidia broadcast & canvas has made it tempting.
@@PsychoBenches Personally I like the AMD software and overlay. I love my 6600xt and so do both of my friends that have 6650xts. I have a bunch of gpus and don't really have a preference of either or.
Unless you are making movies in blender, there is no advantage for nvidia in blender. The blender benchmark is only indicative of one very specific task, for modeling or most work you will run out of ram even with 64gb before you run into any performance limitations with even a 6800xt
YES Wonderful I just asked some Guys on a Stable Diffusion Channel and they told me that ROCm was just available for Linux. Great that the confirmation is out that that is going to change :) I can pass the time until Fall easily without SD. And maybe it even works good enough without ROCm ^^ Sooooo AMD just needs to lower the price a little more to count me on Board. Thanks for bringing the good News!
The architecture focus is gaming workload not so much about compute. The reason RDNA exist is to keep out compute stuff from "bloating" their consumer card (which exist mainly for gaming and rendering 3d graphic) like what happen with GCN. in fact polaris and vega already have ROCm support in the past. Then AMD decided to stop pushing ROCm for consumer card starting with RDNA.
As long as the support is only available in Linux, this is only a side note and nothing more. - the official ROCm Support for AMD cards is weak at least. E.g. While an Rx6800xt works with ROCm out of the box, it's not officially supported. There are even cards you have to "fake" as different supported cards to get it running. We must expect the same for most RDNA3 cards. - Even if everything is fine, Nvidia cards still smoke AMD in that area. It's the same like in raytracing. Maybe it changes a bit with RDNA4, but Nvidia is years ahead in that area. We still have to face that sad truth. ROCm makes the situation for AMD fans a bit better. But not for professional users.
I'm looking at the products Nvidia has been turning out lately and it's almost like they think of their Video Cards as an afterthought since their AI is cruising to new heights. The 40X series is a great example. Why would any company create a modern card with a 128 bit bus? AMD though, seems to be all in with their cards. They don't shortchange the Vram and they have forgotten that a 128 bit bus ever existed. If ROCm turns out to be conductive to great gaming, I'll be switching to an AMD CPU running an AMD GPU. Lets see how this plays out ....
ROCm has nothing to do with gaming. You said nvidia is going with AI while giving gpu as a thing go render game as an after thought, AMD pushing ROCm on RDNA is going the same route as nvidia with AI. the reason AMD make CDNA for compute and RDNA for gaming is so they don't have to push things like AI on consumer cards. In the past Polaris and Vega were already have ROCm support. Then with RDNA AMD completely stop supporting ROCm on consumer card. Now they decided to once again give consumer card ROCm support with RDNA 3. if anything AMD realize that they cannot ignore the potential of generating more income by offering AI support on consumer card rather than make it exclusive to CDNA.
preview driver has MPO issues again during overlay transitions such as steam or AMD overlay toast notification for example. MPO is what makes low power consumption possible on idle
@@AncientGameplays try playing more games its a glitch that shows it self as you start game during the steam toast popup from steam overlay or radeon toast may not show it self every game.
I can run pytorch on a rx6600xt. It uses rocm/hip to use the cuda code... So it's already working with rdna2 - how well supported it is might be another story.
Would you do a follow up on this as it would apply to Linux, Steamdeck, Proton and Vulkan? This seems more relevent as Linux is fast catching Windows as a gaming OS and Windows 11 has actually shed overall downloads compared to Windows 10. Probably due to the requirement for a MS Account to register and complete the download. I know there are instances where the games are significantly better and others where it is significantly worse on Linux, but that's with RDNA 2. I'm not sure if ROCm means having to use Pop or nVidia drivers to gain the greater performance. Any thoughts?
This could be a game changer.....I have actively avoided AMD cards because of the lack of CUDA......I'll be watching this very closely. Thank you for this amazing info :)
If it can run the same CUDA code, with minimal changes, it will be a game changer. A lot of applications can benefit from GPU acceleration. AMD cards are perfectly capable of "GPGPU computing" (GP - general purpose), as evidenced by the tons of game benchmarks coming out each day. OpenCL exists, but it doesn't have the same adoption rate as CUDA, probably because AMD's market share is in the single-digits and Intel is eating THEIR lunch not Nvidia's. And AMD seems to be okay with that, instead of just lowering their prices. 😕 From what I understand the code will still need to be recompiled, just with different headers than "vanilla" CUDA. Same code (mostly), just different headers.
AMD's CUDA alternative is called "HIP" and it can compile down to both CUDA and AMD native instructions. They also have a tool called "HIPify" that automatically converts most CUDA programs to HIP
I went AMD because Nvidia are just price gouging. especially in Australia. 7900xtx was $1500 but 4080 was $2000 and up. i get the shits when i look at prices overseas
Intel arc кстати сейчас отлично соревнуется с CUDA от NVIDIA. До выхода этого видео было только две видеокарты, которые я бы взял, но если AMD всё хорошо сделает со своим HIP и ROCm то я возьму на заметку их видеокарты
Blender? There are now 5 API's to accelerate it's render engine. OneAPI, AMD HIP, Optix, Cuda and metal all targeting different hardware and / or platforms. Imagine if this was also needed for just the 3D viewport. OpenGL for AMD older cards NvidiaGL for Nvidia Direct3D for Intel Metal for Apple Vulkan for newer AMD cards GPGPU is just a mess mostly because of Nvidia's dominant cuda but also because of openCL's failure. When blender stopped support for opencl for having slow performance and "unfixable" bugs, no one had any idea with what it would be replaced. Many including me thought it wouldn't be replaced at all. Or that opencl (with new version) would be re-enabled. Well we got something new. And yesterday blender released version 3.6 long term support. LTS versions are more focused on stability over new features. And we got AMD accelerated rendering, but still broken and not giving proper results in certain cases and is still much slower than comparable Nvidia cards. It was the saddest thing to hear in the official blender 3.6 LTS features video. It's here, but still broken. The irony, Dropping opencl for being buggy and slow, developing HIP in it's place to fix that and after all that time we now have buggy and slow HIP. Khronos group was blamed for opencl's failure. Who gonna be blamed for this now?
I just ordered an asrock Taichi 7900 xtx yesterday, but I dont care for these features at all, since I dont think I will ever need them. But its good that AMD becomes a bigger competition with this now. That always means more and better offers for us customers.
You can already use AMD CPUs for stable diffusion using a special fork of stable diffusion thats uses DirectML, I have stable diffusion running on a humble Vega 56.
Nothing new really. Initially AMD create Stream to compete directly with CUDA. then then they move to OpenCL. Now OpenCL also did not work well for them they have ROCm. And they use nvidia CUDA to make things work for ROCm.
I hope it changes! AMD are much more generous with Vram compared to nvidia. Combine lots of vram with hardware acceleration for professionals and it’s best of the best. Nvidia won’t have a leg to stand on. Even 24gb is getting tiredsome for creators. You don’t need 24gb what’s so ever for gaming, but it’s much more required for professional creators. We’ve had it since all the way back to the RTX Titan. Hopefully AMD continue to grow competitively and nvidia do the same.
Yes. And VRAM is also really important for running AI models on a PC. For example I love Stable DIffusion. And with my RTX 2080 8GB it's nice but If I am looking to upgrade I just want much more VRAM and the price of even the new Super cards are crazy. Max 700 Dollars for the top tier card is what I would say OK. But even the 4080 is 800 MSRP dollars then AIBs put a markup on it and because I live in Europe those cards will cost even more here...
Everything AMD makes is free for all... Everything Nvidia makes is strictly theirs and no one else can adopt to come close to them... IDK who is an unfair competitor...
From now until the foreseen future it's all about AI. AI-generated art, videos, language models/chatbots, etc. Most 3D apps are optimized for CUDA as well. Hopefully AMD's GPUs can catch up. They have the cheapest 24GB of VRAM option in the 790XTX. However, once bitten twice shy - I'm sticking with team green because my livelihood is dependent upon AI.
please don't listen to the creator of this video. ROCm is available for older GPUs for a long time already. Their post was only that they'll add support to the RDNA 3. I've used ROCm for generating images using stable diffusion and AI accelerated translation on my 6900XT running on Ubuntu for a while. The only problem with ROCm that I see it's that it's linux only and can be difficult to setup
To my knowledge some older ROCm version did work on polaris and vega. But the new ROCm most likely on newer hardware only. Heck AMD even already cease to support newer ROCm on their MI100 Instinct.
One of my professors at the university has worked in AMD. He used to write GPU manuals for developers and optimize the efficiency. I hope AMD is able to provide a solid competition to CUDA. Thats the only keeping me from buying an AMD GPU even tho they are the cheaper option.
The cuda acceleration in davinci resolve is gooood , so i will be keeping an eye on this. I am interested that my next gpu is an amd one. I used to have amd, i switched to nvidia and going back to amd :) Like a back and forth
I have a question and need help AMD god! This is random I know but Ihave been told that using MSI afterburner with radeon software is not advised but I don't overclock. I am simply trying to make a custom fan curve. If only making a custom fan curve for the gpu, will afterburner interfere with adrenaline? Again, no overclock and no custom fan curve using adrenaline. What software do you personally use to make a fan curve for your gpu?
My pc runs on Linux, I do a lot of server and virtualization workloads. I have a 6900xt and use it for gaming. I also use it as a Plex server. I have Plex Pass, however, Plex does not support AMD hardware at all. That means my GPU that is more than capable of transcoding sits idle, and my CPU gets crushed. I can only manage 2 4k transcoded streams on my 5900x. Its gotten to the point that I have debated getting a second Nvidia GPU just for transcoding. This is amazing news!!!
I wish youtubers would post on every video what accent they're speaking with because I end up trying to identify it instead of listening to the video. Halfway through I know it's Portuguese so now I can listen to what he's saying instead of listening to how he's saying it 🙂
ROCm sound's great but why didn't they even use their chiplet & wafer stacking technology, with their recent 7900XTX / XT card's AND then gave us a card with a 128bit bus (RX 7600)?
I just received a Sapphire 7900XTX Nitro+ for gaming only. Regardless of not needing ROCm, I’m still glad to have it. “If it’s free, it’s for me” as the saying goes.
Oh hell NO! The HIP-CUDA remapping may cause performance penalties on Radeon GPU performance, I would prefer major software like blender and DaVinci resolves will work on Radeon LLVM like ONNX does, to get the maximum performance of Radeon GPU. Software Devs should work on "ROCm-LLVM" and not on HIP-CUDA.
@@fuku474 AMD to enable ROCm on Windows, add support for some gaming Radeon GPUs issued by Videocardz website mentions ROCm 5.6.0 Alpha support table Radeon RX 6900 XT, RX 6600 and R9 Fury are listed...I think AMD first offers ROCm update to newly lauched rx 7900xtx and Radeon pro W7900...later i hope rdna 2 gpus will also get this update....
the thing I think AMD didn't realize is a lot of the things that push the industry forward are made by passionate professionals, on their home machines while not at work. The windows functionality to double click on a zip file to enter it like any other folder wasn't made by microsoft, it was BOUGHT by microsoft. Even the things that a company does initially release can typically be traced back to some programmer wanting it to happen on their own and then the company realizes afterwards that it's a really good idea. It's basically the selfish altruism effect; if you give people the freedom to make their own things, chances are you'll end up profiting from them in the process. (hell don't some companies like Google even have mandatory paid time allotments to work on personal projects during work-hours for this exact reason?) Nvidia putting CUDA on their consumer line of GPUs meant that tons and TONS of hobbyists could work with their Nvidia GPU, but couldn't work with their AMD GPU. If the choice is between two roughly comperable options but one lets you do something the other doesn't, that one's going to get the adoption. By AMD only putting the tool on their cards designed specifically for that purpose they nuked any chance of a community of innovation developing around it, so more and more projects used CUDA, and most people didn't even know AMD had a competitor to it at all.
@@AncientGameplays AI is basically FP32 and matrix operations and can be run on any modern gpu and CPU, just not as fast as a dedicated hardware instruction set
Hey man, quick thanks for all the videos and tips. But returned my 7900 XTX and got a 4090 - Warzone Caldera and WZ2 and Forza 5 was just too stuttery for me.
That makes absolutely no sense, and I am sure you had some issues there (maybe even a defective card) as Warzone and Forza run much better on the AMD side....
@@AncientGameplays No denying the games had good fps, but textures on high on WZ would cause bad stutter. Changing to low helped. Forza could not get rid of the stutters every 15 seconds or so
I would buy AMD if they did support Linux, and if I could use it like I do my NVIDIA card, I mean I spent 1 year trying to get accelerated ray tracing to work on my AMD graphics card after all that frustration I switched to NVIDIA, and I have everything I want, I have game ray tracing, I have real time ray tracing in blender, I can do a lot of stuff in blender that I cant do with AMD because of lack of support from AMD. The only way I will ever switch to AMD GPU again is if they do what NVIDIA do.
It's coming to 'selected' GPUs ~ that doesn't mean it's coming to gamers. Or home user geeks. Or to all Radian GPUs.... It is a positive move and I do applaud (took 'em long enough) but you don't unroll a ten ~ 20 year head start with an announcement... Correct me if I'm wrong, but any CUDA code will run on any nVidia hardware ~ although it will run much better on modern high-end hardware. So are we going to see all CUDA code run run on all RDNA and newer hardware? When are we going to see that? Is this also going to be true in Linux? I do applaud the move ~ took you long enough ~ but an announcement is not the same as delivered working product. Absolutely critical ~ same as AMD64 being the exact same as Intel64 ~ that CUDA code should run on AMD precisely the same way it does on CUDA. You have to be able to just take all the CUDA code that works, and plug it in without touching it. If it runs properly on nVidia, then it will run properly on ROCm.
"same as AMD64 being the exact same as Intel64 " it's the inverse, actually it's called AMD64 in windows and drivers for a reason, the first 64bit CPU on X86 was the "athlon 64" not the "pentium" of the same era !
@@barbatocedric4246 ~ Yes, perfectly correct. People asked Intel for a 64 bit version, same as we had gone from a 16 bit to a 32, but Intel said 'that's too hard' ~ so AMD developed it. Intel was peddling their own version of 64 bit, called Itanium ~ which was not backwards-compatible. But the idea is all those 'intel' systems are compatible and consistent. Compare to a sparc chip or Arm chip or some other completely different architecture. You can get hold of a program which is 5 years old, and it will run. 10 years, 20 years ... As long as the operating system will allow it, you can run The Dungeons of Umoria in all its 1986 glory without a dosbox or a virtual machine. That game is now almost 40 years old, but I still play it. In terms of providing a software environment for the Enterprise, this is a double edged sword. It means old code still works, but that can & does lock you into supporting 50 year old gold plated virgins who nobody understands any longer, and the original author died the day mankind kicked the moon, murdered by Ritchie & Kernigan. But the whole enterprise depends on that little piece of code running ~ Goldman & Sax discovered a way of controlling the futures market in 1973, with this 50 lines of assembly code ~ and no matter what, all future systems must be able to run an instance of this piece of Fortran or else! Nobody alive today has ever seen the source-code for it... What it does is magic ~ it turns 55 year old men in a boardroom, into willing slaves and cum-buckets, and they'll sign anything, pay anything, suck anything, and agree to anything, as long as that piece of code continues to run.
Follow-up 18 hours later. th-cam.com/video/LKxWTYyplow/w-d-xo.html It will be on Linux, not Windows, and it will support two SKUs ~ one being a high end Pro SKU, the other being a top end gamer SKU. I stand by what I said ~ it is a move in the right direction, but a very small one. They say they intend to grow and expand it, but so far, it's like the cherry on top. That's decoration ~ that's not a course, let alone a meal.
Sorry guys, but it seems I made a mistake and ROCm is coming to LINUX, I hope it does come to windows soon as AMD needs some ground in this department! Cheers!
ROCm is already on Linux, but it doesn't officially support consumer GPUs (but it is already possible to 'trick' that and get it working, even on RDNA2 GPUs like RX 6600).
Formal support for RDNA 3-based GPUs on Linux is planned to begin rolling out this fall, starting with the 48GB Radeon PRO W7900 and the 24GB Radeon RX 7900 XTX, with additional cards and expanded capabilities to be released over time." - well, it's "consumer" xD.
It will be a gamechanger if and when it comes to Windows. At this point, if you're doing anything with AI, AMD is a no go!
as linux fanatic i knew :D but thanks for clearing that to others
Been wanting to learn Linux for a while. So now I can mind as well learn to install dual OS now on my laptop before deciding on a new workstation rig when they release these new RocM drivers
Since AMD GPUs are already VERY good at encoding and decoding, do you think ROCm will make some people migrate to AMD GPUs?
I dunno. It's too early in the video to say.
We will see in the future. In the present day AMD makes great hardware but it lacks in the software department, not so much these days, but in the past were terrible.
I hope AMD will embrace for collaboration with soft devs and make the drivers more efficient like Nvidia has.
At the moment no. I have my doubts that ROCm will be as efficient as CUDA. It'll simply just be "good enough" for the gamers that like to dabble in content creation, but people that are serious with content creation and other professional applications will choose nVidia. Why would you choose AMD to emulate CUDA when you can just get nVidia and run them natively? We'll see how this develops in 5 to 10 years from now.
Depends but whatever AMD does some people won't migrate due to Brand Loyalty. I know some peopel who are adamant they will only use NVIDIA GPUs because they feel like AMD is slow (even though it's not even true in the slightest).
If AMD will provide better value than competition, sure they will. Enterprise always wants the best offer possible.
When they balance each other out, the real competition will start and even more features for customers will come.
This will never happen unfortunately. These 2 companies are partnering together this year hand in hand to extract every dollar out of the consumer. If they both become equals, prices will become equals or barely different.
@@IslamGhonaym I doubt it.
Mainly because, the purpose of both companies is to make profits... they will need to make their options financially attractive to consumers (especially with inflation, etc.).
AMD is already charging less money than NV, but yes, right now, you can't exactly use their GPU's for accellerating workloads via the GPU on Windows, however, once this changes, AMD can easily continue charging less money than NV and people will flock to them because they will have a lot more options.
Any improvement we can get on the AMD side is a bonus and will give them more market share. Thanks for the update.
Thanks for watching!
This is a great news. Cuda is the only reason I'm legit forced to buy Nvidia GPU for so long for my 3D work.
Will they come as an application or drivers? Thanx a lot in advance! Paul Greece.
Both I guess
@@AncientGameplays That's awesome news!
Hmm, I read that ROCM would come to the 6000s series too. Even R9 Fury X supposedly has some compability
Yes, but firstly to RDNA3 it seems
I don't think that this move will shake the professional gpu market but it is a nice feature to have.
I think they know it. It seems it is more about having a complete AI suite. The advantage is that I'll be able to train light language processing models (rather than using the CPU) on my main rig and that's bloody cool. The convenience for developpement is nice to have for a data scientist using a laptop to process large datasets (they all do lol). AMD wants to gain shares in this market
Awesome news! I really love AMD now. Ever since I got my 6700xt I've been team AMD. I just discovered something that has increased my productivity in Ableton 11 (music software). AMD virtual super resolution! I can set my 1080p monitor to 4k! It's only good for this purpose though, the picture quality badly suffers but it doesn't matter if I'm just working on music.
that's a strange thing indeed, but cool!
Well nvidia has the exact same thing, but if you're happy with your card, that's indeed great.
How is that useful?
Makes no sense man 😂
Sadly, I have little faith in ROCm. I would hope they will fix it, but the documentation is chaos. Coding with even the AMD workstation GPUs that support it, is nightmarish. Difficult compatibility, bugs, the support is not there. And we really need a competitor in the GPGPU and computational market
AMD's marketing department should pay attention to this kind of coments. It's not the first time I read people complaining about lack of support and documentatión for devs
it will get there don't worry
if they implement this appropriately this could be a huge win for AMD. I'm sure it still won't be quite as fast as true cuda cores, but if they could hit a solid 80 to 90 percent of cuda speeds with less expensive cards they could pull many professionals over. I worked in the enterprise side of things and nvidia has been pissing their partners off like crazy basically since Volta and Pascal
Definitely, even if the price point will somewhat inflated a little amd would still attract enough new users to further advertise their product.
Amd hip is not translation layer.
Developer needs to convert their cuda codes into hip codes then recompile.
However the conversion is said to be easy as replacing cuda... to hip...
Thanks for the info, although I said "it is like a translation layer" haha
Great news as I was just "forced" to buy Nvidia for blender. Still rocking 6600xt in my gaming rig.
Then you'll be able to test it sooner or later :D
Doesn't AMD already beat Nvidia in photoshop and premier pro because of vram?
@lolthekidison It's why nvidia cards sometimes run better in some games and worse in some games to amd competitors. Nvidia has a vram bottleneck, and amd has a software bottleneck. It's that tradeoff sometimes people will make but I prefer better hardware so I switched from rtx to AMD. Amd better start impressing me though, nvidia broadcast & canvas has made it tempting.
@@PsychoBenches Personally I like the AMD software and overlay. I love my 6600xt and so do both of my friends that have 6650xts. I have a bunch of gpus and don't really have a preference of either or.
Unless you are making movies in blender, there is no advantage for nvidia in blender. The blender benchmark is only indicative of one very specific task, for modeling or most work you will run out of ram even with 64gb before you run into any performance limitations with even a 6800xt
Damn took so Long AMD to Utilize their Ai Accelerator Core on Those Rx 7000
Frickin' awesome. Dude, you're like.... AMD Jesus.
hahahah
YES
Wonderful
I just asked some Guys on a Stable Diffusion Channel and they told me that ROCm was just available for Linux.
Great that the confirmation is out that that is going to change :)
I can pass the time until Fall easily without SD. And maybe it even works good enough without ROCm ^^
Sooooo AMD just needs to lower the price a little more to count me on Board.
Thanks for bringing the good News!
4:03 100% agreed on that. I personally was not considering buying AMD gpu's for this reason. Future looks exciting indeed.
It would feel a bit "wasted" if 6000 series GPUs didnt support ROCm, especially since they are still very strong and have a lot of VRAM
They do it seems, but not on the AI part
The architecture focus is gaming workload not so much about compute. The reason RDNA exist is to keep out compute stuff from "bloating" their consumer card (which exist mainly for gaming and rendering 3d graphic) like what happen with GCN. in fact polaris and vega already have ROCm support in the past. Then AMD decided to stop pushing ROCm for consumer card starting with RDNA.
About time. CUDA has been leaving them in the dust when it comes to professional applications for too long.
Iam abosultely hyped !!! Thanks for the great video and news :) :)
As long as the support is only available in Linux, this is only a side note and nothing more.
- the official ROCm Support for AMD cards is weak at least. E.g. While an Rx6800xt works with ROCm out of the box, it's not officially supported. There are even cards you have to "fake" as different supported cards to get it running. We must expect the same for most RDNA3 cards.
- Even if everything is fine, Nvidia cards still smoke AMD in that area. It's the same like in raytracing. Maybe it changes a bit with RDNA4, but Nvidia is years ahead in that area. We still have to face that sad truth.
ROCm makes the situation for AMD fans a bit better. But not for professional users.
I'm looking at the products Nvidia has been turning out lately and it's almost like they think of their Video Cards as an afterthought since their AI is cruising to new heights. The 40X series is a great example. Why would any company create a modern card with a 128 bit bus? AMD though, seems to be all in with their cards. They don't shortchange the Vram and they have forgotten that a 128 bit bus ever existed. If ROCm turns out to be conductive to great gaming, I'll be switching to an AMD CPU running an AMD GPU. Lets see how this plays out ....
ROCm has nothing to do with gaming. You said nvidia is going with AI while giving gpu as a thing go render game as an after thought, AMD pushing ROCm on RDNA is going the same route as nvidia with AI. the reason AMD make CDNA for compute and RDNA for gaming is so they don't have to push things like AI on consumer cards. In the past Polaris and Vega were already have ROCm support. Then with RDNA AMD completely stop supporting ROCm on consumer card. Now they decided to once again give consumer card ROCm support with RDNA 3. if anything AMD realize that they cannot ignore the potential of generating more income by offering AI support on consumer card rather than make it exclusive to CDNA.
preview driver has MPO issues again during overlay transitions such as steam or AMD overlay toast notification for example.
MPO is what makes low power consumption possible on idle
Currently running it with MPO enabled and have no such issues on the RX 7600/7900XTX
@@AncientGameplays try playing more games its a glitch that shows it self as you start game during the steam toast popup from steam overlay or radeon toast may not show it self every game.
I can run pytorch on a rx6600xt. It uses rocm/hip to use the cuda code... So it's already working with rdna2 - how well supported it is might be another story.
on windows or on linux?
@@willgp6740 Linux.
Linux. Your best bet on linux is to a docker image with rocm(and friends)/pytorch/conda preinstalled. ex: docker pull rocm/pytorch-nightly@@willgp6740
Would you do a follow up on this as it would apply to Linux, Steamdeck, Proton and Vulkan? This seems more relevent as Linux is fast catching Windows as a gaming OS and Windows 11 has actually shed overall downloads compared to Windows 10. Probably due to the requirement for a MS Account to register and complete the download.
I know there are instances where the games are significantly better and others where it is significantly worse on Linux, but that's with RDNA 2. I'm not sure if ROCm means having to use Pop or nVidia drivers to gain the greater performance. Any thoughts?
I can try
Maybe with Maroc we can see the true power of the RDNA3S' architecture...
This could be a game changer.....I have actively avoided AMD cards because of the lack of CUDA......I'll be watching this very closely. Thank you for this amazing info :)
If it can run the same CUDA code, with minimal changes, it will be a game changer. A lot of applications can benefit from GPU acceleration. AMD cards are perfectly capable of "GPGPU computing" (GP - general purpose), as evidenced by the tons of game benchmarks coming out each day. OpenCL exists, but it doesn't have the same adoption rate as CUDA, probably because AMD's market share is in the single-digits and Intel is eating THEIR lunch not Nvidia's. And AMD seems to be okay with that, instead of just lowering their prices. 😕
From what I understand the code will still need to be recompiled, just with different headers than "vanilla" CUDA. Same code (mostly), just different headers.
AMD's CUDA alternative is called "HIP" and it can compile down to both CUDA and AMD native instructions. They also have a tool called "HIPify" that automatically converts most CUDA programs to HIP
I'm confused what you mean about AMD not lowering their prices, considering AMD has been lowering their prices. 🤨
ROCm has been working on RDNA1 and 2 for a while, the RDNA3 support is the new thing. It's also crap
Chill, it hasn't even come officially yet.
I went AMD because Nvidia are just price gouging. especially in Australia. 7900xtx was $1500 but 4080 was $2000 and up. i get the shits when i look at prices overseas
Intel arc кстати сейчас отлично соревнуется с CUDA от NVIDIA. До выхода этого видео было только две видеокарты, которые я бы взял, но если AMD всё хорошо сделает со своим HIP и ROCm то я возьму на заметку их видеокарты
Does this mean my rx 6750 xt can do deep learning now?
This will improve stable diffusion performance?
Humm, don't really know....
good stuff thank you for sharing. 👍👍👍👍
So 6 months for the linux and with some good luck 2025 for all on windows?
Blender? There are now 5 API's to accelerate it's render engine.
OneAPI, AMD HIP, Optix, Cuda and metal all targeting different hardware and / or platforms.
Imagine if this was also needed for just the 3D viewport.
OpenGL for AMD older cards
NvidiaGL for Nvidia
Direct3D for Intel
Metal for Apple
Vulkan for newer AMD cards
GPGPU is just a mess mostly because of Nvidia's dominant cuda but also because of openCL's failure.
When blender stopped support for opencl for having slow performance and "unfixable" bugs, no one had any idea with what it would be replaced. Many including me thought it wouldn't be replaced at all. Or that opencl (with new version) would be re-enabled.
Well we got something new. And yesterday blender released version 3.6 long term support. LTS versions are more focused on stability over new features. And we got AMD accelerated rendering, but still broken and not giving proper results in certain cases and is still much slower than comparable Nvidia cards. It was the saddest thing to hear in the official blender 3.6 LTS features video. It's here, but still broken.
The irony, Dropping opencl for being buggy and slow, developing HIP in it's place to fix that and after all that time we now have buggy and slow HIP.
Khronos group was blamed for opencl's failure. Who gonna be blamed for this now?
I just ordered an asrock Taichi 7900 xtx yesterday, but I dont care for these features at all, since I dont think I will ever need them. But its good that AMD becomes a bigger competition with this now. That always means more and better offers for us customers.
Unrelated to the topic but what game is that at 0:59?
Its Forspoken
Thank you sir!
I have a question:
is there an Nvidia equivalent of
,,Amernime Zone" modded drivers ?
I dont think so.
I know random question but what’s your main gaming monitor ?
Lg 34gn850
Think I just found my new gaming monitor!
when exactly is it coming out?
I believe ROCm 5.6 is already out, linux supp will come soon
any new updates???
Bro is literally the messiah of AMD
hi Fabio, amd hasn't released new gpu drivers for a month, do you know something? Thank you
I do, some big things are coming!
@@AncientGameplays thanks !
That's great news for amd GPU owners. Will you do a benchmark when the rocm update comes to amd GPUs ?
As soon as developers adopt it I will
Will it come to windows as well, I'm looking forward to get pytorch support too on my 6700xt
What do you think about the : AMD Software Adrenalin Edition 23.10.01.41 ?? do they worth the try ???
Just finished testing them
what game is the one shown around the time stamp of 1:00? it looks awesome!
Its forspoken
I wouldn't be so hyped before it really launched and we know the performance improvement
Any improvement is always welcomed
You can already use AMD CPUs for stable diffusion using a special fork of stable diffusion thats uses DirectML, I have stable diffusion running on a humble Vega 56.
By any chance it comes for 6k series?
It always does on Linux I believe
Finally open widely available competition for CUDA.
Nothing new really. Initially AMD create Stream to compete directly with CUDA. then then they move to OpenCL. Now OpenCL also did not work well for them they have ROCm. And they use nvidia CUDA to make things work for ROCm.
Recently bought a oc formula 6950xt, will it support RDNA3?
No, it's based on RDNA2 architecture.. maybe you mean if the 6950XT will support FSR3?
Rdna3 is the architecture. Also, it already does to some extend, just not officially
I hope it changes! AMD are much more generous with Vram compared to nvidia. Combine lots of vram with hardware acceleration for professionals and it’s best of the best. Nvidia won’t have a leg to stand on.
Even 24gb is getting tiredsome for creators. You don’t need 24gb what’s so ever for gaming, but it’s much more required for professional creators. We’ve had it since all the way back to the RTX Titan. Hopefully AMD continue to grow competitively and nvidia do the same.
Companies just want money, that's all
Yes. And VRAM is also really important for running AI models on a PC. For example I love Stable DIffusion. And with my RTX 2080 8GB it's nice but If I am looking to upgrade I just want much more VRAM and the price of even the new Super cards are crazy. Max 700 Dollars for the top tier card is what I would say OK. But even the 4080 is 800 MSRP dollars then AIBs put a markup on it and because I live in Europe those cards will cost even more here...
Hey, as you are doing lots of Videos for Amd Software/drivers, do you know when they release official drivers for 7840u apu? Thanks
When all the main drivers get released...when? Don't know...
Would that improve gaming experience?
Nah, not really
Awesome, as is your videos man.. just keep inserting dem memes even more haha
Thanks hahha
Everything AMD makes is free for all... Everything Nvidia makes is strictly theirs and no one else can adopt to come close to them... IDK who is an unfair competitor...
From now until the foreseen future it's all about AI. AI-generated art, videos, language models/chatbots, etc. Most 3D apps are optimized for CUDA as well. Hopefully AMD's GPUs can catch up. They have the cheapest 24GB of VRAM option in the 790XTX. However, once bitten twice shy - I'm sticking with team green because my livelihood is dependent upon AI.
It would be great if ROCm came to the RX 6950 XT and lower
They already support it, just not officially I believe
please don't listen to the creator of this video. ROCm is available for older GPUs for a long time already. Their post was only that they'll add support to the RDNA 3. I've used ROCm for generating images using stable diffusion and AI accelerated translation on my 6900XT running on Ubuntu for a while. The only problem with ROCm that I see it's that it's linux only and can be difficult to setup
They should support RDNA 1 and 2
they do/will
ROCm works well on 6700xt in Linux, there's many TH-cam videos using it. I've tried that and failed.
Maybe not officially?
@@AncientGameplays The one thing that was missing was rocm windows. But maybe there'll be some improvements.
Translation layer, is called a wrapper also, no ?
Nah, not exactly
This will make blender users happy
will ROCm be supported for older GPUs for example Polaris cards? I'm interested in the core accelerators.
I dont think so
@@AncientGameplays
sad that CUDA's competitor won't be available for these cards. maybe redditors or github will have workaround to this.
To my knowledge some older ROCm version did work on polaris and vega. But the new ROCm most likely on newer hardware only. Heck AMD even already cease to support newer ROCm on their MI100 Instinct.
wait a minute but i remember saw an article that confirms months ago that they ad support for the RX 6600 and RX 6900 XT
It does but not AI
@@AncientGameplays Soo they are using it for deep learning?
until there are real results, this remains again a fiction in the style of fsr 3.0 coming soon.
again you have to wait (amd)
One of my professors at the university has worked in AMD. He used to write GPU manuals for developers and optimize the efficiency. I hope AMD is able to provide a solid competition to CUDA. Thats the only keeping me from buying an AMD GPU even tho they are the cheaper option.
The cuda acceleration in davinci resolve is gooood , so i will be keeping an eye on this.
I am interested that my next gpu is an amd one. I used to have amd, i switched to nvidia and going back to amd :)
Like a back and forth
7900xtx is already faster than a 4090 in resolve, has been for a while, adding ROCm won't really change anything since it was already gpu accelerated.
It depends on the type of Vega GPU.
I don't know if its only the Vega VII or both the Vega VII and the Vega 64.
finally... Im rendering faster with my CPU (R7 5800X) in After Effects / Sony Vegas than with my GPU so this would be great
Great content! name of the Outro song from 6.00 minute?
Let me check
I have a question and need help AMD god! This is random I know but Ihave been told that using MSI afterburner with radeon software is not advised but I don't overclock. I am simply trying to make a custom fan curve. If only making a custom fan curve for the gpu, will afterburner interfere with adrenaline? Again, no overclock and no custom fan curve using adrenaline.
What software do you personally use to make a fan curve for your gpu?
you're completely fine. Have been using them together for years with several builds...
@@AncientGameplays Thank you for the feedback. I greatly appreciate you.
Still running 22.8.2 on my RX 6800 non XT.
Haven't played any games for almost half a year...
What's a good driver these days?
My pc runs on Linux, I do a lot of server and virtualization workloads. I have a 6900xt and use it for gaming. I also use it as a Plex server. I have Plex Pass, however, Plex does not support AMD hardware at all. That means my GPU that is more than capable of transcoding sits idle, and my CPU gets crushed. I can only manage 2 4k transcoded streams on my 5900x. Its gotten to the point that I have debated getting a second Nvidia GPU just for transcoding. This is amazing news!!!
I believe Jellyfish supports AMD GPUs
I wish youtubers would post on every video what accent they're speaking with because I end up trying to identify it instead of listening to the video. Halfway through I know it's Portuguese so now I can listen to what he's saying instead of listening to how he's saying it 🙂
Correct
But is the 7800xt coming?
yes
@@AncientGameplays Any official news from AMD themselves about either a 7800(XT) - Navi31 or 32 -, 7700(XT)? And what about 7600 XT?
ROCm sound's great but why didn't they even use their chiplet & wafer stacking technology, with their recent 7900XTX / XT card's AND then gave us a card with a 128bit bus (RX 7600)?
I just received a Sapphire 7900XTX Nitro+ for gaming only. Regardless of not needing ROCm, I’m still glad to have it. “If it’s free, it’s for me” as the saying goes.
Any new updates?
Yes, now we have ROCM with Linux subsystem on windows
Oh hell NO! The HIP-CUDA remapping may cause performance penalties on Radeon GPU performance, I would prefer major software like blender and DaVinci resolves will work on Radeon LLVM like ONNX does, to get the maximum performance of Radeon GPU. Software Devs should work on "ROCm-LLVM" and not on HIP-CUDA.
I do believe that with ROCm available to use easily, things will move much faster towards that direction!
Thanks for share the information
This is the biggest free update from AMD to RX 6000 and RX 7000 series...
I dont understand. he says rdna 3 at 2:21
why did u include rx 6000 series? I hope you are right cause I have rx 6900 xt right now
@@fuku474 AMD to enable ROCm on Windows, add support for some gaming Radeon GPUs issued by Videocardz website mentions ROCm 5.6.0 Alpha support table Radeon RX 6900 XT, RX 6600 and R9 Fury are listed...I think AMD first offers ROCm update to newly lauched rx 7900xtx and Radeon pro W7900...later i hope rdna 2 gpus will also get this update....
Para quando FSR 3.0?
Nem sei...
the thing I think AMD didn't realize is a lot of the things that push the industry forward are made by passionate professionals, on their home machines while not at work. The windows functionality to double click on a zip file to enter it like any other folder wasn't made by microsoft, it was BOUGHT by microsoft. Even the things that a company does initially release can typically be traced back to some programmer wanting it to happen on their own and then the company realizes afterwards that it's a really good idea. It's basically the selfish altruism effect; if you give people the freedom to make their own things, chances are you'll end up profiting from them in the process. (hell don't some companies like Google even have mandatory paid time allotments to work on personal projects during work-hours for this exact reason?)
Nvidia putting CUDA on their consumer line of GPUs meant that tons and TONS of hobbyists could work with their Nvidia GPU, but couldn't work with their AMD GPU. If the choice is between two roughly comperable options but one lets you do something the other doesn't, that one's going to get the adoption.
By AMD only putting the tool on their cards designed specifically for that purpose they nuked any chance of a community of innovation developing around it, so more and more projects used CUDA, and most people didn't even know AMD had a competitor to it at all.
Can it run games with hardware physx acceleration?
Physix is dead and burried basically
Do you think 5700xt will work ever on windows rocm like it can on linux rocm
Don't know... possibly
I wish it would work with RDNA1.
And Intel is just watching without trying to join the market.
i'm impressed what AMD achieves with lot less workers to compete with Intel + Nvidia, maybe should try AMD products 👌🏽
gonna buy 7800x3d, and an Asus board today 😬 for a second build
@@nivea878 Asus board for 7800x3d after last accident? Bro, you are brave :D
23.10.01.41 preview driver fixed vr and improved power draw for rdna 3 🎉
And more! Already tested them 💪💪
it's coming? already using for 2 yrs
Not for AI my man
@@AncientGameplays AI is basically FP32 and matrix operations and can be run on any modern gpu and CPU, just not as fast as a dedicated hardware instruction set
Is there no way of this supporting my 6700 XT? And this cuda acceleration will benefit in editing software like DaVin ci Resolve right?
Would I be able to use it in a hp proliant ml 150?
Dont know
This is great news for waifu generation. 😅
Is RT and DLSS 2 and 3 worth it for 100-200 dollars more like 7900xtx vs 4080
4080 is way more stable card
Nah, not even close IMO
Hey man, quick thanks for all the videos and tips. But returned my 7900 XTX and got a 4090 - Warzone Caldera and WZ2 and Forza 5 was just too stuttery for me.
That makes absolutely no sense, and I am sure you had some issues there (maybe even a defective card) as Warzone and Forza run much better on the AMD side....
@@AncientGameplays No denying the games had good fps, but textures on high on WZ would cause bad stutter. Changing to low helped. Forza could not get rid of the stutters every 15 seconds or so
I would buy AMD if they did support Linux, and if I could use it like I do my NVIDIA card, I mean I spent 1 year trying to get accelerated ray tracing to work on my AMD graphics card after all that frustration I switched to NVIDIA, and I have everything I want, I have game ray tracing, I have real time ray tracing in blender, I can do a lot of stuff in blender that I cant do with AMD because of lack of support from AMD. The only way I will ever switch to AMD GPU again is if they do what NVIDIA do.
I've put my old 1080Ti in Thunderbolt enclosure and my mom uses for Blender with her ThinkPad - still smokes each and any AMD card T_T
It remains to wait for the ability to run games with physx on AMD video cards and an analogue of nvidia profile inspector
Profile inapector? Yes! Physix? God no
It's coming to 'selected' GPUs ~ that doesn't mean it's coming to gamers. Or home user geeks. Or to all Radian GPUs....
It is a positive move and I do applaud (took 'em long enough) but you don't unroll a ten ~ 20 year head start with an announcement...
Correct me if I'm wrong, but any CUDA code will run on any nVidia hardware ~ although it will run much better on modern high-end hardware. So are we going to see all CUDA code run run on all RDNA and newer hardware? When are we going to see that? Is this also going to be true in Linux?
I do applaud the move ~ took you long enough ~ but an announcement is not the same as delivered working product. Absolutely critical ~ same as AMD64 being the exact same as Intel64 ~ that CUDA code should run on AMD precisely the same way it does on CUDA. You have to be able to just take all the CUDA code that works, and plug it in without touching it. If it runs properly on nVidia, then it will run properly on ROCm.
🙄
"same as AMD64 being the exact same as Intel64 " it's the inverse, actually it's called AMD64 in windows and drivers for a reason, the first 64bit CPU on X86 was the "athlon 64" not the "pentium" of the same era !
@@barbatocedric4246 ~ Yes, perfectly correct. People asked Intel for a 64 bit version, same as we had gone from a 16 bit to a 32, but Intel said 'that's too hard' ~ so AMD developed it. Intel was peddling their own version of 64 bit, called Itanium ~ which was not backwards-compatible. But the idea is all those 'intel' systems are compatible and consistent. Compare to a sparc chip or Arm chip or some other completely different architecture. You can get hold of a program which is 5 years old, and it will run. 10 years, 20 years ... As long as the operating system will allow it, you can run The Dungeons of Umoria in all its 1986 glory without a dosbox or a virtual machine. That game is now almost 40 years old, but I still play it. In terms of providing a software environment for the Enterprise, this is a double edged sword. It means old code still works, but that can & does lock you into supporting 50 year old gold plated virgins who nobody understands any longer, and the original author died the day mankind kicked the moon, murdered by Ritchie & Kernigan. But the whole enterprise depends on that little piece of code running ~ Goldman & Sax discovered a way of controlling the futures market in 1973, with this 50 lines of assembly code ~ and no matter what, all future systems must be able to run an instance of this piece of Fortran or else! Nobody alive today has ever seen the source-code for it... What it does is magic ~ it turns 55 year old men in a boardroom, into willing slaves and cum-buckets, and they'll sign anything, pay anything, suck anything, and agree to anything, as long as that piece of code continues to run.
Follow-up 18 hours later.
th-cam.com/video/LKxWTYyplow/w-d-xo.html
It will be on Linux, not Windows, and it will support two SKUs ~ one being a high end Pro SKU, the other being a top end gamer SKU.
I stand by what I said ~ it is a move in the right direction, but a very small one. They say they intend to grow and expand it, but so far, it's like the cherry on top. That's decoration ~ that's not a course, let alone a meal.
@@Kneedragon1962 Meh, we need windows...not Linux..