https://pytorch.org/blog/pytorch-1.8-released/#beta-amd-gpu-binaries-now-available
\nNvidia has been dominant in the deep learning space with cuda for a long time now. Will be interesting to see how the AMD gpus perform with this and if they are at all competitive. If any of you have AMD gpus around it would be interesting to see some results.
", "messages": [ { "id": "1226940", "postDate": "03/05/2021 03:35:33", "content": "https://pytorch.org/blog/pytorch-1.8-released/#beta-amd-gpu-binaries-now-available
\nNvidia has been dominant in the deep learning space with cuda for a long time now. Will be interesting to see how the AMD gpus perform with this and if they are at all competitive. If any of you have AMD gpus around it would be interesting to see some results.
", "rawMarkdown": "https://pytorch.org/blog/pytorch-1.8-released/#beta-amd-gpu-binaries-now-available\n\nNvidia has been dominant in the deep learning space with cuda for a long time now. Will be interesting to see how the AMD gpus perform with this and if they are at all competitive. If any of you have AMD gpus around it would be interesting to see some results.", "votes": null }, { "id": "1226956", "postDate": "03/05/2021 03:54:54", "content": "That is great news! AMD GPUs are slightly cheaper and have more VRAM per dollar. If it proofs to be competitive, ML person would have more options when it comes to buying GPU.
\nP.S. who am I kidding, no one can get GPU anyways.
", "rawMarkdown": "That is great news! AMD GPUs are slightly cheaper and have more VRAM per dollar. If it proofs to be competitive, ML person would have more options when it comes to buying GPU.\n\nP.S. who am I kidding, no one can get GPU anyways.", "votes": null }, { "id": "1226962", "postDate": "03/05/2021 04:08:03", "content": "GPUs are a bit mythical right now. So hard to find. AMD ones are attainable, but still hard to find. Nvidia it is impossible right now.
", "rawMarkdown": "GPUs are a bit mythical right now. So hard to find. AMD ones are attainable, but still hard to find. Nvidia it is impossible right now.", "votes": null }, { "id": "1227090", "postDate": "03/05/2021 07:31:30", "content": "Can we be sure everything will work fine without Cuda ? I wanted to buy Zotac RTX GPU, even then I'm getting cold feet since it's not exactly NVidia
", "rawMarkdown": "Can we be sure everything will work fine without Cuda ? I wanted to buy Zotac RTX GPU, even then I'm getting cold feet since it's not exactly NVidia", "votes": null }, { "id": "1227091", "postDate": "03/05/2021 07:31:59", "content": "i'll be reserved until I see AMD team at top of the LB :)
\njoke aside, nice to see some moves in this regard
", "rawMarkdown": "i'll be reserved until I see AMD team at top of the LB :)\n\njoke aside, nice to see some moves in this regard", "votes": null }, { "id": "1227129", "postDate": "03/05/2021 08:18:41", "content": "Zotac RTX GPU is Nvidia
", "rawMarkdown": "Zotac RTX GPU is Nvidia", "votes": null }, { "id": "1227173", "postDate": "03/05/2021 08:55:11", "content": "I myself want to buy one but Nvidia 3000 series is selling at twice to thrice the founder edition price because of prices of cryptocurrency and mining :)
", "rawMarkdown": "I myself want to buy one but Nvidia 3000 series is selling at twice to thrice the founder edition price because of prices of cryptocurrency and mining :)", "votes": null }, { "id": "1227249", "postDate": "03/05/2021 10:47:23", "content": "I am curious, if anyone has already tried and if so what are results in comparison with cuda?
", "rawMarkdown": "I am curious, if anyone has already tried and if so what are results in comparison with cuda?", "votes": null }, { "id": "1227884", "postDate": "03/05/2021 22:22:12", "content": "It has been theoretically possible to use pytorch with rocm for a while now, but this is the first time there are binaries included to support it. People would have to build from source and it was a hairy process. I remember seeing some results on reddit at some point but I cant find them anymore. Looked pretty decent on the vega cards
", "rawMarkdown": "It has been theoretically possible to use pytorch with rocm for a while now, but this is the first time there are binaries included to support it. People would have to build from source and it was a hairy process. I remember seeing some results on reddit at some point but I cant find them anymore. Looked pretty decent on the vega cards", "votes": null }, { "id": "1228698", "postDate": "03/06/2021 16:50:03", "content": "This is really exciting. Are there benchmarks available anywhere?
", "rawMarkdown": "This is really exciting. Are there benchmarks available anywhere?", "votes": null }, { "id": "1228831", "postDate": "03/06/2021 20:15:29", "content": "Found the benchmarks I looked at a while ago https://www.reddit.com/r/Amd/comments/9yynds/vega_fe_tensorflowrocm_benchmark/
\nhard to say how much has changed in the two years since this was done
", "rawMarkdown": "Found the benchmarks I looked at a while ago https://www.reddit.com/r/Amd/comments/9yynds/vega_fe_tensorflowrocm_benchmark/\n\nhard to say how much has changed in the two years since this was done", "votes": null }, { "id": "1235252", "postDate": "03/12/2021 02:09:17", "content": "wow，exciting
", "rawMarkdown": "wow，exciting", "votes": null }, { "id": "1236171", "postDate": "03/12/2021 20:42:19", "content": "I wanted and desperately wanted to buy a treatment unit from that company. Can you suggest to me that you have the experience any experience in order to help me choose the most suitable
", "rawMarkdown": "I wanted and desperately wanted to buy a treatment unit from that company. Can you suggest to me that you have the experience any experience in order to help me choose the most suitable", "votes": null }, { "id": "1236364", "postDate": "03/13/2021 04:06:10", "content": "It seems what everyone is missing here is that the 6xxx series doesn't support ROCm - neither does the 5700xt/5700.
\nhttps://github.com/RadeonOpenCompute/ROCm/issues/1390
", "rawMarkdown": "It seems what everyone is missing here is that the 6xxx series doesn't support ROCm - neither does the 5700xt/5700.\n\nhttps://github.com/RadeonOpenCompute/ROCm/issues/1390", "votes": null }, { "id": "1237106", "postDate": "03/13/2021 18:48:39", "content": "They only support old hardware like AMD Radeon RX Vega 64. About two generations behind the current technology.
That is great news! AMD GPUs are slightly cheaper and have more VRAM per dollar. If it proofs to be competitive, ML person would have more options when it comes to buying GPU.
\nP.S. who am I kidding, no one can get GPU anyways.
", "votes": null, "replies": [ { "id": 1226962, "author_name": "ryches", "author_url": "", "post_date": "03/05/2021 04:08:03", "content": "GPUs are a bit mythical right now. So hard to find. AMD ones are attainable, but still hard to find. Nvidia it is impossible right now.
", "votes": null, "replies": [] } ] }, { "id": 1227090, "author_name": "nyleve", "author_url": "", "post_date": "03/05/2021 07:31:30", "content": "Can we be sure everything will work fine without Cuda ? I wanted to buy Zotac RTX GPU, even then I'm getting cold feet since it's not exactly NVidia
", "votes": null, "replies": [ { "id": 1227129, "author_name": "philippsinger", "author_url": "", "post_date": "03/05/2021 08:18:41", "content": "Zotac RTX GPU is Nvidia
", "votes": null, "replies": [] }, { "id": 1227173, "author_name": "vickygoyal", "author_url": "", "post_date": "03/05/2021 08:55:11", "content": "I myself want to buy one but Nvidia 3000 series is selling at twice to thrice the founder edition price because of prices of cryptocurrency and mining :)
", "votes": null, "replies": [] } ] }, { "id": 1227091, "author_name": "valanm", "author_url": "", "post_date": "03/05/2021 07:31:59", "content": "i'll be reserved until I see AMD team at top of the LB :)
\njoke aside, nice to see some moves in this regard
", "votes": null, "replies": [] }, { "id": 1227249, "author_name": "ademyanchuk", "author_url": "", "post_date": "03/05/2021 10:47:23", "content": "I am curious, if anyone has already tried and if so what are results in comparison with cuda?
", "votes": null, "replies": [ { "id": 1227884, "author_name": "ryches", "author_url": "", "post_date": "03/05/2021 22:22:12", "content": "It has been theoretically possible to use pytorch with rocm for a while now, but this is the first time there are binaries included to support it. People would have to build from source and it was a hairy process. I remember seeing some results on reddit at some point but I cant find them anymore. Looked pretty decent on the vega cards
", "votes": null, "replies": [] }, { "id": 1228831, "author_name": "ryches", "author_url": "", "post_date": "03/06/2021 20:15:29", "content": "Found the benchmarks I looked at a while ago https://www.reddit.com/r/Amd/comments/9yynds/vega_fe_tensorflowrocm_benchmark/
\nhard to say how much has changed in the two years since this was done
", "votes": null, "replies": [] } ] }, { "id": 1228698, "author_name": "anjum48", "author_url": "", "post_date": "03/06/2021 16:50:03", "content": "This is really exciting. Are there benchmarks available anywhere?
", "votes": null, "replies": [] }, { "id": 1235252, "author_name": "rungogo", "author_url": "", "post_date": "03/12/2021 02:09:17", "content": "wow，exciting
", "votes": null, "replies": [] }, { "id": 1236171, "author_name": "mohamedbakrey", "author_url": "", "post_date": "03/12/2021 20:42:19", "content": "I wanted and desperately wanted to buy a treatment unit from that company. Can you suggest to me that you have the experience any experience in order to help me choose the most suitable
", "votes": null, "replies": [] }, { "id": 1236364, "author_name": "stanleyjzheng", "author_url": "", "post_date": "03/13/2021 04:06:10", "content": "It seems what everyone is missing here is that the 6xxx series doesn't support ROCm - neither does the 5700xt/5700.
\nhttps://github.com/RadeonOpenCompute/ROCm/issues/1390
", "votes": null, "replies": [ { "id": 1237106, "author_name": "steamedsheep", "author_url": "", "post_date": "03/13/2021 18:48:39", "content": "They only support old hardware like AMD Radeon RX Vega 64. About two generations behind the current technology.