5 ms·

Ooo, that means its probably time for me to get a used 2080, or maybe even a 3080 if I'm feeling special

by voidUpdate 2y ago

Kelteseth 2y ago
Why not go for AMD? I just got a 7900XTX for 850 euros, it runs ollama or comfyUI via WSl2 quite nicely.
- whywhywhywhy 2y ago
  Pointless putting yourself through the support headaches or having to wait for support to arrive to save a few dollars because the rest of the community is running Nvidia
  - Kelteseth 2y ago
    Nah it's quite easy these days. Ollama runs perfectly fine on Windows, comfyUI still has some not ported requirements, so you have to do stuff through WSL2.
- viraj_shah 2y ago
  Do you have a good resource for learning what kinds of hardware can run what kinds of models locally? Benchmarks, etc? I'm also trying to tie together different hardware specs to model performance, whether that's training or inference. Like how does memory, VRAM, memory bandwidth, GPU cores, etc. all play into this. Know of any good resources? Oddly enough I might be best off asking an LLM.
  - holoduke 2y ago
    To prevent custom implementations is recommended to get a Nvidia card. Minimum 3080 to get some results. But if you want video you should go for either 4090 or 5090. ComfUI is a popular interface which you can use for graphical stuff. Images and videos. Local text models I would recommend to use the Misty app. Basically a wrapper and downloader for various models. Tons of youtube videos on how to achieve stuff.
  - Kelteseth 2y ago
    I tested ollama with 7600XT at work and the mentioned 7900XTX. Both run fine with their VRAM limitations. So you can just switch between different quantization of llama 3.1 or the vast amount of different models at https://ollama.com/search https://ollama.com/search
- orphea 2y ago
  AMD driver quality is crap. I upgraded from GTX 1080 to RX 6950 XT because I found a good deal and I didn't want to support nvidia's scammy bullshit of launching inferior GPUs under the same names. Decided to go with AMD this time, and I had everything: black screens, resolution drops to 1024x768, total freezes, severe lags in some games (BG3) unless I downgrade the driver to a very specific version.
  - mldbk 2y ago
    It is an outdated claim. I have both 4090 (workstation) and 7900XT (to play some games) and I would say that 7900XT was rock solid for me for the last year (I purchased it in Dec 2023).
    - orphea 2y ago
      It is an outdated claim. You assume that. I wish it was. This is my actual recent experience with AMD and their drivers. It did get better in the past few months, but I still occasionally get total freezes when only a hard reset helps. Good for you though.
- williamDafoe 2y ago
  AMD is an excellent choice. NVidia UI has been horrible and AMD adrenaline has been better than NVidia for several years now. With NVidia, you are paying A LOT of extra money for trickery and fake pixels, fake frames, fake (ai) rendeering. All fakeness. All hype. When you get down to the raw performance of these new cards, it must be a huge disappointment, otherwise, why would Jensen completely forget to mention anything REAL about the performance of these cards? These are cut-down cards designed to sell at cut-down prices with lots of fluff and whipped cream added on top ...
- satvikpendem 2y ago
  DLSS is good and keeps improving, as with DLSS 4 where most of the features are compatible with even the 2000 series cards. AMD does not have the same software feature set to justify a purchase.
Macha 2y ago
The 2080 was a particularly poor value card, especially when considering the small performance uplift and the absolute glut of 1080 Tis that were available. A quick look on my local ebay also indicates they're both around the €200-250 range for used buy it now, so it seems to make way more sense to go to a 3080.
- qingcharles 2y ago
  2080 TI though is a really good sweet spot for price/performance.
vonneumannstan 2y ago
a 4070 has much better performance for much cheaper than a 3080...
- rtkwe 2y ago
  Any the 4070 Super is relatively available too. I just bought one with only a small amount of hunting. Bought it right off of Best Buy, originally tried going to the Microcenter near my parent's house while I was down there but should have bought the card online for pickup. In the 2 days between my first check and arriving at the store ~20 cards sold.
  - alkonaut 2y ago
    What was the drop in 3070 pricing when the 4070 was released? We should expect a similar drop now I suppose?
    - rtkwe 2y ago
      It took a while according to the first price chart I found. The initial release of the 4070 Ti/FE in Jan 2023 didn't move the price much but the later release did start dropping the price. Nvidia cards are pretty scarce early in the generation so the price effect takes a minute to really kick into full force. I just upgraded from a 2080 Ti I had gotten just a few weeks into the earliest COVID lockdowns because I was tired of waiting constantly for the next generation. https://howmuch.one/product/average-nvidia-geforce-rtx-3070-8gb/price-history https://howmuch.one/product/average-nvidia-geforce-rtx-3070-...

Ooo, that means its probably time for me to get a used 2080, or maybe even a 3080 if I'm feeling special

Kelteseth 2y ago

Why not go for AMD? I just got a 7900XTX for 850 euros, it runs ollama or comfyUI via WSl2 quite nicely.

whywhywhywhy 2y ago
Pointless putting yourself through the support headaches or having to wait for support to arrive to save a few dollars because the rest of the community is running Nvidia
- Kelteseth 2y ago
  Nah it's quite easy these days. Ollama runs perfectly fine on Windows, comfyUI still has some not ported requirements, so you have to do stuff through WSL2.
viraj_shah 2y ago
Do you have a good resource for learning what kinds of hardware can run what kinds of models locally? Benchmarks, etc? I'm also trying to tie together different hardware specs to model performance, whether that's training or inference. Like how does memory, VRAM, memory bandwidth, GPU cores, etc. all play into this. Know of any good resources? Oddly enough I might be best off asking an LLM.
- holoduke 2y ago
  To prevent custom implementations is recommended to get a Nvidia card. Minimum 3080 to get some results. But if you want video you should go for either 4090 or 5090. ComfUI is a popular interface which you can use for graphical stuff. Images and videos. Local text models I would recommend to use the Misty app. Basically a wrapper and downloader for various models. Tons of youtube videos on how to achieve stuff.
- Kelteseth 2y ago
  I tested ollama with 7600XT at work and the mentioned 7900XTX. Both run fine with their VRAM limitations. So you can just switch between different quantization of llama 3.1 or the vast amount of different models at https://ollama.com/search https://ollama.com/search
orphea 2y ago
AMD driver quality is crap. I upgraded from GTX 1080 to RX 6950 XT because I found a good deal and I didn't want to support nvidia's scammy bullshit of launching inferior GPUs under the same names. Decided to go with AMD this time, and I had everything: black screens, resolution drops to 1024x768, total freezes, severe lags in some games (BG3) unless I downgrade the driver to a very specific version.
- mldbk 2y ago
  It is an outdated claim. I have both 4090 (workstation) and 7900XT (to play some games) and I would say that 7900XT was rock solid for me for the last year (I purchased it in Dec 2023).
  - orphea 2y ago
    It is an outdated claim. You assume that. I wish it was. This is my actual recent experience with AMD and their drivers. It did get better in the past few months, but I still occasionally get total freezes when only a hard reset helps. Good for you though.
williamDafoe 2y ago
AMD is an excellent choice. NVidia UI has been horrible and AMD adrenaline has been better than NVidia for several years now. With NVidia, you are paying A LOT of extra money for trickery and fake pixels, fake frames, fake (ai) rendeering. All fakeness. All hype. When you get down to the raw performance of these new cards, it must be a huge disappointment, otherwise, why would Jensen completely forget to mention anything REAL about the performance of these cards? These are cut-down cards designed to sell at cut-down prices with lots of fluff and whipped cream added on top ...
satvikpendem 2y ago
DLSS is good and keeps improving, as with DLSS 4 where most of the features are compatible with even the 2000 series cards. AMD does not have the same software feature set to justify a purchase.

Macha 2y ago

The 2080 was a particularly poor value card, especially when considering the small performance uplift and the absolute glut of 1080 Tis that were available. A quick look on my local ebay also indicates they're both around the €200-250 range for used buy it now, so it seems to make way more sense to go to a 3080.

qingcharles 2y ago
2080 TI though is a really good sweet spot for price/performance.

vonneumannstan 2y ago

a 4070 has much better performance for much cheaper than a 3080...

rtkwe 2y ago
Any the 4070 Super is relatively available too. I just bought one with only a small amount of hunting. Bought it right off of Best Buy, originally tried going to the Microcenter near my parent's house while I was down there but should have bought the card online for pickup. In the 2 days between my first check and arriving at the store ~20 cards sold.
- alkonaut 2y ago
  What was the drop in 3070 pricing when the 4070 was released? We should expect a similar drop now I suppose?
  - rtkwe 2y ago
    It took a while according to the first price chart I found. The initial release of the 4070 Ti/FE in Jan 2023 didn't move the price much but the later release did start dropping the price. Nvidia cards are pretty scarce early in the generation so the price effect takes a minute to really kick into full force. I just upgraded from a 2080 Ti I had gotten just a few weeks into the earliest COVID lockdowns because I was tired of waiting constantly for the next generation. https://howmuch.one/product/average-nvidia-geforce-rtx-3070-8gb/price-history https://howmuch.one/product/average-nvidia-geforce-rtx-3070-...