Question Need help to choose the right GPU for my usage

[Filll999-]

hello, let me specify immediately "ignoring gaming" (which I'm not very interested in) I would like to understand which are the most important general key specifications of a gpu ignoring the brute power of the card (rasterization) and ignoring areas such as gaming and content-creation, rather than other for the best possible visual experience (film, video streaming, YouTube), therefore what really matters and more realistically (if necessary, also take into account the gaming area). What is often left out to make room for pure gaming performance (which, for heaven's sake, is also right, but I don't think that's all), I can understand because most of the users to whom the gpu brands refer are certainly gamers, and fewer workers seek more productivity. Inquiring I ran into terms such as: decompression and decoders, AV1, Media Engineer, I would like to know what these technical specifications are useful for better video streaming quality (no gaming). After this, if you could answer me, I would like to know more specifically which are the most important specifications that are most important in a use and in fields of programming and statistical calculations: - AI, deep learning and neural networks - Machine Learning - Software Engineer. So knowing (in these areas) what I really need and what more is needed for such use. If you want, you can also recommend me some gpu models, you do what you think is most appropriate, but I would prefer to receive answers to these questions and therefore know what is important and what are the technical specifications of the video cards that I have to take into account for my first request and for this fairly specific type of use (not too specific because I couldn't tell you the specific actions because I'm fully learning these areas of AI, and maybe what I'll do in 2 years will be different). I really ask you to know what the gpu specifications are (vram, clock, core number, architecture, codec, etc.) to evaluate myself which model is best for me and make an informed choice. Thank you have a nice day

 

[Aeacus]

How about adding some spacing into your wall of text? So that anyone who reads it, doesn't hurt their eyes?

I really ask you to know what the gpu specifications are (vram, clock, core number, architecture, codec, etc.)

This should answer most, if not all,
article: https://www.binarytides.com/graphics-card-specs-explained/

to evaluate myself which model is best for me and make an informed choice.

This, at almost all times, comes down to the budget anyone has for GPU. And if there is no budget at all, get the best GPU out there.

I would like to know more specifically which are the most important specifications that are most important in a use and in fields of programming and statistical calculations: - AI, deep learning and neural networks - Machine Learning - Software Engineer.

I guess the GPUs you've looked at thus far, are gaming GPUs, namely AMD Radeon and Nvidia GeForce GPUs, right? If so, then for the tasks you list, i'd actually look towards AMD FirePro (Radeon Pro) or Nvidia Quadro (Quadro RTX) GPU, since those are designed and built exactly for this kind of work you describe.

Downside of FirePro/Quadro is that you can not game on them, then again, they aren't built for gaming, but for actual real work. And they do excel in that as well. While gaming GPUs are best at gaming.

Further readings;
GeFroce vs Quadro: https://www.gpumag.com/nvidia-quadro-vs-geforce/
Latest Quadro lineup: https://www.nvidia.com/en-us/design-visualization/desktop-graphics/
Latest Radeon Pro lineup: https://www.amd.com/en/graphics/workstations

 

[Filll999-]

How about adding some spacing into your wall of text? So that anyone who reads it, doesn't hurt their eyes?



This should answer most, if not all,
article: https://www.binarytides.com/graphics-card-specs-explained/



This, at almost all times, comes down to the budget anyone has for GPU. And if there is no budget at all, get the best GPU out there.



I guess the GPUs you've looked at thus far, are gaming GPUs, namely AMD Radeon and Nvidia GeForce GPUs, right? If so, then for the tasks you list, i'd actually look towards AMD FirePro (Radeon Pro) or Nvidia Quadro (Quadro RTX) GPU, since those are designed and built exactly for this kind of work you describe.

Downside of FirePro/Quadro is that you can not game on them, then again, they aren't built for gaming, but for actual real work. And they do excel in that as well. While gaming GPUs are best at gaming.

Further readings;
GeFroce vs Quadro: https://www.gpumag.com/nvidia-quadro-vs-geforce/
Latest Quadro lineup: https://www.nvidia.com/en-us/design-visualization/desktop-graphics/
Latest Radeon Pro lineup: https://www.amd.com/en/graphics/workstations

I will keep it in mind😅
However I would feel safer buying a consumer gpu like GeForce 3000, Radeon 7000 or Intel arc:
1)Because from what I understand Titan, Quadro and FireGL really make the difference compared to "gaming" gpu in AI work and machine learning but very specific, while carrying out more general and if we want less complex jobs, the performance gap between the framework and GeForce should flatten, this from what I was able to understand.
2) I don't really like the idea of buying so old gpu's. Having an i5-13600k I was quite inclined to take an Intel arc, and I was evaluating only if 8gn of memory probably for the management of more advanced models and for the statistical calculations of machine learning, AI and Deep learning is not enough, in which case should I take the A770 model with 16 gb of vram and not 8

 

[Aeacus]

I don't really like the idea of buying so old gpu's.

Which ones are the "old" GPUs that you talk about?
Nvidia Quadro RTX A5000/A5500/A6000 are Ampere architecture, same as GeForce RTX 30-series (e. RTX 3080/3090), that you plan to buy.
While Nvidia Quadro RTX 6000 is Ada Lovelace architecture, same as GeForce RTX 40-series (e.g RTX 4080/4090), making the GPU "too new?".

In a nutshell:
GeForce GPUs - jack of all trades, master of one (gaming)
Quadro GPUs - jack of all trades, master of all but one (everything else, except gaming)

and if we want less complex jobs, the performance gap between the framework and GeForce should flatten, this from what I was able to understand.

There aren't much (if any) studies about overall performance between Quadro and GeForce, since when Quadro is bought, it is put into service (usually there is a dire need) and there isn't time to test the Quadro against GeForce. Especially since both GPUs are designed for completely different tasks and thus, aren't comparable in a meaningful way.

On an analogy; it's like comparing passenger car with semi-truck. Passenger car has good acceleration and is very maneuverable, while semi-truck has loads of pulling power and can haul loads of stuff in one go. And based on your needs, what you need, is semi-truck (aka Quadro) for moving loads of stuff (making calculations), rather than passenger car (GeForce) for entertainment (gaming). Sure, you can haul stuff with passenger car as well, but it would take far longer to move the same amount of stuff, that one semi-truck can in one go. And yes, you can hook trailer to passenger car to increase the load capacity and flatten the gap between semi-truck carrying capacity, but it still isn't comparable.

Due to this, i think that Quadro is better for your needs. And if you still need some entertainment, you can always buy 2nd PC, just for gaming.

In the end, it's your money and when gaming aspect is crucial, look towards RTX 4090, since it's the best gaming GPU currently on the market.