• Happy holidays, folks! Thanks to each and every one of you for being part of the Tom's Hardware community!

[SOLVED] which Graphics Card for heavy workloads

BearTech

Distinguished
Jan 21, 2015
69
0
18,660
Hey!

I'm just in the time of upgrading my computer with a new graphics card.

Computer today:
2700X water-cooled
16GB Frame (will upgrade to 32GB soon)
1060 - 6gb
ASUS ROG STRIX X470-F GAMING
EVGA G2 850W

My primary workloads:
3D Modeling, Rendering, Game Development in VR on Valve Index and Games (VR and "Regular Games")

I don't have the Valve Index yet but I think I need an upgrade to make my daily work go faster since I also use this computer in my visualization business.
There have also been some rumours of a 2080 TI Super, so is it worth waiting?

I have looked at some different graphics cards and prices and this is what I find:
2080 TI for 12000 SEK
2080S for about SEK 8500
2080 for just under SEK 8000

That's a 32% performance difference on the 2080 TI vs 2080S with a price difference of about 30%. While between 2080S vs 2080 there is a 7% difference with a price difference of about 10%. Data were taken from UserBenchmark.

I'm really unsure which card will bring me the most value. So I start this thread to discuss this.

Thanks in advance.
 
Last edited:
Solution
No, the fastest video card on the market is best for your use case...Comes down to your costs and how much an improvement a reduction in rendering time will make to you. If you use this professionally, then your output will go up. If you use it casually, you will still be able to get more done, or be able to use use your computer for something else more often.

I did a little research on Cycles, couldn't find anything that indicates memory speed is all that important. Memory amount yes. But that doesn't mean the memory speed has no impact on the overall performance of a GPU during renders. Would have to find some better sources on that one.

Nvidia is still using TSMC's 12nm process node for their GPUs. AMD already made the jump to...
Depends on what your time is worth to you, I suppose.

Performance/Cost always goes down as you go up in performance. Usually this makes the best deal somewhere around the 70 series cards. But if you don't think that is enough, then you pay the premium.

How do your workloads respond to memory speed? That would be one reason to avoid the non-super cards. I suspect any potential 2080Ti Super will just have the memory speed increase. Doubt they will pull out the stops and go with fully functional silicon.
 
Depends on what your time is worth to you, I suppose.

Performance/Cost always goes down as you go up in performance. Usually this makes the best deal somewhere around the 70 series cards. But if you don't think that is enough, then you pay the premium.

How do your workloads respond to memory speed? That would be one reason to avoid the non-super cards. I suspect any potential 2080Ti Super will just have the memory speed increase. Doubt they will pull out the stops and go with fully functional silicon.

To be honest I don't know how my workloads respond to memory speed. I use Cycles in Bender as my main render engine.

So 2070 would be better for my use case? Are there any rumour on if Nvidia or AMD is releasing their next-gen, and is it worth waiting?
 
No, the fastest video card on the market is best for your use case...Comes down to your costs and how much an improvement a reduction in rendering time will make to you. If you use this professionally, then your output will go up. If you use it casually, you will still be able to get more done, or be able to use use your computer for something else more often.

I did a little research on Cycles, couldn't find anything that indicates memory speed is all that important. Memory amount yes. But that doesn't mean the memory speed has no impact on the overall performance of a GPU during renders. Would have to find some better sources on that one.

Nvidia is still using TSMC's 12nm process node for their GPUs. AMD already made the jump to TSMC's 7nm process. It will certainly be interesting to see what their architecture can do with a shrink. Given the size of the GPUs they can make now, you might see 6000 CUDA cores or so on their big chips, if they even go down that route. Aside from compute, not sure there is much of a market, or if just getting the same core count at a lower power envelope is more desirable.

Old news says they'll have something new in the first half of 2020, but that doesn't mean reasonably priced consumer graphics cards will be amongst them.

AMD is in a similar state, near as I can tell. They are filling out the bottom of their stack now, with 5700XT (roughly a 2070 Super) being their top card at the moment. With the expectation of a release in mid 2020 of Navi 23. So bigger versions of the designs they have already.

It is never worth waiting. Buy when you have a need, regret nothing. If you keep a regular cycle, just buy when the time is right. I would only recommend waiting when release dates have at least been soft announced.

Contrary to what I said earlier looks like performance per dollar rises in the Nvidia stack, might be some local sales skewing the numbers though. But the most bang for the buck is the cheapest 2080Ti, but only if you can afford it. 2080 Super isn't a bad deal either.
 
Solution
No, the fastest video card on the market is best for your use case...Comes down to your costs and how much an improvement a reduction in rendering time will make to you. If you use this professionally, then your output will go up. If you use it casually, you will still be able to get more done, or be able to use use your computer for something else more often.

I did a little research on Cycles, couldn't find anything that indicates memory speed is all that important. Memory amount yes. But that doesn't mean the memory speed has no impact on the overall performance of a GPU during renders. Would have to find some better sources on that one.

Nvidia is still using TSMC's 12nm process node for their GPUs. AMD already made the jump to TSMC's 7nm process. It will certainly be interesting to see what their architecture can do with a shrink. Given the size of the GPUs they can make now, you might see 6000 CUDA cores or so on their big chips, if they even go down that route. Aside from compute, not sure there is much of a market, or if just getting the same core count at a lower power envelope is more desirable.

Old news says they'll have something new in the first half of 2020, but that doesn't mean reasonably priced consumer graphics cards will be amongst them.

AMD is in a similar state, near as I can tell. They are filling out the bottom of their stack now, with 5700XT (roughly a 2070 Super) being their top card at the moment. With the expectation of a release in mid 2020 of Navi 23. So bigger versions of the designs they have already.

It is never worth waiting. Buy when you have a need, regret nothing. If you keep a regular cycle, just buy when the time is right. I would only recommend waiting when release dates have at least been soft announced.

Contrary to what I said earlier looks like performance per dollar rises in the Nvidia stack, might be some local sales skewing the numbers though. But the most bang for the buck is the cheapest 2080Ti, but only if you can afford it. 2080 Super isn't a bad deal either.
Thank you so much for setting aside time to answer my questions! I'll probably see if I can find a good deal on Black Friday on a 2080 TI.