Hi all,
I’m currently on the lookout for new workstations and bumped into a wall, here… I was presented with the option of getting a NVidia RTX A4000 16GB or a more familiar RTX 38080 10GB for basically the same price.
What do you guys think would be the better option? I’m inclined towards the A4000 (less CUDA cores, more VRAM). What do you think?
Tks!
Hello,
The 2 GPUs shouldn’t be the same price, one of them is either on a good deal or overpriced
a 3080 should be between 800 to 900 USD, the A4000 will cost at least 1200 USD
I would go with A4000 personally, it runs silently because of 140 W TDP and occupies a single slot in width. And you have 6 more GB of memory
But yeah the 3080 is faster(Depending on the brand) it could be 30-40% faster, if your scenes are not big and you prefer performance it is not a bad choice
Best,
Muhammed
Exactly my thoughts. Tks, Muhammed!
Back with more questions, lol… Our IT guy said Dell is trying to sell us a new server that would feature a Nvidia A100 with 40GB vRAM (10vGPU), claiming each node would have 200GB available to net render using V-Ray GPU. I’m not a hardware expert, but this sounds a bit fishy, doesn’t it?
It looks like this is the card in question: NVIDIA A100 PCIe 40 GB Specs | TechPowerUp GPU Database
Notice that it has “DirectX: N/A”, so Vantage won’t work (since this is the Vantage forum, even though you spoke about V-Ray GPU). Vantage also doesn’t share the memory pool. But V-Ray GPU can, with nvlink. Did you mean “10x GPU” when you wrote “10vGPU”? 10x 40GB would total 400GB, not 200. This assumes the server has one of those fancy nvlink switches for full connectivity. Normal desktops can only nvlink pairs of cards. Not sure if someone has tried something as crazy as 10x A100 with V-Ray GPU 
It is true, could even go higher than 200 GB of GPU memory.. for the price of 180,000 USD per node 
Bingo! Lol… I knew there was a catch somewhere. Since I’m not dealing with Dell directly, it’s always bits of information here and there…
Tks!
That’s the thing… Dell is probably trying to sell as much as they can, hence their offer… I don’t even think it makes sense, tbh… I mean, it’s A LOT of money for a system like that…
Wait for the 4090 coming end of summer.
Unfortunately we can’t wait for those to be available… 
anyone know the prices of 4090’s? i’m thinking i’d stick with 3080ti hybrids.
USD 1,699.00, afaik.
well, was going to do the 3080’s but now we decided to get the 4080 or 90. i think 4080s are just about same performance as 90’s but less expensive. they just did their unveiling a couple days ago. i can only fit 2 of them on my mobo. could do 2 more in another, identical machine tho.
i think 4080s are just about same performance as 90’s but less expensive.
Based on the core count mentioned on nVidia’s site, this doesn’t seem to be true this time around, unfortunately.
The gap between 4080 and 4090 seems huge, unlike the gap between 3080 and 3090.
But until the GPUs are out, nothing can be certain.
Greetings,
Vladimir Nedev
Pretty sure the cuda core count is hugely different…the 4090 will be a lot faster…and you also have to be careful about which 4080 youre talking about too, since Nvidia have been sly by calling a stripped down card also a 4080 with less memory, but it also has fewer cuda cores…a 4070 in disguise basically.
ah, interesting. i’ll be sure to check that out more.
The specs between the announced 4080s are quite different, the 1200 USD model has 16 GB and more Cuda cores. It will be faster for sure than the 900 USD model with 12 GB and less Cuda cores
It is a bigger gap in specs than the 3080 and 3090, I think calling both cards 4080s is not helpful. It should have been 4080 and 4070 or 4080/4080Ti
Like what Vlado Nedev said, we will know more when the benchmarks for these cards are out soon
Best,
Muhammed