If TechPowerUp's rumour is accurate we're looking at $2k for the rumoured 3090..
You already pay 2.5K for the current titan rtx and my personal opinion is that the titan is going to be 'rebranded' as the 3090 (the titan has 24GB memory and the 3090 seems like it might as well) so there is already a (stupid) pricing precedent. Why do I think it's a rebranding exercise, basically they're running out of names and going number based is easier.
Do I agree with that level of pricing, hell no..... even as the 'half way house' between geforce and quadro it doesn't warrant that price imo, unfortunately it's a 'relative' bargain for professionals so it likely sells pretty well none the less.
To be fair even that has sort of blurred now with nvidia studio drivers, while not supporting some of the programs that force you to use a quadro (by hard coding, for no reason in most cases) they do support some of the ones that the quadro drivers are 'optimised' for... not that I've had any issues with normal geforce drivers outside of the programs that force you to use a quadro (they still work in most cases, just 'slower')
I think it'll actually be cheaper than last Titan. AMD have made them a bit cautious, so rather than having out of orbit pricing on what would have been the titan, they're releasing it as a 3090 in order to say "we still beat AMD in the 'regular' range", while allowing them to charge more than a 3080. Progress, of sorts.
Honestly, as I said earlier, I don't think there will be a titan this time, if ever again, it's already a confusing mess trying to work out which version is which lol.
What I suspect is going to happen is we're going to get
GA102-400-A1 with 5,376 CUDA cores and 24 GB of 17 Gbps VRAM - originally planned as a titan, renamed to 3090 priced the same or similar to the current titan around £2.5k, maybe lowered slightly depending on AMD
GA102-300-A1 with 5,248 CUDA cores and 12 GB of 21 Gbps VRAM - originally planned as 3090, renamed to 3080ti - I reckon £1200 range
GA102-200-Kx-A1 with 4,352 CUDA cores and 10 GB of 19 Gbps VRAM - 3080 - £1000 range
I could be wrong mind you lol
This is Nvidia who made the Titan V which was over £2000. There probably will be one - I wouldn't be surprised if the RTX3090 is a salvage part,or is worse in some way. All they are doing is making another tier above the 80TI,hence it now becomes a lower tier gaming part.
Hence the 80TI now drops down the rank - instead of it being a second level salvage,it will be 3rd level.
The current titan rtx is just shy of £2500 and has 24GB of vram....
As said above maybe it's AMD rumours (and I'm sure they've had access to info from partners, both ways) making them be a bit more cautious and/or branding things a little different, nvidia will want to hold onto the 'gaming crown' after all. I wouldn't be shocked if the titan/3090 isn't the 'full chip' either (supposedly ampere is Up to 128 SMs / 8192 GPU cores for GA100) but if rumours/info is true it does have a slower vram throughput which is consistent with current titan.
Realistically the pricing really depends on AMD and I can fully see AMD pricing theirs 'high' as well but just a touch cheaper than nvidia at relative performance.
Nvidia like Apple will find new ways to exploit the market. The Titan V was £3000,and there was already another Pascal Titan underneath it,and that had also more expensive Star Wars Editions.
Even if as you say there is no Titan this year,it doesn't mean we won't have an RTX4090 and a Titan the next generation either. Nvidia is just setting the groundwork to make the 80TI a tier below what its now. After all the 80 series used to be the best GPU they did,but now its been pushed down two tiers,ie,80TI and Titan.
You only need to look at where the mainstream tier now stands. A 60 series GPU used to be the 3rd/4th fastest GPU in the stack. Now the RTX2060 and RTX2060 Super are 7th and 8th down the stack,with the GTX1660 series now covering 9th,10th and 11th places.
So if you look at each generation,the maximum performance jump has been relatively consistent,but its now spread over more and more models.So if you buy via price,then things are getting slowly worse and worse.
worse in terms of bang-per-buck as a % of the top tier card, then yes. But overall performance at a given price point still increases vs previous models that were sold at that price in previous generations. It's not like buying the newer generation at that price gives you worse performance. It just wrankles that instead of being on a decent card it's the relative dregs of the stack that is now affordable for the mainstream.
Actually the performance jumps at the top are relatively consistent for Nvidia. So 70~100% for each node,and around 40~50% when on the same node,but ONLY if you compare the properly aligned models. However,by cost its definitely getting worse and worse. For instance the doubling of performance at a given cost in the mainstream is getting slower IMHO.
It even gets worse when you consider it by ranking relative to the top. Your GTX460 was the 3rd fastest GPU Nvidia did in the Fermi stack after the GTX470 and GTX480. In the Maxwell stack it was the GTX980. In the Pascal stack it was the GTX1080(or you could say the GTX1080TI as there were two Pascal based Titan GPUs with co-existed with each other). In the Turing stack,it is the RTX2080 Super(or the RTX2080 if you go by the original launch lineup).
There are currently 1 users browsing this thread. (0 members and 1 guests)