Well, the people want it when it makes things look really good, but also only when it can actually be done with the hardware that exists and not faked so horribly....
Well, the people want it when it makes things look really good, but also only when it can actually be done with the hardware that exists and not faked so horribly....All this upscaling garbage is because of ray tracing, no other reason. And virtually every internet tech poll shows the majority of people don't want ray tracing.
So here's my take on this refresh and I expect the 4080 Super to be a dud too based on it having nothing more to offer than a few extra CUDA cores.
Nothing still comes close to the 4090 I've had for over a year now and it's a weird thing the halo card is the best value of the whole 4000 series. This "Super" refresh is just disappointing all the way around. The 4070 is decent, but I still expect a 256bit memory bus and 16GB+ of VRAM for $600+ video cards. Even the 4090 is a bit disappointing in that its 72MB cache is holding back an additional 15%+ performance at 1440p and lower resolutions.
The intentional nerfing of the current generation 4000 series to sell future products is strong here, just like it was with the low cache, low VRAM 3000 series of cards. Going back to the 2000 series, nVidia offered almost no performance increases but added RT, which was too weak in that generation to be usable. In summary, all I see is feature additions, large price increases, and not much to show with regard to actual "real" (aka raster) performance increases.
MLID says that his sales contacts have only gotten inquiries for the 4080 Super and not for the other cards. He expect a short term surge by people who are done waiting for the 4090 to come down in price.So here's my take on this refresh and I expect the 4080 Super to be a dud too based on it having nothing more to offer than a few extra CUDA cores.
The 3090 was a horrible value. The 3080 and the 3080 Ti were within spitting distance of its performance. The 4090 is a solid 50% faster than the 4080 in comparison. Even considering the price adjustment for the 4080 Super, they're about equal values given you're getting 50% less memory bandwidth and 60% less CUDA cores for basically 60% less cash. Not to mention the 4090 has been available for over a year.The best part of the 4080 Super "refresh" is going to be the price adjustment to $999 if that sticks, IMO. The performance uplift is sort of a bonus on a long overdue pricing adjust for a clear "2nd Place" card by a big measure. I assume this will, in turn, push the 7900 XTX to ~$949 or less as its a poor head to head match up, IMO.
Just like the 3090, the 4090 at launch was the best "value" of the entire family if you really were looking for leading edge performance.
The 3090 was a horrible value. The 3080 and the 3080 Ti were within spitting distance of its performance. The 4090 is a solid 50% faster than the 4080 in comparison. Even considering the price adjustment for the 4080 Super, they're about equal values given you're getting 50% less memory bandwidth and 60% less CUDA cores for basically 60% less cash. Not to mention the 4090 has been available for over a year.
Even considering the price adjustment for the 4080 Super, they're about equal values given you're getting 50% less memory bandwidth and 60% less CUDA cores for basically 60% less cash.
Yeah, I got it new for $1600 back in December 2022. So far so good. I do agree the 4090s available at my local Micro Center are $1850 and $2000, and at those prices the value calculations I'm using get altered drastically. The whole product stack across the board needs to get about a 30% price shrink to get back to pre-COVID shortage prices. That's a pipe dream right now since AI is here to replace COVID, Mining, etc. as the new reason to price gouge though.I suppose you aren't wrong - it just had the biggest window at MSRP IIRC. If you bought one before ETH and Covid spiked the price, you looked unexpectedly brilliant as all GPUs went to megabucks. And it was unabashedly the fastest card at that time, especially for aspiring 4K gamers. That's a lot of context I didn't mention.
Not only is the 4090 a year old, but it also costs significantly more now that it MSRP'd for at launch due to non-gaming demand.
You made a great purchase so long as it causes you no problems.
Very true now, as the performance gap is more like 30-35%. As faster CPUs get released, I expect that gap to grow. That said, there's nothing except maybe Cyberpunk 2077 with RT turned on that's not getting 144+ FPS (limit of my current monitor) when paired with my 7800X3D, so I feel like I'm golden for a good while without any upgrades.Performance doesn't scale linearly with the number of CUDA cores, though. Especially that is in part also because the 4090 is so fast that in many games the CPU can't keep up.
I think you mean a 40-50% price shrink to get back to pre-Covid prices. The top dog cards then were only $1000-1100 out the door (and were actually MSRP or less). If you look at the history, pre-Covid, for 3-4 years prior, the average price of a GPU sold/purchased was between $250-340, and that aligned very closely to between a xx60 and xx70 product from Nvidia, so not the top cards, but also not the bottom of the stack (with the stack having xx50 and even xx40 or xx30 cards).Yeah, I got it new for $1600 back in December 2022. So far so good. I do agree the 4090s available at my local Micro Center are $1850 and $2000, and at those prices the value calculations I'm using get altered drastically. The whole product stack across the board needs to get about a 30% price shrink to get back to pre-COVID shortage prices. That's a pipe dream right now since AI is here to replace COVID, Mining, etc. as the new reason to price gouge though.
Yeah I may be jumping on the 4080 Super if I can manage to get one from Best Buy or Nvidia for $999. It seems like it would be a nice upgrade from my 3070. And more importantly the old 3070 will be a nice upgrade for my wife's rig which still has a 1070!The best part of the 4080 Super "refresh" is going to be the price adjustment to $999 if that sticks, IMO. The performance uplift is sort of a bonus on a long overdue pricing adjust for a clear "2nd Place" card by a big measure. I assume this will, in turn, push the 7900 XTX to ~$949 or less as its a poor head to head match up, IMO.
Just like the 3090, the 4090 at launch was the best "value" of the entire family if you really were looking for leading edge performance.
And you can talk to her all about RTX and she will know exactly what you mean!And more importantly the old 3070 will be a nice upgrade for my wife's rig which still has a 1070!
The 4080 Super should be a major upgrade. So will the 3070. Very nice.Yeah I may be jumping on the 4080 Super if I can manage to get one from Best Buy or Nvidia for $999. It seems like it would be a nice upgrade from my 3070. And more importantly the old 3070 will be a nice upgrade for my wife's rig which still has a 1070!
How do you know? GDDR7 could provide a decent benefit at a cheaper price than GDDR6X.Blackwell will do the bare minimum in price/perf
If AMD actually makes 3+ Ghz work, then AMD could force Nvidia prices down. Or we can go for AMD cards.The <$500 market should actually be a nice perf/$ bump for once, assuming AMD doesn't miss twice in a row. Beyond that, there won't be a meaningful leap forward from AD102 perf until 2026.
I think you mean a 40-50% price shrink to get back to pre-Covid prices. The top dog cards then were only $1000-1100 out the door (and were actually MSRP or less). If you look at the history, pre-Covid, for 3-4 years prior, the average price of a GPU sold/purchased was between $250-340, and that aligned very closely to between a xx60 and xx70 product from Nvidia, so not the top cards, but also not the bottom of the stack (with the stack having xx50 and even xx40 or xx30 cards).
How do you know? GDDR7 could provide a decent benefit at a cheaper price than GDDR6X.
Kind of have to assume GDDR7 is going to be decently more expensive. The faster speeds & 3 GB is just begging NV to cut bus width.
GDDR7 will be more expensive, how much is still up in the air. Of course any chance to trim down analog area on a new node is a win, but NV will probably take it too far, top end should still be 384-bit but will they give that to client straight up?How do you know? GDDR7 could provide a decent benefit at a cheaper price than GDDR6X.
Ampere had competition, and Samsung N8 is really cheap so NV could price competitively while maintaining margins, so long as enough people bought the 3090.And in the past we've seen Nvidia react to a bad generation by giving more to the next gen (2000-series vs 3000).
You can always go for AMD cards, they are only completely lost in stuff that only 1% of buyers use.If AMD actually makes 3+ Ghz work, then AMD could force Nvidia prices down. Or we can go for AMD cards.
This is the scenario if RDNA4 does its job. If it doesn't, then NV could do some very disgusting segmentation.More like a recipe for keeping bus width the same. If there were 3GB chips available, the current 60 cards could be 12 GB, and 70 cards could be 18GB, without resorting to clamshell designs. Which seems about where people wanted them.
This is the scenario if RDNA4 does its job. If it doesn't, then NV could do some very disgusting segmentation.
Remember GDDR7 baseline is 2GB chips, 3GB is a premium product. High end could be really nasty, similar to Ampere.
RDNA4 is in the lab, has been for some time considering expected Computex announcement and availability soon after.Design will be set long before you can know if RDNA4 "Did it's job".
There is more than one way to skin a cat, AMD will never win by simply copying NV verbatim.The job AMD needs to do is software more than hardware. DLSS was a major deciding factor for me. AMD needs competitive neural network scaling. It's embarrassing, that even Intel XeSS is better than FSR scaling.
There is more than one way to skin a cat, AMD will never win by simply copying NV verbatim.
FSR2 can work nearly as well, it just requires careful integration from the ground up, fully replacing the native AA.
This is why we can't have nice things. Just as legions of gamers could not resist supporting scalpers, they cannot resist supporting the company that is doing precisely what you describe in this quote.Just don't reward a company with obscene greed and dominance for doing the bare minimum when they don't care the slightest in the market right now.
Who says that an NN approach is the only way to improve from a handwritten algo?They are certainly going to lose making third rate implementations behind both NVidia and Intel. If Intel gets it's HW in order before AMD catches up on SW, it's going to be rough for AMD.
An algorithmic approach simply can't compete to the NN approach for this kind of problem.
IMO, it's only a question of when, not if AMD will switch it's scaling to a NN approach. The longer the delay the more embarrassing it is.
Until then, they will just have a third rate solution, relegating them to only the AMD fan market.
Who says that an NN approach is the only way to improve from a handwritten algo?