Looks like a great gpu at a bad price. The 4090 takes everyone that can just about afford $1600 to those with millions (a sizeable market). The market below that of those that can only afford $1200 but not $1600 must be tiny - surely Nvidia needed this card to be at most $1000 to sell?
It's a great card at an abhorrent price. It's one of those releases that had it not been for the MSRP, it would have obliterated the market and outsold everything else in this segment. The mind boggles. I truly don't understand the reasoning here. Maybe the yields are simply not there and this is the equilibrium. Because greed doesn't explain it - the demand in the market is there, and it's always better to sell more.
So it peaks at around 320W and 61C while being quiet enough to be background noise. So... why did it come with a 3-slot heatsink? It just unnecessarily limits who would buy this, both in terms of whose PC it'll fit into and in terms of added cost (shipping being the bigger issue, not materials). The price point is already hard to justify, so why would they give people more reasons to not buy it?
Also kind of interesting how you can really see where the 4090 gets CPU bottlenecked.
I'm sorry, but how are we supposed to read this diagram? I see three values for each card, one probably is the max value of the samples per minute. But what about the other two?
Ok that we are in tough times, but adding some letters doesn't increase the consume of ink in this case.
Also, Open CL isn't supported anymore by Blender since version 3.0: it's time to update your bench test and separate Cuda results from Optix results, because they are different things.
Jensen may be a good engineer, but he's terrible at assessing product value. It will go under $1k when reality sinks in, as it did when he unlaunched the 4080 12gb.
My FTW3 was almost a grand after taxes and shipping.
I planned ahead and set a decent budget for this gen, I stayed realistic.
The same higher level perf card comparable to a FTW is going to be $1400+ this go around. The price is way off right now. Hopefully AMD's 7900 cards are good and push these prices down.
So it peaks at around 320W and 61C while being quiet enough to be background noise. So... why did it come with a 3-slot heatsink? It just unnecessarily limits who would buy this, both in terms of whose PC it'll fit into and in terms of added cost (shipping being the bigger issue, not materials). The price point is already hard to justify, so why would they give people more reasons to not buy it?
Also kind of interesting how you can really see where the 4090 gets CPU bottlenecked.
I think AMD is nailing how to do this with the 7900 cards. They are smaller cards not pushed extremely hard so cooling can be less. However, for AIB cards I do like the larger heat sinks. The MSI 4080 cards are coming in under 30 db's at load which is pretty nice for those of us that like near silent computers however having no 2 slot options is very limiting. I really feel like Nvidia's choices on pushing the stock cards hard and the pricing are pretty terrible but otherwise they are nice cards.
It's a great card at an abhorrent price. It's one of those releases that had it not been for the MSRP, it would have obliterated the market and outsold everything else in this segment. The mind boggles. I truly don't understand the reasoning here. Maybe the yields are simply not there and this is the equilibrium. Because greed doesn't explain it - the demand in the market is there, and it's always better to sell more.
Maybe they don't want to sell that many just yet as they are trying to unload the old stock? That is all I can think right now. If that's true then we should see a big drop in December.
So it peaks at around 320W and 61C while being quiet enough to be background noise. So... why did it come with a 3-slot heatsink? It just unnecessarily limits who would buy this, both in terms of whose PC it'll fit into and in terms of added cost (shipping being the bigger issue, not materials). The price point is already hard to justify, so why would they give people more reasons to not buy it?
Also kind of interesting how you can really see where the 4090 gets CPU bottlenecked.
The massive cooler on the 4090 is nice, once you've managed to squeeze it into your case. I really appreciate the noise improvement over my previous 3080ti. But I get that some people might prefer a smaller, if noisier card.
I'll say it again, not that it's needed. Good card. Bad price.
If the cheapest 4090 in the UK i can see is £1800, the 4080 will probably be £1500 and upwards.
If the 7900xtx hits at £1200 its a winner winner chicken dinner (as long as Raster perf is +/> 4080)
For me it feels like the performance between 4090 and 4080 is too big. Even for a 4080ti to fit in later. The 4080 feels more like a 4070ti. I don't see any reason why someone would pick the 4080 over the 4090 when the price is so close, they use the same power connectors and hold the same risks. If the 4080 was $799 I'd feel differently. But seeing some 4080's with extra cooling coming close to generic 4090 pricing...this is just absurd. Nice review Hilbert.
The massive cooler on the 4090 is nice, once you've managed to squeeze it into your case. I really appreciate the noise improvement over my previous 3080ti. But I get that some people might prefer a smaller, if noisier card.
I'll say it again, not that it's needed. Good card. Bad price.
For a 4090, the massive cooler is a requirement because it gets massively hot. The 4080 has a pretty reasonable wattage. Considering how cool it runs, you could probably do a dual slot heatsink, keep the fan speeds the same, and it'll still be far enough from thermal throttling without any additional noise.
I'm sorry, but how are we supposed to read this diagram? I see three values for each card, one probably is the max value of the samples per minute. But what about the other two? Ok that we are in tough times, but adding some letters doesn't increase the consume of ink in this case. Also, Open CL isn't supported anymore by Blender since version 3.0: it's time to update your bench test and separate Cuda results from Optix results, because they are different things.