Bojji
Member
You mean that chart?
33% and 35% in apples to apples comparisons is a software patch?
One with RT and the second one with RT and frame gen. Zero raster tests so far.
You mean that chart?
33% and 35% in apples to apples comparisons is a software patch?
It's still an apples to apples test, unless you think for some bizarre reason, regular frame gen on the RTX 50 works a lot better? Far Cry's RT is weak as hell and wouldn't explain that disparity. Either this chart is made up or it isn't.One with RT and the second one with RT and frame gen. Zero raster tests so far.
Right . But expecting the performance in resta to be lower than the one with RT. Simply because the newer RT core is working with the resta power just to gain this advantage over older RT cores.One with RT and the second one with RT and frame gen. Zero raster tests so far.
It's still an apples to apples test, unless you think for some bizarre reason, regular frame gen on the RTX 50 works a lot better? Far Cry's RT is weak as hell and wouldn't explain that disparity. Either this chart is made up or it isn't.
Right . But expecting the performance in resta to be lower than the one with RT. Simply because the newer RT core is working with the resta power just to gain this advantage over older RT cores.
If the newer RT cores and resta power can only do that much combined, then resta in theory should be lower . But who knows
The hit would have to be comically higher to balloon up to a 35% advantage. Based on those apples to apples charts at least, the 5080 certainly is faster than the 4080. Not Turing to Ampere levels or even Ampere to Blackwell, but people have been suggesting the gains will be less than 5% and that seems impossible.DLSS3 frame gen has performance hit, you could have 60fps without FG but when FG is enabled it doubles from ~50FPS to 100FPS for example (so base FPS drops). I think there i a chance that this performance hit might be lower on Blackwell? Or not, we don't know but that would skew results for sure.
RT is indeed improved on the 5000 series.You mean that chart?
33% and 35% in apples to apples comparisons is a software patch?
What data? Where is there a raster benchmark with verifiable and credible data? I don’t think it has improved by 50%, but it most certainly isn’t a "software patch".RT is indeed improved on the 5000 series.
Raster performance is unlikely to be much better as per the data.
He's referring to the cu counts. Those have barely increased, so any kind of noteworthy uplift (if any) has to come from the architecture itself.What data? Where is there a raster benchmark with verifiable and credible data? I don’t think it has improved by 50%, but it most certainly isn’t a "software patch".
The hit would have to be comically higher to balloon up to a 35% advantage. Based on those apples to apples charts at least, the 5080 certainly is faster than the 4080. Not Turing to Ampere levels or even Ampere to Blackwell, but people have been suggesting the gains will be less than 5% and that seems impossible.
I think we’re looking at around a 20% performance advantage in general going from a 4080>5080, not 30-35%. Those games are likely strong performers on Blackwell.
That would mean a titanic performance increase in ray tracing. FC6 has a 21% performance penalty when toggling RT on on a 4080. If the 5080 and 4080 are in the same ballpark with RT off, but the 5080 suddenly becomes 25% more performant with RT on, then we’re looking at the biggest gen-on-gen rt performance increase so far.Combine it with RT teraflop increase.
We don't have even one pure raster bench from nvidia so no wonder people are not confident about performance jump. Most games still are raster only (plus software lumen UE5 games).
That would mean a titanic performance increase in ray tracing. FC6 has a 21% performance penalty when toggling RT on on a 4080. If the 5080 and 4080 are in the same ballpark with RT off, but the 5080 suddenly becomes 25% more performant with RT on, then we’re looking at the biggest gen-on-gen rt performance increase so far.
I doubt though if the numbers of BMW are anything to go by. The 5090 should be wiping the floor with the 4090 by well over 60%, not just 30-40%.
Yeah, so wait for reviews instead of claiming the 5080 won’t be an improvement over the 4080. It won’t be a big one, but they won’t be equal.We know nothing without independent tests. Nvidia focuses on MFG and don't even show BW TF numbers (when they show them for Ada):
Yeah, so wait for reviews instead of claiming the 5080 won’t be an improvement over the 4080. It won’t be a big one, but they won’t be equal.
Based on the core count it's going to be a small improvement. And probably why they're releasing 5080 reviews on release date only.Yeah, so wait for reviews instead of claiming the 5080 won’t be an improvement over the 4080. It won’t be a big one, but they won’t be equal.
That Plague tale note is odd. FP8 vs FP4. Might be that DLSS runs faster on 5000 series with the new Transformer model due to that. Unless I misunderstood what they meant by flux.It's still an apples to apples test, unless you think for some bizarre reason, regular frame gen on the RTX 50 works a lot better? Far Cry's RT is weak as hell and wouldn't explain that disparity. Either this chart is made up or it isn't.
Not only that, memory bandwidth has taken a rather large jump upwards. Games that are heavily memory bound can see a larger than expected boost just due to that.He's referring to the cu counts. Those have barely increased, so any kind of noteworthy uplift (if any) has to come from the architecture itself.
Every gpu ever released had rnd. It was always a costThat huge and profitable high end gpu market that AMD and Intel can't survive in? Even if it's just that they make shit cards that can't compete, Nvidia and the AIB vendors sure do leave a lot of money on the table by producing so few of those cards.
Even if the cards only have $400-$500 of hardware in them, you can't just ignore the R&D and marketing costs, your thoughts on owners of those spending more on games and DLC is just pure conjecture.
yeah and he is obviously exaggeratingI've been listening to Jeff for 20 years. I'm not going to say he's wrong, but the degree to which he's implying the additional latency impacts his ability to perform in game compared to his skill level relative to competitive fighting game and FPS players seems out of whack. He's always been a bit above average at both of those genres. Not to say he needs to "deal with it" or anything. I just think the latency introduced by FG is one of those things that some people are going to feel and others aren't. The people who don't feel it just aren't sensitive to it and probably aren't playing a lot of competitive multiplayer games.
Exactly . That’s why we don’t have any information about resta performance. But I would assume if it was more than 30%, they would have at least one graph. But nope nothing.DLSS3 frame gen has performance hit, you could have 60fps without FG but when FG is enabled it doubles from ~50FPS to 100FPS for example (so base FPS drops). I think there i a chance that this performance hit might be lower on Blackwell? Or not, we don't know but that would skew results for sure.
RT teraflops performance went significantly up (like over 30% between 4070tiS and 5070ti) so it should boost hybrid RT games to some extend. Meanwhile raster TFs are closer.
Every gpu ever released had rnd. It was always a cost
Yeah I understand. Gpu is more than bill of materialsI never implied they didn't, I was saying just build cost of a board isn't the actual total cost so counting that margin as profit isn't accurate.
It's possible to calculate flops based on the available data.We know nothing without independent tests. Nvidia focuses on MFG and don't even show BW TF numbers (when they show them for Ada):
I think it depends on the use case. For gaming,, seems about right. For AI workloads I can see much bigger improvements due to 32GB VRAM, 512bit bus and increase in Tensor cores.When the actual hardware (based on CUDA/shader counts) has such limited improvement over the predecessor, yes
NVIDIA's own charts show the 5080 is a software patch over the 4080.
5090 is likely to be 25-30% faster, but also a 28.9% increase in total draw power.
If you're getting 25-35% more performance for 25-35% more money, what the fuck else is it?
Clickbait or not, it's difficult to argue with the raw figures.
Overall the 5000 series has been a sideways refresh with a software patch. The 5090 has an increase in performance - with increased power consumption and total price. Rest of the cards? Raster performance is no better than 2 year old GPU.
If you look at their graphs, that is what they're claiming for RT performance improvement.isnt that number only referring to the raster perfomance? thats not really where im focused.
Yes, MFG (FGx4) was used on all RTX50 series cards in their benchmarks.so in a raytracing heavy game the 5080 would double my performance? seems like a bargain does that test use framegen?
Impressive
Yes that's fine. 12VHPWR and 12v-2x6 are interchangeableAlright guys, PSUs confuse the hell out of me. This is what I have:
COMBAT FG 850W ATX 3.1 | SUPER FLOWER
Please contact the distributor in your country for the availability and quotationwww.super-flower.com.tw
For the 5080, will I just use the 12vhp that came with the PSU? I believe it's a 450w cable.
Some more sildes leak
Glad to see those nativists losing the resolution war.
Glad to see those nativists losing the resolution war.
its around 15% according to slide.That 5070 vs 4070 don't make much sense...
its around 15% according to slide.
Maybe a typo on 3d centre's end
With NO framegeneration on
60-65 fps on 5090 at 4K DLSS quality Path tracing.
How much does 4090 Get?
I'm sure I'll have a hard time finding the small fits in my case founders edition.Am I crazy in thinking these cards might not be THAT hard to get at launch?
Am I crazy in thinking these cards might not be THAT hard to get at launch?
- No bitcoin mining craze to deal with
- No COVID related supply chain issues to deal with
- Cards are at really high prices and more and more people are struggling
- General enthusiasm around the reveal has been lukewarm
- Switch 2 announcement may steal some of the thunder
I'm sure the initial batch will still sell out, but I wonder if they'll sell out in an hour or two this time around instead of the first 3 minutes.
Hard to know. I mean wasn't the deal in the 40 series was that you could get them as long as you weren't trying to get the 4090? (They were so plentiful that tech tubbers started claiming that the 4070 was a flop. Of course tech epiphany and Steam GPU survey suggest otherwise.)Am I crazy in thinking these cards might not be THAT hard to get at launch?
- No bitcoin mining craze to deal with
- No COVID related supply chain issues to deal with
- Cards are at really high prices and more and more people are struggling
- General enthusiasm around the reveal has been lukewarm
- Switch 2 announcement may steal some of the thunder
I'm sure the initial batch will still sell out, but I wonder if they'll sell out in an hour or two this time around instead of the first 3 minutes.
Glad to see those nativists losing the resolution war.
I guarantee you the RTX 5090 will fly off the shelves. The first and only 32gb GeForce card and the best performance money can buy.Am I crazy in thinking these cards might not be THAT hard to get at launch?
- No bitcoin mining craze to deal with
- No COVID related supply chain issues to deal with
- Cards are at really high prices and more and more people are struggling
- General enthusiasm around the reveal has been lukewarm
- Switch 2 announcement may steal some of the thunder
I'm sure the initial batch will still sell out, but I wonder if they'll sell out in an hour or two this time around instead of the first 3 minutes.