News RTX 5080 is 22% faster than the RTX 4080 per leaked benchmarks — falls short of the RTX 4090

[Admin]

Geekbench puts the RTX 5080 (up to) 22% faster than its predecessor.

RTX 5080 is 22% faster than the RTX 4080 per leaked benchmarks — falls short of the RTX 4090 : Read more

 

[artk2219]

With half the die being wasted, I really wonder if it wasn't a better idea to have a separate tape out, either that or the yields are horrendous enough to make it worth it.

 

[Eximo]

Just looking at the core count and such it was clear it is an updated replacement for the 4080/4080 Super, not intended as an improvement. They want that massive gap between the 5080 and 5090.

You might convince me to get a 5070 Ti at this point. But the value add for the extra money seems minimal. Just like the 4080 at launch.

 

[TheyStoppedit]

The 50 series is just a 40 series refresh. Everybody already knows this.
The 5080 is really the 4080Ti
The 5070Ti is what the 4080S should have been
The 5070 is what the 4070Ti Super should be

My message to NVidia is this: You give me fake frames, I'll give you fake dollars. The 5090 is 3x price what I can get a 4080S for. Here in Canada, a 5090 will be 3500 Canadian pesos minimum, on the absolute bottom end. I can get a 4080S used for 1200. I would have 60% of 5090 performance for 30% of the price. I can get a 4090 for 1800 CAD easy, and I would have 75% of the performance while only using 70% of the the 5090s power consumption for 50% of the 5090 price.

Bottom line is: The 50 series can only appeal to those who are willing to pay real dollars for fake frames. I'm not one of them and I know most others arent either. I'm sure 60 series will do 6 fake frames instead of just 3 or 4 and they will be like "Oh look, 5090 performance in a 6060 for only $700 USD." Sorry, not happening. I want to be wrong but I'm not holding out much hope

 

[Jame5]

AMD's Zen 5 architecture and features a large 64MB SRAM chiplet atop the CCD for 104MB of total L2+L3 cache

Underneath, not atop. That's half the reason these CPU's do as well as they do. The CCD isn't being smothered by the extra L3 cache.

 

[aero1x]

The 50 series is just a 40 series refresh. Everybody already knows this.
The 5080 is really the 4080Ti
The 5070Ti is what the 4080S should have been
The 5070 is what the 4070Ti Super should be

My message to NVidia is this: You give me fake frames, I'll give you fake dollars. The 5090 is 3x price what I can get a 4080S for. Here in Canada, a 5090 will be 3500 Canadian pesos minimum, on the absolute bottom end. I can get a 4080S used for 1200. I would have 60% of 5090 performance for 30% of the price. I can get a 4090 for 1800 CAD easy, and I would have 75% of the performance while only using 70% of the the 5090s power consumption for 50% of the 5090 price.

Bottom line is: The 50 series can only appeal to those who are willing to pay real dollars for fake frames. I'm not one of them and I know most others arent either. I'm sure 60 series will do 6 fake frames instead of just 3 or 4 and they will be like "Oh look, 5090 performance in a 6060 for only $700 USD." Sorry, not happening. I want to be wrong but I'm not holding out much hope

damn you complain a lot

 

[PaladinNH]

The 50 series is just a 40 series refresh. Everybody already knows this.
The 5080 is really the 4080Ti
The 5070Ti is what the 4080S should have been
The 5070 is what the 4070Ti Super should be

My message to NVidia is this: You give me fake frames, I'll give you fake dollars. The 5090 is 3x price what I can get a 4080S for. Here in Canada, a 5090 will be 3500 Canadian pesos minimum, on the absolute bottom end. I can get a 4080S used for 1200. I would have 60% of 5090 performance for 30% of the price. I can get a 4090 for 1800 CAD easy, and I would have 75% of the performance while only using 70% of the the 5090s power consumption for 50% of the 5090 price.

Bottom line is: The 50 series can only appeal to those who are willing to pay real dollars for fake frames. I'm not one of them and I know most others arent either. I'm sure 60 series will do 6 fake frames instead of just 3 or 4 and they will be like "Oh look, 5090 performance in a 6060 for only $700 USD." Sorry, not happening. I want to be wrong but I'm not holding out much hope

It's a slap in the face to current 40 series owners, for sure, but for everyone on older cards, it's still a worthy refresh. Bad naming, agreed. But if I can upgrade to the "4080" for $750 now in the form of the 5070 Ti and with DLSS 4, that's a far better deal than anything Nvidia previously offered last gen.

That said, the 5070 is an awful "upgrade" from the 4070S and the 5090's not worth $2000, but people are paying over $2000 for the 4090 already, so it will sell out.

 

[Bubba Jones]

22%?? Not a freakin' chance! It was obvious that the 5080 would never be "allowed" to get anywhere near the 4090 and the SM count is only trivially higher than the 4080, and in the 5080 things like L2 cache and the 256bit bus stay the same. Everyone knew that even an average 15% uplift over a 4080 would be VERY optimistic and it looks like 10%-ish might be the better bet.

The 5090 barely scales with the SM and power increase over the 4090 and the GDDR7 bandwidth seems almost irrelevant, so the 5080 kind of has nowhere to go, other than raw TDP and (very) small silicon gains giving uplift.

 

[Phyzzi]

With half the die being wasted, I really wonder if it wasn't a better idea to have a separate tape out, either that or the yields are horrendous enough to make it worth it.

My guess is that yields really are that bad. These chips are pushing the limits of what can be designed using semi-classical approximations, as well as at the limits of what can be build using photons that interact with electrons more than nuclei, so using "masking" and etching instead of crystallographic holography and I don't even know what (which I am not even sure could be done for something this complex, much less anything but the most general hand waving vague ideas of how it might happen). Add that this is a new design on a pretty new process and having a few hitches is To Be Expected. While I have no real idea, given the price structure and the fact that the 5090 went up significantly from the 4090 but the 5080 is staying similarly priced, and looking at the price difference between them, I am guessing they expect something like a quarter of their PUs to fail at some level, and probably a fifth or so to fail completely (assuming each chip has two complete PUs), which probably means more are failing now and fewer will fail later and also that they will be spending the next few (or several) months figuring lol out how to salvage as much as possible from the "failed" boards to resell them as 5070's and below, as well as starting to work out how to do a "Super" or "Ti" refresh with chips that have enough extra working among the parts that failed QA for one side or the other to give things a little bump or have a few more be "mostly okay".

That expected failure rate would be over the whole run though, so right now things are probably pretty miserable, and maybe the staggered release of lower level chips is because they aren't really sure what will be salvagable if the whole thing doesn't work. Of course, that's a lot of guess work on my part, but it's not blind guesswork.

 

[Pigpig]

Now intel become savior for poor gamers🙄
wait for battlemage B770

 

[artk2219]

Now intel become savior for poor gamers🙄
wait for battlemage B770

Im all for it, with the way the GPU market currently is, there is the space for multiple options.