AMD Announces The Radeon RX 7700 XT & RX 7800 XT Graphics Cards

[stormcrow]

It'll do quite well against 4060ti, which is ALSO a joke.

If the 7700XT does well in the benchmarks versus the Nvidia card then it's quite likely to be a contender if one isn't a brand... person. The extra VRAM is going to be situational, even assuming there's no memory bottleneck as sometimes happens once all the VRAM is in use (like has regularly happened in the past with Nvidia two tiered models). My experience is that VRAM often isn't the bottleneck on future games many people assume. The GPU & memory bandwidth are usually the bottleneck beyond games around 18 - 24 months out from point of sale. But the price could be compelling for current generation 1440p gamers if actual game play performance statistics are favorable... at least, for games. If it's for compute loads, then AMD loses regardless because AMD's compute software package offerings are awful to deal with.

[Mike Frett]

The $450 price of the 7700 XT will go down

I think you mean the price will go up.

[Dehir]

I don't like that the 7700XT has a higher TDP than the 6700XT personally, I'll keep mine and pass on this gen, I'd like the same perf for 180W or something like that.

Well you could easily undervolt it. Was it like you can default lower the power target -10% with software with that much of an penalty in performance. That lowers it to 220w. Thou another way specially if you are playing/sticking with 1080p is to lock the FPS on lower which also lower the stress on GPU. Older games probably wont need that much power anyway.

[drakonas777]

I don't like that the 7700XT has a higher TDP than the 6700XT personally, I'll keep mine and pass on this gen, I'd like the same perf for 180W or something like that.

It is what it is. PC community has accepted high power consumption as a new norm. Today a mid range PC consumes about the same amount of energy as an original HEDT. "But but but muh efficiency" - yeah whatever. It's good for business though. Now you have to spend several hundred dollars more for cooling solutions (starting from higher tier videocards for better cooling perf ending with quality meshed cases) on top of the everything.

Thanks to such internet clowns as JaysTwoCents who made such bullshit as water cooling in electronics a almost "desirable" norm.

[artivision]

RDNA3 is a joke. The 4xFP never works and so half the units do 2xFP and half 2xFP + 2xINT. Tho is advertised and payed normally. My opinion is that is time for some mass lawsuits. This product should never be released.

[avis]

Am I the only one that absolutely hates how physically large graphics cards are?

I know this isn't anything new, the Voodoo 5 5500 was huge and that was 23 years ago, but ram, CPUs and GPUs have advanced to the point where you can get a laptop with a 16 core processor, 64gb, and Quadro GPU, the AMD AMD Ryzen 9 7900X has 12 cores and reasonably capable graphics in a relatively compact design, yet add-in graphics cards are huge, consume significant amounts of juice, generate a lot of heat and require 1 or 2 power connectors.

I think we should boycott any graphics card that is so physically large that it looks like someone is trying to compensate some other shortcoming.

It's inevitable. In the past you could shrink transistors and get performance for free. For the past couple of years, transistors haven't really shrunk much but OEMs need to release products with increased performance, so they've had no choice but to use bigger and bigger dies which results in more power consumption and much beefier coolers, i.e. substantially larger cards.

We need new physics or completely new materials (photonics?) to tackle the issue.

[creative]

sophisticles

Not a fan of large GPU's, ironically in general the bigger they are the better they are cooled, three fans help.

[bridgman]

RDNA3 is a joke. The 4xFP never works and so half the units do 2xFP and half 2xFP + 2xINT. Tho is advertised and payed normally. My opinion is that is time for some mass lawsuits. This product should never be released.

4X FP ? As far as I know all we did (and all we claimed) was add a second ALU to each stream processor.

Doing that doesn't double performance but it does remove (or raise) a performance limit in certain compute bound scenarios. It's like having more ROPs - only makes a difference in certain scenarios but significant in those scenarios.

I'm curious where you saw the "4xFP" reference other than maybe packed instructions working on small data types (eg 2 FP16 instead of 1 FP32). I don't remember seeing any reference to 4x in our slides or ISA guide.

The only "4xFP" I can think of would be the combination of dual-issue and packed FP16, but even that would not be 4x since RDNA2 also supports packed FP16 IIRC.

[artivision]

4X FP ? As far as I know all we did (and all we claimed) was add a second ALU to each stream processor.

Doing that doesn't double performance but it does remove (or raise) a performance limit in certain compute bound scenarios. It's like having more ROPs - only makes a difference in certain scenarios but significant in those scenarios.

I'm curious where you saw the "4xFP" reference other than maybe packed instructions working on small data types (eg 2 FP16 instead of 1 FP32). I don't remember seeing any reference to 4x in our slides or ISA guide.

The only "4xFP" I can think of would be the combination of dual-issue and packed FP16, but even that would not be 4x since RDNA2 also supports packed FP16 IIRC.

That is the way you count it: 6K cores * 4xFP32 * 2.5Ghz = 60Tflops
I couldn't get 60Tflops to anything tho. I guess is 30Tflops + 15Tiops + Your Imagination

[Mechanix]

Good. Perhaps this means the price of the RTX4070 will drop.