Associate
- Joined
- 14 Aug 2017
- Posts
- 1,196
So … we don’t actually know what that means. Great!
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
RDNA 3 rumours Q3/4 2022
- Thread starter Grim5
- Start date
- Status
So … we don’t actually know what that means. Great!
It means @TheLurkingDev might be forum member uscool.
Last edited:
That is completely inaccurate, as if you never actually looked into this at all. AMD has used integrated cores for a long time now, which contain stream processors (what you called shader cores, which in itself is not accurate anymore) but also RT accelerators (and other elements). RT accelerators are dedicated transistors that can only accelerate RT and nothing else - and they are present already in RDNA 2. Aside them being inside CU and not a separate unit, they're already dedicated "cores" for RT acceleration. The main difference is that NVIDIA's ones can accelerate more functions than AMD's one in RDNA 2 - one of the reasons of the speed difference. RDNA 3 improves on that, but there's always more space for improvement. CU contains AI acceleration transistors in RDNA 3 (like tensor cores, but again inside in CU and not separate unit). This is just how AMD designs their GPUs with integrating everything they can in CUs, sharing cache in between things etc. It also does NOT mean these things are part of "shaders".
Theres also a gigantic gap between the price of these cards
Associate
- Joined
- 19 Sep 2022
- Posts
- 1,010
- Location
- Pyongyang
i think amd uses a logic unit that doubles as a ray intersection block or a TMU in one clock cycle, thats been the case with rdna2 and leaked numbers suggest that they have continued in the same direction with rdna3..
the identity is like 1 tmu or 4 ray box intersections per clock, so amd will generally report a tmu:rt core in 1:4 ratio, so no dedicated rt hardware for rdna3
also nvidia has moved away from deep BVH structures to flatter BVH structures for data organization, atleast thats what i could interpret form the white paper, and new acceleration structures for rt workloads, the gap will only widen as new games start implementing nvidia's methods
Last edited:
Why not? What’s your budget? I personally would not want to spend more than £700 and for that on a new gen card would want a decent improvement over what I have in both rasta and rt. Thats why it seems to me I will be skipping this gen.
Pretty sure it will be safe to upgrade next gen. 4090 performance will probably drop into my price range, whether that is new or used.
Associate
- Joined
- 19 Sep 2022
- Posts
- 1,010
- Location
- Pyongyang
lol next gen is going to be 600w, atleast thats what i read in few articles and also the reason why amd and nvidia came up with this great idea of a new atx standard.. nex gen is going to be more expensive - unless intel pulls out something magical
Last edited:
how do you figure that? pay 33% more for what exactly. Someone who has only 1k to spend on a card isnt going to pony up more to "get the best on the market' They also sure as **** arnt buying a 4080 16gb with 30% less raster performance which is 200 more
The best part of the RDNA3 reviews isnt seeing how close it is to the 4090, but how much if crushes the 4080 for less coin
Had the pound not dumped and was at 1.4 like it used to be for a long while, these AMD GPU’s would be looking very good to most of us. Prices may go up but if you look at dollar value you can see AMD have done well to improve on price for performance. Same will happen again next gen in my opinion.
£700 and below, I don't need the fastest thing on the block. It's a shame nvidia and amd don't want to offer anything decent in that range and instead just nickel-dime us at 1k+ price. They treat gamers with contempt as a side job to their data business.
Last edited:
Associate
- Joined
- 19 Sep 2022
- Posts
- 1,010
- Location
- Pyongyang
i assumed from your sig that you would be buying a 80 class graphic card, but i believe the 5070 will be able to comfortably do 4K at a more reasonable price - so maybe lets see
maybe you can but i doubt most people can afford to just up their outlay by 33%. only someone with money to spare or an idiot thinks an extra 500 is nothing
the same goes for your 4080 comment, you dismiss it now because its looks ******* stupid at that price. jensen tried to sell you lies about 5nm costing a ton more, but **** if amd arnt charging anymore are they absorbing the costs? **** no, jensen is just trying to drive
up the price to performance tiers and nvidia fanboys cant help themselves but defend it
So looking at the extrapolated numbers from Linus at min;
The 7900xtx gets 24fps
The 3090Ti gets about 26/27fps
So for the 7900xtx to have 29.4% less RT Intersections per second than the 3090Ti and get basically the same performance is actually pretty solid surely?
Or the 4090 is not 5.8x / 484% faster than the 7900xtx either. It is appearing to be about 2x faster. So diminishing returns for that side of things.
No I couldn't that is daft. The extra £500 is an extra 50% what a stupid comment.
Soldato
- Joined
- 26 Oct 2013
- Posts
- 4,132
- Location
- Scotland
And the cheapest 4090 is more like £1800, so it's actually an extra 80% (on the AMD RRP, no idea how much they will actually cost).
What GPU do you have now? Why not grab a 3080 Ti for under £600? Even a 3090 went for £610 recently and those cards cost almost treble that not so long ago..
I only have a 3080 Ti because I managed to get it for £575. XX80 class GPU’s should be £700, but looks like that ship has sailed together with our weak ass currency.
But one could say if you can afford £700 you can add £500 or less on top and get a 7900XTX
I agree though, the 4080 price is stupid and makes no sense.
- Status