-
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.
AMD RDNA3 unveiling event
- Thread starter CAT-THE-FIFTH
- Start date
Last edited:
Associate
The xtx will further inch up in price. A large quantity of units have been taken out of supply. Maybe explore a liquid cooling block (which i personally consider to be waste of money)
Last edited:
Jensen, is that you?
Shhh, not so loud, I'm trying to get a feel for any interest in a GeForce leather range.
Just $299, billions of rays per second, maxed out multi-bounce indirect lighting effects, a feast for the eyes! And my bank account.
Last edited:
Benchmarking the RDNA3 architecture
chipsandcheese.com
Some pretty interesting bits in there you won't see in other reviews. I found the exploration of AMD's claimed dual issue compute abilities to be quite interesting
"RDNA 3’s dual issue mode has limited impact. It relies heavily on the compiler to find VOPD possibilities, and the compilers are frustratingly stupid at seeing very simple optimizations. For example, the FMA test uses one variable for two inputs, which should make it possible for the compiler to meet dual issue constraints. But obviously the compiler didn’t make it happen."
"It’ll be interesting to see how RDNA 3 performs once AMD has more time to optimize for the architecture, but they’re definitely justified in not advertising dual issue capability as extra shaders cores. Typically, GPU manufacturers use shader count to describe how many FP32 operations their GPUs can complete per cycle. In theory RDNA3's dual issue would double FP32 throughput per WGP with very little hardware overhead besides the extra execution units. But it does so by pushing heavy scheduling responsibility to the compiler. AMD is probably aware that its compiler technology is not up to the task. AMD can optimize games by replacing the shaders with hand-optimized assembly instead of relying on this compiler code generation."
So that's perhaps where RDNA3's rumoured gaming performance went missing - in order for a game to make full use of RDNA3 architecture, game specific hand written code needs to be used. Perhaps games need to built from the ground up for it, or otherwise AMD needs to do it themselves by releasing hand written per game optimizations for its drivers.
AMD's software and driver teams are going to be very, very busy in 2023 if AMD intends to eventually unlock RDNA3's full performance potential
Chips and Cheese | Substack
The Devil is in the Details! Deep dives into computer hardware and software and the wider industry... Click to read Chips and Cheese, a Substack publication with thousands of subscribers.
chipsandcheese.com
Some pretty interesting bits in there you won't see in other reviews. I found the exploration of AMD's claimed dual issue compute abilities to be quite interesting
"RDNA 3’s dual issue mode has limited impact. It relies heavily on the compiler to find VOPD possibilities, and the compilers are frustratingly stupid at seeing very simple optimizations. For example, the FMA test uses one variable for two inputs, which should make it possible for the compiler to meet dual issue constraints. But obviously the compiler didn’t make it happen."
"It’ll be interesting to see how RDNA 3 performs once AMD has more time to optimize for the architecture, but they’re definitely justified in not advertising dual issue capability as extra shaders cores. Typically, GPU manufacturers use shader count to describe how many FP32 operations their GPUs can complete per cycle. In theory RDNA3's dual issue would double FP32 throughput per WGP with very little hardware overhead besides the extra execution units. But it does so by pushing heavy scheduling responsibility to the compiler. AMD is probably aware that its compiler technology is not up to the task. AMD can optimize games by replacing the shaders with hand-optimized assembly instead of relying on this compiler code generation."
So that's perhaps where RDNA3's rumoured gaming performance went missing - in order for a game to make full use of RDNA3 architecture, game specific hand written code needs to be used. Perhaps games need to built from the ground up for it, or otherwise AMD needs to do it themselves by releasing hand written per game optimizations for its drivers.
AMD's software and driver teams are going to be very, very busy in 2023 if AMD intends to eventually unlock RDNA3's full performance potential
Last edited:
Soldato
- Joined
- 14 Aug 2009
- Posts
- 3,190
Same thing it was previously when only using Mantle you could get the best out of their cards.Benchmarking the RDNA3 architecture
Chips and Cheese | Substack
The Devil is in the Details! Deep dives into computer hardware and software and the wider industry... Click to read Chips and Cheese, a Substack publication with thousands of subscribers.
Some pretty interesting bits in there you won't see in other reviews. I found the exploration of AMD's claimed dual issue compute abilities to be quite interesting
"RDNA 3’s dual issue mode has limited impact. It relies heavily on the compiler to find VOPD possibilities, and the compilers are frustratingly stupid at seeing very simple optimizations. For example, the FMA test uses one variable for two inputs, which should make it possible for the compiler to meet dual issue constraints. But obviously the compiler didn’t make it happen."
"It’ll be interesting to see how RDNA 3 performs once AMD has more time to optimize for the architecture, but they’re definitely justified in not advertising dual issue capability as extra shaders cores. Typically, GPU manufacturers use shader count to describe how many FP32 operations their GPUs can complete per cycle. In theory RDNA3's dual issue would double FP32 throughput per WGP with very little hardware overhead besides the extra execution units. But it does so by pushing heavy scheduling responsibility to the compiler. AMD is probably aware that its compiler technology is not up to the task. AMD can optimize games by replacing the shaders with hand-optimized assembly instead of relying on this compiler code generation."
So that's perhaps where RDNA3's rumoured gaming performance went missing - in order for a game to make full use of RDNA3 architecture, game specific hand written code needs to be used. Perhaps games need to built from the ground up for it, or otherwise AMD needs to do it themselves by releasing hand written per game optimizations for its drivers.
AMD's software and driver teams are going to be very, very busy in 2023 if AMD intends to eventually unlock RDNA3's full performance potential
Considering the market share AMD has is pretty safe to say that not a lot of studios will bother to code specifically for them.
Best of luck to the driver's team and to players hoping that their favourite games get the necessary attention.
Soldato
That’s a very wild assumption. Completely ignoring that AMD is used in all consoles except the Switch and their implementations are all open source so cheaper.
Soldato
- Joined
- 14 Aug 2009
- Posts
- 3,190
Open source is meaningless unless is actively pushed by them is not going to make much of a difference.
Consoles are different. Back in r290 day i only got the best of that card with Mantle. The hardware scheduler and async compute worked best only when it was properly coded for them. Usually it wasn't and you could see that when dx11 code stumbled due to draw calls and and such, in heavy scenes, where nVIDIA was more stable.
LE: consoles are RDNA 2 , not RDNA3, so... pointless.
Last edited:
I don't get why AMD are not interested in gaining market share, even if it is at the expense of some profits for a gen or two. Surely they would have improved market share quite a bit had they called the 7900 XTX a 7800 XT and priced it $799 at most.
In one or two gens they may end up at 5% market share the way they going
Soldato
It is being actively pushed. We are seeing FSR and the rest appearing in most games.
Consoles aren’t that different. Only the operating system. How the game operates on the hardware is just the same.
AMD will optimise their drivers for all games over time. As has been pointed out in reviews the games that AMD performed really well on had driver optimisations so we know the potential is there.
Last edited:
I wish people would stop saying amd being in consoles will benefit them, we (including me! 
) have been saying that since the day they got into consoles yet here we are still waiting for it... if anything, nvidia actually benefit more when looking at sonys/ps ports Does make you wonder though if amd weren't in consoles, how much worse would games run on the desktop side? 
I did raise the concern with this chiplet design a while back and who would have to optimise for it to get the best from it:
And even before that but can't find the post now.
AMD prefers open source as it means they can do an over the fence approach and let others do the work for them (this has always been made clear ever since Roy made a comment more or less saying that), this is why open source is great as you get a whole community looking into your code and able to recommend improvements i.e. create a branch from the main branch, add, remove, change etc. code then create a pull request and get someone from amd (or others that are assigned the perms) to review the code changes and if all looks good, they merge it into the main branch. That and also because of amds marketshare, they wouldn't be able get away with closed source.
) have been saying that since the day they got into consoles yet here we are still waiting for it... if anything, nvidia actually benefit more when looking at sonys/ps ports Does make you wonder though if amd weren't in consoles, how much worse would games run on the desktop side? I did raise the concern with this chiplet design a while back and who would have to optimise for it to get the best from it:
And even before that but can't find the post now.
AMD prefers open source as it means they can do an over the fence approach and let others do the work for them (this has always been made clear ever since Roy made a comment more or less saying that), this is why open source is great as you get a whole community looking into your code and able to recommend improvements i.e. create a branch from the main branch, add, remove, change etc. code then create a pull request and get someone from amd (or others that are assigned the perms) to review the code changes and if all looks good, they merge it into the main branch. That and also because of amds marketshare, they wouldn't be able get away with closed source.
Last edited:
Soldato
- Joined
- 14 Aug 2009
- Posts
- 3,190
They have a "holly" duty, apparently, towards their investors to maximize profits now, with not much regard to the future. Since they are "luxury" items, perhaps we are seeing this wrong - the less people have them, the more their value increases! So AMD GPUs are actually worth more than Nvidia's due to lower market share!I don't get why AMD are not interested in gaining market share, even if it is at the expense of some profits for a gen or two. Surely they would have improved market share quite a bit had they called the 7900 XTX a 7800 XT and priced it $799 at most.
In one or two gens they may end up at 5% market share the way they going
)Consoles are RDNA 2, not 3.
Soldato
At the moment. The Xbox and PS5 updates are expected to be RDNA 3.They have a "holly" duty, apparently, towards their investors to maximize profits now, with not much regard to the future. Since they are "luxury" items, perhaps we are seeing this wrong - the less people have them, the more their value increases! So AMD GPUs are actually worth more than Nvidia's due to lower market share!
Consoles are RDNA 2, not 3.
On profits AMD make big money still. They exceeded Nvidia on gaming on the last financial results ($1.60bn vs $1.57bn). Nvidia become a small player for personal computing and consoles once you look at whole market where Intel is obviously the dominant player for both CPU and GPU on PC and then AMD for PC and consoles. If anything that is creating pressure on Nvidia as it’s their market to lose and they do not operate in the other personal computing market segments. Nvidia know this so tried to buy ARM but that fell through. Nvidia do win again once you look at industrial/enterprise, AI and cloud computing which is arguably their biggest business.
Last edited:
Soldato
- Joined
- 14 Aug 2009
- Posts
- 3,190
Expected is one thing. Released with games actually made with that arch in mind that do make use of RDNA3 (and to what degree?) is another. Untile then the next or next-next series of cards will be out.
and the consoles will still cost the same no doubt but will turn a better profit for amd, sony and ms, new process smaller die's better cooling savings all around. apart from the buyer, because money
They have a "holly" duty, apparently, towards their investors to maximize profits now, with not much regard to the future. Since they are "luxury" items, perhaps we are seeing this wrong - the less people have them, the more their value increases! So AMD GPUs are actually worth more than Nvidia's due to lower market share!
Caporegime
Benchmarking the RDNA3 architecture
Chips and Cheese | Substack
The Devil is in the Details! Deep dives into computer hardware and software and the wider industry... Click to read Chips and Cheese, a Substack publication with thousands of subscribers.
Some pretty interesting bits in there you won't see in other reviews. I found the exploration of AMD's claimed dual issue compute abilities to be quite interesting
"RDNA 3’s dual issue mode has limited impact. It relies heavily on the compiler to find VOPD possibilities, and the compilers are frustratingly stupid at seeing very simple optimizations. For example, the FMA test uses one variable for two inputs, which should make it possible for the compiler to meet dual issue constraints. But obviously the compiler didn’t make it happen."
"It’ll be interesting to see how RDNA 3 performs once AMD has more time to optimize for the architecture, but they’re definitely justified in not advertising dual issue capability as extra shaders cores. Typically, GPU manufacturers use shader count to describe how many FP32 operations their GPUs can complete per cycle. In theory RDNA3's dual issue would double FP32 throughput per WGP with very little hardware overhead besides the extra execution units. But it does so by pushing heavy scheduling responsibility to the compiler. AMD is probably aware that its compiler technology is not up to the task. AMD can optimize games by replacing the shaders with hand-optimized assembly instead of relying on this compiler code generation."
So that's perhaps where RDNA3's rumoured gaming performance went missing - in order for a game to make full use of RDNA3 architecture, game specific hand written code needs to be used. Perhaps games need to built from the ground up for it, or otherwise AMD needs to do it themselves by releasing hand written per game optimizations for its drivers.
AMD's software and driver teams are going to be very, very busy in 2023 if AMD intends to eventually unlock RDNA3's full performance potential
It does make you wonder if AMD have forgotten that they had a dual issue architecture once before and moved away from it for good reason.
Caporegime
My guess is...
Working:
Not working
Working:
Not working
Oh my, let's hope this is not a new Vega...My guess is...
Working:
Not working
