AMD: RDNA 3 Speculation, Rumours and Discussion

[Flappy Pannus]

what a load of crap,

raster + RT + upscale , thats covers about 99% of the consumer GPU buying market.

by your logic not a single console would be sold..... but wait.......

I think 'not enough' might apply to $1000+ GPU's. There's obviously a market for these GPU's for gamers only, but it's a relatively small one.

The impact of 'mind share' from the lack of these flagship champions is debatable, though. If AMD can swoop in and make a clear $500 champion over Nvidia, that's how they get their gamer mindshare, as those are the cards the majority actually buy. Their problem up until this point is that they haven't really been able to compete in that area either.

 

[xpea]

what a load of crap,

raster + RT + upscale , thats covers about 99% of the consumer GPU buying market.

by your logic not a single console would be sold..... but wait.......

Console and PC are different markets with different competitive landscape, different software workloads, and requirements. I don't even know why I have to type this...
And no, PC gamers are not satisfied by multi years stagnant console specs, otherwise it would not sale more RTX dGPUs than PS5. Sorry for being Captain Obvious again ...

 

[Bondrewd]

the price of gpus would come down

Yes, of NV ones, to retain the MSS.
That's how it's been going since 2007.
Makes for a cycle where no one makes much money.

raster + RT + upscale , thats covers about 99% of the consumer GPU buying market.

Bingo but AMD brand rep was never quite there since 2007 to turn into real money or mss.

If AMD can swoop in and make a clear $500 champion over Nvidia

That was HD 5k all those eons ago and it surely never booked AMD much money or any crazy mss swings.
For the foreseeable future non-NV dGP brands exist as a pricing lever against NV, unless they don't like the previous cycle.

 

[Flappy Pannus]

That was HD 5k all those eons ago and it surely never booked AMD much money or any crazy mss swings.
For the foreseeable future non-NV dGP brands exist as a pricing lever against NV, unless they don't like the previous cycle.

The 5x00 line were 'comparable' to Turing, a somewhat better value in raster, but certainly nothing like a slam dunk. The main problem also was that around ~8 months after it hit the market, so did covid. Hard to have a price war when neither vendor can keep anything in stock.

 

[LordEC911]

*Deleted*
I think I misunderstood the discussion.

 

[Bondrewd]

The 5x00 line were 'comparable' to Turing, a somewhat better value in raster, but certainly nothing like a slam dunk. The main problem also was that around ~8 months after it hit the market, so did covid. Hard to have a price war when neither vendor can keep anything in stock.

Oh no, not the Navi1 parts but the Evergreens of 2009.

 

[techuse]

Oh no, not the Navi1 parts but the Evergreens of 2009.

Evergreen took them up almost 15 percentage pints of marketshare.

 

[LordEC911]

Evergreen took them up almost `5 percentage pints of marketshare.

That's a mighty fine estimate to what really happened...

 

[JoeJ]

PC gamers are not satisfied by multi years stagnant console specs, otherwise it would not sale more RTX dGPUs than PS5. Sorry for being Captain Obvious again ...

This raises two questions:
Why are the averaged PC gamer specs always close to the actual console generation at the same time?
What do we do if increasing HW power is no longer possible without increasing prices as well, which seemingly is already the case right now?

It feels obvious enthusiast PC gaming has no bright future.

 

[yuri]

AMD has been sitting in that niche for so long because they couldn't afford the R&D to compete with both Intel and nVidia, and chose to prioritize CPUs. R&D investment, and especially GPU R&D investment, has risen very dramatically because of their success on the CPU side. AMD spent more (even after inflation) on R&D last quarter than they spent in the entire year of 2016.

Yea, R&D budgets are important. Although one might argue that AMD is still far behind either Intel or nV. Competing with both of them (CPUs && GPUs) would still not be viable and thus keeping the current prioritization going on.

Does the public R&D spending data differentiate between the AMD divisions? If not we still do not know where is the spending aimed.

IF AMD and Intel were competitive with Nvidia in terms of performance and features then the price of gpus would come down as they try and compete with each other .

IMO it wouldn't work for Intel/AMD from the financial standpoint - the GeForce brand is just too strong.

Imagine AMD setting a goal to reach 95% of nV features + performance. The ROI would not be there - spending on developing monsters like 4090Ti, spending on building the whole CUDA-compatible environment, OptiX-compatible environment, building all the RTX features, having dev support like nV, etc. All that in the "nV quality". The final 95% product would be cool, but GeForce would still be 100%. The gaming crowd would not buy the 95% product. The $$$ would still flow to the 100% product aka GeForce.

 

[Putas]

Why are the averaged PC gamer specs always close to the actual console generation at the same time?

Because they exist at the same time. The same technology enables similar devices.

 

[JoeJ]

Because they exist at the same time. The same technology enables similar devices.

And most importantly: Enables similar games.

However, the curious aspect is this: From Moores Law slowing down we could assume the similarity across devices increases. But actually we see the opposite:
Low end handhelds, which feels increasingly popular.
APU consoles, which feels like the sweet spot.
High end PC, which feels like pretending halo products like Ada and chiplet RDNA3 would be the norm.

But we want and need similar games regardless, so the enthusiast section becomes a bad investment for both devs and gamers on the long run, i expect.

 

[Jawed]

You mean those LLVM/AMDGPU commits? They can be changed as well as during Navi1 and Navi2 launches

There's also corresponding patent documents. Finally, rumours point towards more ALU lanes per WGP, though that's the weakest support for my opinion.

I don't remember anything significant being "misleading" about commits for Navi 1/2, when comparing pre-launch and post-launch versions of the code.

 

[Kaotik]

Pretty sure it's 18 then 20 then 24 as far as G6 goes.
20Gbps parts are even listed somewhere iirc.

Double checked, you're right, Samsung has 20 gbps bin in there too

 

[Kaotik]

I agree on Nvidia pricing but I don't see anyone buying a Radeon GPU above $1k, no matter how good is it. At least, not before AMD improve their ecosystem (Broadcast, TRX noise, RTX Remix now) and global offer for the target audience (performance and support on AI, ML, Compute, rendering, RT, video encoding workloads). Be only good at raster gaming is not enough in 2022
However, in gaming mainstream value market (below $500), it's less of a problem even if mind share hurts a lot. AMD still needs a strong halo product for 2 or 3 generations to change their brand perception

"RTX Remix" which so far is still unproven for anything but sandbox fun is the only one they're missing from your list. They have working streaming tools and driver built-in noise suppression.

 

[Bondrewd]

Evergreen took them up almost 15 percentage pints of marketshare.

They've failed to breach 50% mss ceiling with a royal NV fuckup and margin shredder lineup.
That's a bad example.

 

[Silent_Buddha]

Evergreen took them up almost 15 percentage pints of marketshare.

To be fair HD 4xxx got them most of that way, from 31% to 40.6% HD 5xxx just carried that momentum (to 43.8%) by taking the performance crown. While GTX 5xx reversed most of that it's interesting that the lower parts in the HD 5xxx stack managed to win market share back almost immediately and exceeded the market share that HD 58xx had gained. 43.8% for HD 58xx effect and a slower gain to 44.5% due to the lower (and more affordable) HD 5xxx parts.

If anything that shows that price has a larger influence than performance as long as performance is relatively close. Basically proof that while halo parts can be important, perf/price is even more important in terms of shifting significant numbers of cards.

I'd say that's potential evidence that AMD doesn't need to beat NV in order to win back market share. What they need are parts that are close enough in performance that provide a noticeable cost savings over the NV counterparts. Basically repeat what they did with HD 4xxx and all HD 5xxx parts lower than 58xx.

The big question is whether or not AMD wants to dedicate enough valuable silicon to GPUs? Doing so potentially reduces their more lucrative CPU sales. I'd say AMD are more willing to sacrifice GPU sales volume than they are to sacrifice CPU sales volume.

So, if the plan is already to not claw back market share due to dedicating most of their silicon wafer starts to CPU, then it makes some sense to price their GPUs as high as they can such that it can move the relatively limited number of GPUs that they plan to produce. Basically, they won't price them to move a lot of GPUs, instead they'll price them to move a limited number of GPUs. And that's regardless of how performant they are as CPU margins are likely higher for AMD right now than GPU margins.

Regards,
SB

 

[Esrever]

Do people think Nvidia is just going to sit there an take it if AMD lowered prices? The only reason AMD was able to even gain market share back in 2009 was cause Nvidia screwed up so much that AMD actually could sell GPUs. When AMD released the rx480 for a good price, Nvidia didn't do anything except have a decent competitive model also on the market and AMD didn't gain any market share. They just released the 1060 3gb for less price and less performance which people bought anyways. Even tho the 1060 3gb is a cut down card compared to the 1060 6gb.

 

[Bondrewd]

Correct, in the current market AMD mostly exists as a pricing lever in the eyes of consumer population.
So they're not gonna actively compete on price or try to push major GPU volumes etc.

 

[BRiT]

Moved the OFFTOPIC nVidia 4080 Unlaunch to a more fitting thread. This topic is about AMD RDNA3. Stay on target!