Question Speculation: RDNA3 + CDNA2 Architectures Thread

[uzzi38]

Man I have been dying to make this one for a while now.

First rumours for RDNA3 are here so new thread time!

Just going to start off with this one for now: kopite7kimi on Twitter: "@VideoCardz Ah, I mean a simple mcm design with 10240 cores is not enough. Because the lift from RDNA2 to RDNA3 is much bigger than from RDNA1 to RDNA2. We should expect many big improvements in GFX11. 🤔" / Twitter

 

[jpiniero]

So AMD is, at the same time, having too much unsold N33 inventory resulting in them dumping chips in the desktop market but also having "not that much" excess inventory, resulting in them dumping at high prices. Fascinating.

The unsold inventory might not be that much. But I'm guessing that sales are going to be slow regardless of what AMD prices the 7600 at, given their inability to wipe out RDNA2 supply.

 

[Heartbreaker]

You meant a repeat of most recent Radeon launches then!?

And a repeat of the forum "experts" claims.

Every generation the Forum "experts":
"NVidia is overpriced, AMD are morons if they don't undercut them by a large amount and take market share".

But AMD never does. Maybe AMD are morons, or maybe the forum "experts" underestimate how much it costs to produce a modern GPU and how much room to maneuver AMD has. I lean to the latter explanation.

 

[psolord]

What kind of performance increase are we looking, from the 6600 to the 7600?

Techpowerup shows 9tflops for the 6600 and 21tflops for the 7600. Ok there are other metrics as well, maybe that's why they show the real difference at 8%, but will their theoretical computational power be so different? Is there a test that can highlight this (when the 7600 comes out)?

 

[igor_kavinski]

What kind of performance increase are we looking, from the 6600 to the 7600?

AMD Radeon RX 7600 Specs Seemingly Leaked

In line with the RX 6600 XT, updated with RDNA3 silicon magic.

www.tomshardware.com

If it offers 25% average perf increase with lower power draw and better RT, it may cause the RX 7600 to sell out and RX 6600 to tank further in price. Good for us.

 

[jpiniero]

What kind of performance increase are we looking, from the 6600 to the 7600?

Should be close to the 6700.

 

[KompuKare]

And a repeat of the forum "experts" claims.

Every generation the Forum "experts":
"NVidia is overpriced, AMD are morons if they don't undercut them by a large amount and take market share".

But AMD never does. Maybe AMD are morons, or maybe the forum "experts" underestimate how much it costs to produce a modern GPU and how much room to maneuver AMD has. I lean to the latter explanation.

But the price to produce a modern GPU must go down massively after launch, as the 7900 XT has now been cut by what? $200 and counting.

Yes, occasionally their BoM must be very high (did AMD not hint at that during their HBM consumer misadventures?), but mostly they launch at high prices, get bad reviews and low sales, and then reduce their prices. I mean I don't mind as I have never been a launch-day purchases, but this has been their modus operandi for years now.

Anyway, my argument is that with such high fixed design costs, volume is important. Very important.

 

[maddie]

What kind of performance increase are we looking, from the 6600 to the 7600?

Techpowerup shows 9tflops for the 6600 and 21tflops for the 7600. Ok there are other metrics as well, maybe that's why they show the real difference at 8%, but will their theoretical computational power be so different? Is there a test that can highlight this (when the 7600 comes out)?

Doubled 32b shaders in RDNA3 that appear to be marginally effective. Divide the 21 by 2.

 

[coercitiv]

Every generation the Forum "experts":
"NVidia is overpriced, AMD are morons if they don't undercut them by a large amount and take market share".

But AMD never does. Maybe AMD are morons, or maybe the forum "experts" underestimate how much it costs to produce a modern GPU and how much room to maneuver AMD has. I lean to the latter explanation.

Lean away!

 

[Heartbreaker]

But the price to produce a modern GPU must go down massively after launch, as the 7900 XT has now been cut by what? $200 and counting.

Yes, occasionally their BoM must be very high (did AMD not hint at that during their HBM consumer misadventures?), but mostly they launch at high prices, get bad reviews and low sales, and then reduce their prices. I mean I don't mind as I have never been a launch-day purchases, but this has been their modus operandi for years now.

Anyway, my argument is that with such high fixed design costs, volume is important. Very important.

Production cost will go down some, but not massively. Prices dropping later are not an indication that production costs have dropped that much, but that it's just in a different part of the product life cycle.

If we go back to before all the mining nonsense:

Cards would launch at the MSRP kind of aimed at early adopters, and when sales slowed months later you would get a price cut, and they would sell more a while at that price and then near the end they get a final clearance price cut. It's a lifecycle strategy aimed at both maximizing margin initially and volume later.

Overall launching high lets them improve the ASP, and it's NOT a bad strategy.

7900 XT pricing was not that, and kind of unique. It was just bad pricing, unless it's point was to sell more XTX cards, by making them look much better in comparison.

 

[MrTeal]

My feeling on that is yields are quite good on N31, and there's just not a lot of stock of dies they have to sell in cut down SKUs. Pricing the 7900XT high reduces demand, and keeps them out of the situation that they're disabling fully functional dies just to feed that market.

 

[coercitiv]

My feeling on that is yields are quite good on N31, and there's just not a lot of stock of dies they have to sell in cut down SKUs. Pricing the 7900XT high reduces demand, and keeps them out of the situation that they're disabling fully functional dies just to feed that market.

This was the explanation I saw on day one as well, that they priced the cut-down card in line with their expected supply, not their production cost. Apparently the market was sympathetic to AMDs problematic situation, so they held off buying the cards, allowing them to build up inventory and be more flexible with pricing.

 

[PJVol]

AMD Radeon RX 7600 Specs Seemingly Leaked ​

In line with the RX 6600 XT, updated with RDNA3 silicon magic.

www.tomshardware.com

Lol at the title. What specs are they talking about or did I miss something? )

If it offers 25% average perf increase with lower power draw

Not sure about power draw, but 25% is the difference between 6600 and 6650xt. Adding rumored 11% (MLID) gives 38% over 6600.

 

[igor_kavinski]

Lol at the title. What specs are they talking about or did I missed something? )

Not sure about power draw, but 25% is the difference between 6600 and 6650xt. Adding rumored 11% (MLID) to the lattee gives 38%.

This screenshot:

https://twitter.com/x/status/1656541226831855616

 

[PJVol]

This screenshot:

I don't have access to twitter.

 

[igor_kavinski]

I don't have access to twitter.

Not sure why it's slanted.

 

[Dribble]

You are aware Intel is currently operating at a loss, right?

That was the point - x86 is slowly being pushed out of markets due to architecture age and restrictions on who can make it. That said I would say it's AMD finally being competitive with Intel that's hurt them the most. Hence not surprising AMD are focusing more on x86 then gpu's right now.

 

[PJVol]

Not sure why it's slanted

Ah...) I've seen it in TH post. It's likely from the box photo leaked by videocardz, and it doesn't say anything new, such as gfx and vram clocks, tbp, etc.

 

[Mopetar]

3080 also only had 10GB of VRAM vs 16GB on 4080.

So there is likely significant cost growth in the two most expensive components, GPU chip and VRAM, plus some margin increase.

An extra 6 GB of VRAM is insignificant when examining the total BOM for a GPU. The only reason cost would have gone up is that the 4080 uses the fastest GDDR6X memory available (at the time) whereas the 3080 used slightly slower rated chips.

The 3080 has a larger bus, so it actually would have used more chips (10 vs. 8) than the 4080 so if they were buying the same general cost level per chip, the 4080 would have a lower cost there. The supply chain isn't as crouched as it was during the pandemic so the prices for VRAM are probably coming down or should be assuming the manufactures aren't playing the same game as NVidia and AMD and pricing higher to squeeze extra revenue and profit.

Unless a company is buying premium VRAM, the costs aren't anywhere near what a consumer can expect to pay for the extra memory. It's not quite as bad as Apple charging an extra $100 for something that costs them $3, but there's a lot of markup.

 

[tajoh111]

AMD Radeon RX 7600 Specs Seemingly Leaked

In line with the RX 6600 XT, updated with RDNA3 silicon magic.

www.tomshardware.com

If it offers 25% average perf increase with lower power draw and better RT, it may cause the RX 7600 to sell out and RX 6600 to tank further in price. Good for us.

Just looking at the original MSRP of the RX 6600xt, it pretty crazy that a 230mm2 was priced at 400 dollars.

I don't think the market is going to be happy with 7600xt as AMD poisoned the market with their 8GB crusade along with their comparisons with last gens performance vs todays products.

The 7600xt performance is going to be pretty mediocre compared to the 6600xt(10% or less) and has 8gb of video memory. This is going to lead to the card being hammered by reviewers when they take into account current pricing of AMD's last gen.

 

[Ranulf]

I don't think the market is going to be happy with 7600xt

I think you mean the 7600. The XT version isn't coming out this month (if ever). Then again I wouldn't be too surprised if they release a 7600XT with 8gb of ram at $300 or more.

 

[Heartbreaker]

I think you mean the 7600. The XT version isn't coming out this month (if ever). Then again I wouldn't be too surprised if they release a 7600XT with 8gb of ram at $300 or more.

Do you mean a 7600Xt with 16GB?

Because the 7600 is apparently already the full 32 CUs...

 

[Kepler_L2]

Do you mean a 7600Xt with 16GB?

Because the 7600 is apparently already the full 32 CUs...

The 7600 XT could be a 40 CU, 8 GB version of N32.

 

[Aapje]

The 7600 XT could be a 40 CU, 8 GB version of N32.

That makes absolutely no sense. If it is at least 12 GB, it would actually sell.

 

[coercitiv]

That makes absolutely no sense.

Sure it does, it would be the perfect card to look bad against the 6700XT. We can even give it a new nickname: 7600XT "Close your browser to free up 0.4GB of VRAM" Edition.

 

[GodisanAtheist]

If N32 is going to have any cuts, it's going to be a 56 CU 7700. At most I can see a 48CU 7600xt. This is of course assuming 7700xt is 64 CU full die and 70 CU N31 is the 7800.

But doing a cut all the way down to 40CU on a 200mm GCD seems nuts.

AMD might have botched Chiplets in the weirdest kind of way, where they're getting great yields on underperforming parts and now product segmentation has gone to the dogs.