Discussion RDNA4 + CDNA3 Architectures Thread

Page 85 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DisEnchantment

Golden Member
Mar 3, 2017
1,747
6,598
136
1655034287489.png
1655034259690.png

1655034485504.png

With the GFX940 patches in full swing since first week of March, it is looking like MI300 is not far in the distant future!
Usually AMD takes around 3Qs to get the support in LLVM and amdgpu. Lately, since RDNA2 the window they push to add support for new devices is much reduced to prevent leaks.
But looking at the flurry of code in LLVM, it is a lot of commits. Maybe because US Govt is starting to prepare the SW environment for El Capitan (Maybe to avoid slow bring up situation like Frontier for example)

See here for the GFX940 specific commits
Or Phoronix

There is a lot more if you know whom to follow in LLVM review chains (before getting merged to github), but I am not going to link AMD employees.

I am starting to think MI300 will launch around the same time like Hopper probably only a couple of months later!
Although I believe Hopper had problems not having a host CPU capable of doing PCIe 5 in the very near future therefore it might have gotten pushed back a bit until SPR and Genoa arrives later in 2022.
If PVC slips again I believe MI300 could launch before it :grimacing:

This is nuts, MI100/200/300 cadence is impressive.

1655034362046.png

Previous thread on CDNA2 and RDNA3 here

 
Last edited:

leoneazzurro

Golden Member
Jul 26, 2016
1,052
1,716
136
Neither Ian or phoenix21 have claimed this, and your usage of quotes to make it seem like you are quoting someone, when you are not, is actually the misleading part here.
A video called "AMD Ryzen 9 7950X: cost to manufacture?" with a 69$ in the thumbnail, when in the first seconds (0:06 ) we heaar "we are going to calculate how much it costs to amd to manufacture this..." and with the conclusion "Your new shiny CPU cost 69$ to make" (13:30) is not misleading?
Did you even see the video?
Yes, after he says this he puts a written "Bill of Material" under the 69$ but before that he says twice that he was to calculate "the cost to manufacture".
These are both WRONG.
First, what he is listing is not the full BOM: sure, these are probably the most expensive parts of the BOM but BOM includes also auxiliary/interconnect materials, external packaging (blisters, carton boxes, down to instructions fliers).
Second, BOM is NOT cost to manufacture. Cost to manufacture includes other things than direct costs of the materials. It includes labour, testing, binning, direct overheads, shipping costs between production sites. And these are only the direct costs of the production, because in the total cost of manufacture there are many other expenses as well, R&D and tooling are among the first.
Yes, there is a disclaimer "I can be wrong ant this may not be the full BOM" at the end. Well, then at least try to estimate and speak about what it is missing. Presenting a flat 69$ figgure is outright misleading.
And no,I am not quoting anyone. These are cost accounting basics.

EDIT: not surprisingly, there are similar criticism to that video in the comments below it.
 
Last edited:

soresu

Diamond Member
Dec 19, 2014
3,228
2,515
136
Straying a wee bit off topic here for a GPU thread?

tumblr_ohsj6v3Zp21r115hxo2_250.gif
 

leoneazzurro

Golden Member
Jul 26, 2016
1,052
1,716
136
People were using a video giving a correct data (the estimation of die costs for an AMD CPU) presented in a largely incorrect way (as the cost of the manufacture of a AMD CPU) for estimating what a AMD GPU should cost to the final user.
As one of my tasks for more than 15 years was about production costs accounting, am I allowed to say that this is, at least, not the correct way of estimating a GPU cost in a GPU thread? Thanks.
 

phoenix21

Junior Member
Apr 6, 2024
7
11
41
People were using a video giving a correct data (the estimation of die costs for an AMD CPU) presented in a largely incorrect way (as the cost of the manufacture of a AMD CPU) for estimating what a AMD GPU should cost to the final user.
I think there is misunderstanding, I agree the video doesn't have full BOM of product but I was specifically talking about 6nm IO die cost and trying to make a point that a similarly sized 4nm N44 doesn't significantly increase the die cost, infact it is about the same as similarly spec 204mm² N33. But Rx 7600 MSRP does include costs of other parts, R&d etc, and the inflation in those parts is not that much and can be easily managed under $270-$300 gap.
If a 122mm² 6nm io die cost just $21 to manufacture and even if 4nm wafer price is 75% more than 6nm, a ~130mm² 4nm die would only cost ~$40. That's very similar cost to a 204mm² 6nm N33.

Now N44 with half the CU of N48, with slightly higher clocks can be over 33% faster than RX 7600 yet 19% slower than RX 7700XT, any MSRP between $270-$300 is acceptable.
The point is for ISO specs(μarch), the cost of more advanced node chip always lower than less advanced node, that's why they keep improving fabrication, to keep their business model running. Otherwise they will not only lose to competition, but also themselves with their last gen products.
I just noticed RX7700 XT has already dropped to $360 on Newegg. And N44 will be 15% slower than that at best.
Well I respect your 15 year service and you are more than welcome to share your thoughts.
 

leoneazzurro

Golden Member
Jul 26, 2016
1,052
1,716
136
I think there is misunderstanding, I agree the video doesn't have full BOM of product but I was specifically talking about 6nm IO die cost and trying to make a point that a similarly sized 4nm N44 doesn't significantly increase the die cost, infact it is about the same as similarly spec 204mm² N33. But Rx 7600 MSRP does include costs of other parts, R&d etc, and the inflation in those parts is not that much and can be easily managed under $270-$300 gap.
OK, I understand your reasoning, the point is that as told raw die cost is only one part of the costs, and in this case the part of the total cost of the VGA board is even smaller, and we have one problem more adding to the costs: the OEM.
The supply chain for the CPU, if bought on the market and not in a complete SKU from a OEM, is AMD->Reseller->Dealer-> user, every one of these steps adds to the cost because every one of these subjects has a margin (they need money to continue to operate). Sometimes the chain lenght is shorter, very often it is longer.
In the case of the GPU, we have AMD->OEM->Reseller->Dealer->user, so an higher cost to market respect to the CPU case.
In the case of a VGA board, we have a lot of components to be added to the GPU, especially the RAM, the cooler (can be very sensitive to raw metals price) and so on and so on.
And, from my experience, the cost of the labour, depreciation, R&D, and so on, adds a lot to the cost of the materials. I work in a power semiconductor company, where the power modules we produce are relatively simple compared to VGA, and even with highly automated production lines and high material cost (Silicon Carbide dies) the labour/depreciation/overhead easily represents half of the direct cost of production. Let's add that, to have a profit, you should add a margin higher than the indirect costs of production...
The point is for ISO specs(μarch), the cost of more advanced node chip always lower than less advanced node, that's why they keep improving fabrication, to keep their business model running. Otherwise they will not only lose to competition, but also themselves with their last gen products.
I just noticed RX7700 XT has already dropped to $360 on Newegg. And N44 will be 15% slower than that at best.
Well I respect your 15 year service and you are more than welcome to share your thoughts.
This was true some years ago, lately unfortunately the $/transistor ratio is not really so favorable with the latest technologies, at least until the machinery is depreciated and R&D expenses are amortised (it takes years, unfortunately).That only to say that AMD's focus in improving PPA has a reason to exist outside the competition with Nvidia. So why there is the need to push for improved processes, even when the costs will not go down (per transistor)? Well, simply because you can pack more transistors and have better performances, so CPU/GPU makers can create the need for new products and keep the market going.
My opinion in RDNA4 will take form when the first reviews will come out, at that point one can evaluate if such a GPU will satisfy his needs or not.
 

Mopetar

Diamond Member
Jan 31, 2011
8,113
6,768
136
seems going off from shelves before RDNA4. We don't know about cutdown N48 perfomance.

Based off recent product generations, I can't see them doing more than a 10% cut. If there's anything binned more heavily than that, it will likely only sold in overseas markets or only to OEMs.

The cutdown N48 die should be nearly as good as the full die.
 

SolidQ

Senior member
Jul 13, 2023
539
629
96
Last edited:

RnR_au

Platinum Member
Jun 6, 2021
2,062
5,006
106
Interesting that's true RDNA4 for SteamDeck 2?
There is a rumour that RDNA 4 will only be in the field for a year. Its a stop gap. Which seems to suggest that RDNA 5 would be available for the steamdeck 2 in that time frame... if Valve is willing to pay for it ofcause. RDNA 4 could be good enough and cheap enough.
 

soresu

Diamond Member
Dec 19, 2014
3,228
2,515
136
Also ye, there's a fair number of published papers on gfx there.

Not nearly so many as nVidia, but then they don't have the money to go nearly so far, not to mention that they haven't had nearly so long back in profitability to build bridges with academia in the gfx/light transport field of study.
 

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
The ambition just keeps digging deeper into the cave of Nothing.
240mm² biggest die, GDDR6 when 7 is available, 18Gbps, and still around $500 because AMD under Lisa wants margins above anything else.
The most sleep-inducing product. I'm so unhyped.
 

branch_suggestion

Senior member
Aug 4, 2023
392
875
96
The ambition just keeps digging deeper into the cave of Nothing.
240mm² biggest die, GDDR6 when 7 is available, 18Gbps, and still around $500 because AMD under Lisa wants margins above anything else.
The most sleep-inducing product. I'm so unhyped.
The gold standard for a new generation historically was 30% perf/$ improvement for mainstream parts. AMD will price to achieve this at a minimum.
 

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
The gold standard for a new generation historically was 30% perf/$ improvement for mainstream parts. AMD will price to achieve this at a minimum.
N48 (whether it's called 8800xt or 8700xt) should have roughly 7% better raster perf than 7900 xt ($800) and about a 4070 TI's RT (also $800).
That takes us to $500. $560 even.
So it will be an arch that fixes RDNA 3 and gets us a BVH walker and a bunch of goodies. It will still produce a thoroughly sleep inducing product that competes with the last gen.
I'm not saying it's a bad product, I'm saying the hype is negative.
 

branch_suggestion

Senior member
Aug 4, 2023
392
875
96
N48 (whether it's called 8800xt or 8700xt) should have roughly 7% better raster perf than 7900 xt ($800) and about a 4070 TI's RT (also $800).
That takes us to $500. $560 even.
So it will be an arch that fixes RDNA 3 and gets us a BVH walker and a bunch of goodies. It will still produce a thoroughly sleep inducing product that competes with the last gen.
I'm not saying it's a bad product, I'm saying the hype is negative.
For a PPA connoisseur like myself I think it is kinda neato.
Gaming segment revenue needs an injection due to consoles sales dropping off a cliff, this does the trick and should increase margins overall, naturally this means a market share gain back to old historical levels to set the foundation for the big boy generation after.