RDNA4

Seems doable if Intel is at 2.8. Remember RDNA3 clocks are “broken”.
Intel is at 2.4:

clock-vs-voltage.png


We'll see how BMG will do on clocks soon.
 
AMD has announced that they are stepping out of the high-end GPU market in the RDNA4 generation, that is sure to put a damper on any hype-train.
I know that, but the thing is most customers don’t actually buy the high end stuff anyways so there’s plenty reason to be excited for all 3 IHVs. We just know nothing and there’s no hype behind any of it, even Nvidia.
 
Oh yes I remember now. Still hoping for a miracle.

I don't think anyone's hoping for a miracle. It likely to compete in the 5060 to 5070 ti range and if it manages to offer similar performance and/or more VRAM that is enough.

3.3-3.4 GHz sustained would be kind of one considering where RDNA3 is.
Seems doable if Intel is at 2.8. Remember RDNA3 clocks are “broken”.

Keep in mind even the 6nm N33 had a boost clock of up to 2755 mhz, somehow higher than the 5nm parts even. Not sure if the clocks are really "broken" or if there was a design bottleneck or whatever, but Strix Point with RDNA 3.5 showed an appreciable increase in perf and perf/W in particular.

Compared to RDNA 3, they should also have the benefits of N4P which should offer slightly higher performance/efficiency. So > 3 Ghz sustained is quite possible. We'll find out in a month anyway.
 
I know that, but the thing is most customers don’t actually buy the high end stuff anyways so there’s plenty reason to be excited for all 3 IHVs. We just know nothing and there’s no hype behind any of it, even Nvidia.
To be fair, most consumers don't buy AMD anyways (looking at Steam numbers)
 
3.3-3.4 GHz sustained would be kind of one considering where RDNA3 is.

RDNA3 do quite high sustained clocks... on some compute loads. It's clearly uneven, in that parts of the GPU can run much faster than other parts. I suspect that moving back to monolithic helps some, and presumably they have put a lot of work on the parts that were a problem with rdna3. And it's on a better process.

I think all of that combined at least keeps the hope alive. Though, AMD hype train has a long history of derailing.
 
AMD has said multiple times mid-range. 7900xtx level performance is not mid-range.
Well the definition of mid-range is kind of fluid as well and performance is generally expected to go up gen on gen. For the next gen from Nvidia I expect more of a performance increase at the high end and relatively lower for mid and low end. But I agree with you, I expect RDNA 4 performance to match 7900xt at best, with the two chips spanning the 5060 to 5070ti range.
 
Where? I can only remember references to skipping highest end which isn't the same thing.
“So, my number one priority right now is to build scale, to get us to 40 to 50 percent of the market faster. Do I want to go after 10% of the TAM [Total Addressable Market] or 80%? I’m an 80% kind of guy because I don’t want AMD to be the company that only people who can afford Porsches and Ferraris can buy. We want to build gaming systems for millions of users.”

80% TAM is Nvidia xx60 TI and below. Not 500mm dies.

Edit: He even mentions the $499 price point of a PlayStation. If AMD could do 4080S performance at $499 they would have.
 
Last edited:
This is not an analysis either way of what RNDA4 will target but just exploring the idea of the mid range and addressing the largest possible market.

This is market share for 4xxx series on the desktop according to Steam -

GPUSteam %RTX 4xxx %RTX 4xxx Cumulative %
4060
3.83​
24.47%​
24.47%​
4060ti
3.18​
20.32%​
44.79%​
4070
2.51​
16.04%​
60.83%​
4070 Super
1.64​
10.48%​
71.31%​
4070ti
1.25​
7.99%​
79.30%​
4070ti Super
0.69​
4.41%​
83.71%​
4080
0.78​
4.98%​
88.69%​
4080 Super
0.74​
4.73%​
93.42%​
4090
1.03​
6.58%​
100.00%​

So it seems like the 4070ti and lower would account for 80% of the desktop market in terms of unit volume. Now in terms revenue share it would likely be higher in term of SKUs, but we can't really calculate that exactly without knowing ASP for Nvidia. But for fun if we used MSRP (yes it's poor) as a stand in -

GPUSteam %MSRPRTX 4xxx %RTX 4xxx Cumulative %
4060
3.83​
$300​
11.74%​
11.74%​
4060ti
3.18​
$400​
13.00%​
24.74%​
4070
2.51​
$600​
15.39%​
40.12%​
4070 Super
1.64​
$600​
10.05%​
50.18%​
4070ti
1.25​
$800​
10.22%​
60.40%​
4070ti Super
0.69​
$800​
5.64%​
66.04%​
4080
0.78​
$1200​
9.56%​
75.60%​
4080 Super
0.74​
$1000​
7.56%​
83.16%​
4090
1.03​
$1600​
16.84%​
100.00%​

Then you see the 4080 and lower would account for about 75% of revenue and the 4080 Super and lower 83% of all revenue.

Now if you look at AMD's perspective for 7xxx desktop product page the 7700xt and 7800xt sit in the middle of the 6 product stack. RNDA3 is really a 3 die product stack, cutting out Navi 31 would still leave the 7800xt and lower serviced by 2 dies.

4080/Super performance sounds high end now but keep in mind last gens high end does get moved down the stack. Nvidia's 5070 or 5070ti at the very least will likely be where the 4080/s sits now. So AMD's biggest die for RNDA4 would only need to be competiting against Nvidia's 2nd biggest die cut down, or even it's 3rd biggest die (and GB205 itself seems like it will sit lower relative to previous 04 dies).

As commentary. Yes I know people have some dislike of the absolute costs of GPUs but in terms of actual relative purchasing the "mid range" isn't as low as people might think.
 
Back
Top