RDNA 4 speculation: chip size of N43 (3nm) = 5700 XT & N44 (4nm?) = 7600

M

Marees

2[H]4U
Joined
Sep 28, 2018
Messages
2,117
MLID source claims AMD looking at 64 CU for Navi 43 & 32 CU for Navi 44

MLID speculates that AMD is mainly targetting laptop designs with these 2 chips

& in future if there is remergence of chip shortage (due to crypto/AI etc) then AMD is ready with their complex & multi-chip "big Navi" designs to cater to the high end "demand"


View: https://youtube.com/watch?v=7cEKTr70YBY&t=855s
 
My 2 cents:

If AMD had planned for these chip sizes BEFORE scrapping big RDNA 4, then it makes sense to increase bus width of the remaining N43 & N44 (n44 especially because n43 might be constrained by laptop limits)

So increase the 32 CU of navi 33 to 40+ & die shrink to 4NP with 128 bus (12gb VRAM) for the performance of a 6800 / 6800 xt (with much better RT ? ) in $300 to $400 range & that should make a good 1080p high/max card

But most importantly release it within 1 year. Don't delay it unnecessarily like navi 33 was delayed for 9 months, despite being in stock !!!
 
This is the exact strategy I said could work for AMD in the now-locked thread about them ditching the high end. Focus on laptop parts that can double as desktop parts, and move on to the next gen as fast as possible. The key to this is that AMD needs to lead the market when it comes to pricing.

This could be their Polaris 2.0 moment. It's their victory to lose.
 
News from the same leaker, who claimed that big RDNA 4 has been scrapped:

The smaller RDNA 4 chip(s) — Navi 43 and/or Navi 44 is/are on track to release next year. (Likely H2 2024)

This could force nvidia to release a auper version of 4060 ti / 4070 / 4070 ti as they feel the competitive heat.

Keep in mind, that there is many a slip between the cup & the lip, and anything can happen in next 1 year.

For example same leaker hyped 4ghz of RDNA 3. While that leak was proven true by benchmarks, in real gaming usages, due to power leakage, RDNA 3 could never hit above 2.8ghz !!!
 
Speculations are that RDNA4 will be an optimized RDNA3 with more features to enhance speed, not new functions. The drivers and developpement will then be up to the hardware AMD puts on sale, which is currently not the case for some time now. There won't be any RDNA 4 card to upgrade the RX 7900XT/XTX and RDNA 4 cards will stop at middle range.
On the contrary there may be a big gap with RDNA 5 which will bring back high end cards maybe on par with Nvidia high end (I mean xx90). Actually AMD is kind of focusing on RDNA 5 and leaves the RDNA4 as an in between update.
 
Last edited:
Jandor said:
Speculations are that RDNA4 will be an optimized RDNA3 with more features to enhance speed, not new functions. The drivers and developpement will then be up to the hardware AMD puts on sale, which is currently not the case for some time now. There won't be any RDNA 4 card to upgrade the RX 7900XT/XTX and RDNA 4 cards will stop at middle range.
On the contrary there may be a big gap with RDNA 5 which will bring back high end cards maybe on par with Nvidia high end (I mean xx90). Actually AMD is kind of focusing on RDNA 5 and leaves the RDNA4 as an in between update
Click to expand...

What we know for sure is that packaging demand unexpectedly tripled at TSMC & they are able to meet only 80% of capacity right now. They are ramping up but it will take atleast until end of next year for supply to catch up.

AMD also has expressed ambitions of using FPGA for AI.

Now both FPGA & big radeon (Navi 41 & Navi 42) compete for packaging. This issue doesn't exist for monolithic small radeon chips such as Navi 43 (rumoured to be a monolithic die shrunk navi 32) & Navi 44 (rumoured to be die shrunk navi 33)

so assuming AMD can get big RDNA 5 (navi 51/navi 52) ready by late 2025/early 2026, then it makes sense to speed up small RDNA 4 release (navi 43 & navi 44), as if it is a fixed version of RDNA 3 (aka RDNA 3.5)

Only question is how soon can AMD release navi 43 & navi 44?
Rumour is that atleast one of them has taped out & all going well (fingers crossed) is expected to be released in 2024 itself.

Marees said:
Until TSMC can bring additional capacity online, Nvidia's H100 and older A100 – which power many popular generative AI models, such as GPT-4 – are at the heart of this shortage. However, it's not just Nvidia. AMD's upcoming Instinct MI300-series accelerators – which it showed off during its Datacenter and AI event in June – make extensive use of CoWoS packaging technology.


AMD's MI300A APU is currently sampling with customers and is slated to power Lawrence Livermore National Laboratory's El Capitan system, while the MI300X GPU is due to start making its way into customers' hands in Q3.

We've reached out to AMD for comment on whether the shortage of CoWoS packaging capacity could impact availability of the chip and we'll let you know if we hear anything back.


https://www.theregister.com/AMP/2023/09/08/tsmc_ai_chip_crunch/
Click to expand...
 
You must log in or register to reply here.
Back
Top