Soldato
RDNA2 was fine.so how come AMD can make efficient cpu's but not efficient gpus. Radeon(AMD) division haven't hired intels cpu dept by any chance have they
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD GPU sales tanking
- Thread starter tamzzy
- Start date
because cpu and gpu are different things.so how come AMD can make efficient cpu's but not efficient gpus. Radeon(AMD) division haven't hired intels cpu dept by any chance have they
architecture makes a bigger difference. nvidia didnt have a process node advantage over amd in the past and were still able to be more efficient
Caporegime
I've explained it at the top of this page, its silly how many unnecessary volts AMD pump their GPU's with, they need to find a way of fine tuning them on mass... like Nvidia do.
Last edited:
Caporegime
Don't just dismiss 5nm vs 4nm out of hand because its inconvenient to the agenda, 4nm is 22% more efficient than 5nm.
en.wikipedia.org
5 nm process - Wikipedia
Caporegime
If there was no difference between them, like some are trying to make out then AMD wouldn't be moving off 5nm to 4nm for Zen 5.
Its more efficient, and by quite a chunk.
Its more efficient, and by quite a chunk.
Last edited:
Soldato
This is sad I thought they were doing OK to .
As stated before it's basically clickbait, its factoring in consoles as well and not just discrete gpus.
Better to wait and see, I doubt 7900s are gonna vanish overnight when the new boards arrive.
I think others have covered this, but console SOCs really would not work with DIMMs at all. So unless those LPDDR5X module become a thing, any such SOC has to be a soldered motherboard. Not that LPDDR5X is fast enough really compared to GDDR. It has far better latency but not bandwidth and consoles generally go for bandwidth.
Also, the console SOCs really are semi or more custom - didn't the PS5 drop something which would be crazy for a mainstream CPU? Certainly on a console where developers have to use the console vendors compilers and tools, dropping x87 or even SSE is totally possible.
Volumes and R&D budget make a huge different. While we have to back to 28nm, Hawaii beat GK110 pretty well: it was compute monster and was only 440mm² vs GK110's 560mm².
After that? Well Nvidia were already raking it in on the server and workstation market and Maxwells was the start of splitting the two while AMD with Fury and Vega etc tried to serve both markets.
The last few gens, a lot of efficiency is/was how much die gets thrown at cache, and which node (but cheaping out with Samsung worked quite well for Nvidia with the GPU shortage - however if AMD hadn't using such a huge % of their TSMC 7nm allotment for console and didn't also have far more profitable Zen3 on TSCM 7nm... Well if they had gone for a volume push then RDNA2 vs Ampere could have gone quite differently. This gen Nvidia played it save and paid up the TSMC premium.
wait till 4 angstrom is out.
As Jen Psaki would say.... "I'll circle back to that"
I can't read it right now I'm dying with hay fever and I'm Oliver twist. Good night Gentleman. *
*And fair ladies (doubt there's any here but worth a shot)
Last edited:
RDNA 3 would have looked fine also if Nvidia had chosen to go with an inferior Samsung node again.
Associate
- Joined
- 13 Sep 2010
- Posts
- 2,027
Happy ending?
Caporegime
Associate
- Joined
- 13 Sep 2010
- Posts
- 2,027
Yes..
.... Jokes aside, I really should tweak my 7900XT at some point, Seen it hit a reported 500+W occasionally (think it's usually more in the 350-370 range).Is it just a case of setting a static offset like in the good ol' days, or is it a curve based thing like Ampere was?
Soldato
RDNA2 and RDNA1 were on the same 7NM node. The RX6600XT(Navi23) beat out the RX5700XT(Navi10) using less memory bandwidth, whilst introducing RT for pretty much the same amount of transistors and similar chip size. It was also 33% more energy efficient than an RX5700XT. Considering Vega 7 was on 7NM,looking at TPU, AMD got over a 50% improvement on the same node in under three years.
The RTX3060(GA106) had more transistors, more memory bandwidth, etc but overall isn't faster in most games.
Last edited:
Caporegime
Is it just a case of setting a static offset like in the good ol' days, or is it a curve based thing like Ampere was?
I can get mine to 300 watts, it tops out there, but with the undervolt i need to run it at around 3.1Ghz, which it will do quite happily but its pointless, its doesn't scale past about 2.9Ghz, which is interesting and i know why, but i might make a dedicated post about that, you see clock for clock this 60 CU GPU is about 30% faster than the 60 CU RX 6800.
But the 80 CU RX 7900 GRE is clock for clock only about 5% faster than the 80 CU 6950 XT, odd that isn't it?
Last edited:
Laaate reply, but high end APU would choke on DDR4, not enough bandwith (they said)
Soldato
They are making one for laptops with a 40CU IGP called Strix Halo:
AMD reaffirms Zen5-based "Strix Point" APUs on track for 2H 2024 launch - VideoCardz.com
Zen5 APUs: Strix Point on track to launch this year, Strix Halo reportedly features 40, 32, 24 and 20 CU RDNA3.5 variants AMD is reaffirming that Zen5 Ryzen products will be introduced this year. CEO Dr. Lisa Su claims that the introduction of these next-gen Ryzen “Strix Point” mobile processors...
videocardz.com
Last edited: