*** The AMD RDNA 4 Rumour Mill ***

[muon]

How much was the 529 mm 7900 gre again? Nvidia is ripping people off with 5070 spec at a grand.

Navi 31 actually only had a 304mm^2 GCD and up to 17% of the chip could be defective and still be usable in a 7900 GRE. The tiny 37mm^2 MCDs which probably had close to 100% yield was on a much cheaper process.

Whilst the poster is obviously wrong in the price, this post isn't really a rebuttal. The above doesn't compare to a fully enabled 390mm^2 die.

 

[TheRealDeal]

Agreed. Well said.

Not like Nvidia can't bring out supers in 12 months if it turns out vram is becoming an issue and impacting sales.

Amd just need to release their cards with as small margin as they can bare this time round. Stay relevant and maybe even claw some market share back and hope their next gen brand new architecture is a hit.

I am more excited about their next gen udna cards tbh. Naively hoping it is the chiplets design we have been hearing for years.

I can see the Supers coming earlier if AMD have something good and eat into Nvidia's market share as what Nvidia are bringing looks to be pretty poor. It's not suprising if it does happen. Nvidia used a better process node last Gen and now there is parity so AMD are no longer at a disadvantage in this regard. Nvidia are a smart company so won't hang around. It will take something pretty special from AMD to force there hand as the mindshare will probably get them through any how. 7900xtx performance right on top of the 5070 price should do it but gpu buyers are crazy so maybe not.

 

[FredFlint]

Nvidia must be paying more for memory (GDDR7) than AMD (GDDR6). Would be interesting to know how much the difference is.

 

[TheRealDeal]

Nvidia must be paying more for memory (GDDR7) than AMD (GDDR6). Would be interesting to know how much the difference is.

Yea AMD can get away with it because of the infinity cache. I am surprised Nvidia have not developed something similar to make more profits using older memory types.

 

[Phixsator]

the 9070 better come soon and for a good price cause I'm already loathing my temporary 980ti I just put in the machine this evening.

 

[muzzydark]

I can see the Supers coming earlier if AMD have something good and eat into Nvidia's market share as what Nvidia are bringing looks to be pretty poor. It's not suprising if it does happen. Nvidia used a better process node last Gen and now there is parity so AMD are no longer at a disadvantage in this regard. Nvidia are a smart company so won't hang around. It will take something pretty special from AMD to force there hand as the mindshare will probably get them through any how. 7900xtx performance right on top of the 5070 price should do it but gpu buyers are crazy so maybe not.

It might even happen a soft transition without changing labels (less than a year for "supers" is not a good impression for the company) letting AIB's throwing 3GB VRAM modules on their PCB and boom 24GB 5080, 18GB 5070, 12GB 5060

 

[Bidley]

I mean would any of you pay £650+ for a 9070XT or £550+ for a non-XT?

I would in a heartbeat if the 9070xt sat ahead of the 5070ti in raster and RT, and MFG was the only difference. In fact I would even if performed the same.

I'm not expecting to pay less just because it's got a Radeon logo on it. The reality is, if I don't get a 9070xt then my options are; pay more for the equivalent Nvidia card or stick with my 3080 (that isn't performing how I want).

Price to performance is what I look for in an upgrade, not the likelihood of it gaining the corporation any (sigh) marketshare.

 

[Sumanji]

It seems that AMD wanted the XT to sell for $999 and the non-XT to sell for $799, but nvidia's cheaper pricing on the 5070/ti messed everything up.

Dude, you need to at least source if you're gonna post outrageous claims like that

I'm not expecting to pay less just because it's got a Radeon logo on it.

Sense++

 

[RavenXXX2]

I'm not expecting to pay less just because it's got a Radeon logo on it.

Frank Azor, is that you?

 

[TheRealDeal]

It seems that AMD wanted the XT to sell for $999 and the non-XT to sell for $799, but nvidia's cheaper pricing on the 5070/ti messed everything up.

I take it AMD never factored in the 5080 at $999 either. Best laugh i have had in a while.

 

[muon]

Yea AMD can get away with it because of the infinity cache. I am surprised Nvidia have not developed something similar to make more profits using older memory types.

You do know that "infinity" cache is just cache right.

The 7900 XTX has 96MB of L3 cache ("infinity" cache).

AMD Radeon RX 7900 XTX Specs

AMD Navi 31, 2498 MHz, 6144 Cores, 384 TMUs, 192 ROPs, 24576 MB GDDR6, 2500 MHz, 384 bit

www.techpowerup.com

The RTX 4080 has 64MB of faster L2 cache (no special branding).

NVIDIA GeForce RTX 4080 Specs

NVIDIA AD103, 2505 MHz, 9728 Cores, 304 TMUs, 112 ROPs, 16384 MB GDDR6X, 1400 MHz, 256 bit

www.techpowerup.com

edit:

Just to add a little more context. When "infinity" cache did come out.

The 6800 XT had 128MB of L3 cache

AMD Radeon RX 6800 XT Specs

AMD Navi 21, 2250 MHz, 4608 Cores, 288 TMUs, 128 ROPs, 16384 MB GDDR6, 2000 MHz, 256 bit

www.techpowerup.com

massively beating the 5MB L2 cache of the RTX 3080

NVIDIA GeForce RTX 3080 Specs

NVIDIA GA102, 1710 MHz, 8704 Cores, 272 TMUs, 96 ROPs, 10240 MB GDDR6X, 1188 MHz, 320 bit

www.techpowerup.com

but AMD has reduced cache since then.

 

[TheRealDeal]

You do know that "infinity" cache is just cache right.

The 7900 XTX has 96MB of L3 cache ("infinity" cache).

AMD Radeon RX 7900 XTX Specs

AMD Navi 31, 2498 MHz, 6144 Cores, 384 TMUs, 192 ROPs, 24576 MB GDDR6, 2500 MHz, 384 bit

www.techpowerup.com

The RTX 4080 has 64MB of faster L2 cache (no special branding).

NVIDIA GeForce RTX 4080 Specs

NVIDIA AD103, 2505 MHz, 9728 Cores, 304 TMUs, 112 ROPs, 16384 MB GDDR6X, 1400 MHz, 256 bit

www.techpowerup.com

My understanding is Infinity cache boosts bandwidth. I ain't no expert on it but it's called infinity cashe for a reason and not just cache.

https://forums.overclockers.co.uk/threads/the-amd-rdna-4-rumour-mill.18984360/page-35

Maybe it was just the 6 series where it helped and Nvidia did use something similar for the 4 series.

 

[muon]

My understanding is Infinity cache boosts bandwidth. I ain't no expert on it but it's called infinity cashe for a reason and not just cache.

https://forums.overclockers.co.uk/threads/the-amd-rdna-4-rumour-mill.18984360/page-35

With smaller buses AMD don't seem to be at any disadvantage using slower gddr.

This is just AMD marketing. It's all just L1, L2 and L3 cache. Nvidia doesn't use L3 cache, but that's because they have way more L2 cache which is faster than L3 cache.

All cache helps a processor. Ideally you'd have lots of L1 cache, but that would sacrifice other things in a chip design.

 

[TheRealDeal]

This is just AMD marketing. It's all just L1, L2 and L3 cache. Nvidia doesn't use L3 cache, but that's because they have way more L2 cache which is faster than L3 cache.

All cache helps a processor. Ideally you'd have lots of L1 cache, but that would sacrifice other things in a chip design.

AMD's Infinity Cache improves memory bandwidth and performance compared to Nvidia's GPUs by using a smaller memory bus and less expensive VRAM.

Explanation

• Memory bandwidth: Infinity Cache improves memory bandwidth without using a more power-hungry 512-bit memory bus.
  
  
• Performance: Infinity Cache improves performance by reducing page faults and fetch cycles to VRAM.
  
  
• Power savings: Infinity Cache improves performance while using less power and latency.
  
  
• Cost: Infinity Cache allows AMD to use less expensive VRAM.
  

Infinity Cache is a small memory store that's located between the L1 and L2 caches and the GDDR6 and DDR4 memory. It's designed to help reuse temporal and spatial data when rendering complex images.


Infinity Cache was introduced with the RDNA 2 architecture and is found in the following graphics cards: RX 6800 XT, RX 6800, and RX 6900 XT.

According to this it's a bit more than that.

 

[muon]

AMD's Infinity Cache improves memory bandwidth and performance compared to Nvidia's GPUs by using a smaller memory bus and less expensive VRAM.

Explanation

• Memory bandwidth: Infinity Cache improves memory bandwidth without using a more power-hungry 512-bit memory bus.
  
  
• Performance: Infinity Cache improves performance by reducing page faults and fetch cycles to VRAM.
  
  
• Power savings: Infinity Cache improves performance while using less power and latency.
  
  
• Cost: Infinity Cache allows AMD to use less expensive VRAM.

Infinity Cache is a small memory store that's located between the L1 and L2 caches and the GDDR6 and DDR4 memory. It's designed to help reuse temporal and spatial data when rendering complex images.


Infinity Cache was introduced with the RDNA 2 architecture and is found in the following graphics cards: RX 6800 XT, RX 6800, and RX 6900 XT.

According to this it's a bit more than that.

That's just what cache is.

It's like how the X3D chips have a lot of L3 cache.

 

[TheRealDeal]

That's just what cache is.

So then why when it was introduced did AMD cards not suffer against the 30 series which used faster memory and bigger buses.

 

[muon]

So then why when it was introduced did AMD cards not suffer against the 30 series which used faster memory and bigger buses. It had to do something as AMD named it as if it was a big new feature.

They just had more available cache than Nvidia. Nvidia also has cache on their chips.

AMD, Nvidia and Intel like coming up with fancy names for things that have alternative, boring but more accurate, names.

 

[TheRealDeal]

They just had more available cache than Nvidia. Nvidia also has cache on their chips.

"Infinity Cache is a small memory store that's located between the L1 and L2 caches."

So that's normal then and what you called the L3 cache. Whatever they did it did seem to work for the 6 series.

 

[muon]

"Infinity Cache is a small memory store that's located between the L1 and L2 caches."

So that's normal then

It isn't located between L1 and L2. Read you sentence properly.

"Infinity Cache is a small memory store that's located between the L1 and L2 caches and the GDDR6 and DDR4 memory."

It would actually sit between L1/L2 and the GDDR6 memory. That is the L3 cache. If your L3 cache has what you need, then you don't need to go to the GDDR6 memory to retrieve information. (If your L2 cache has what you need, then don't need to go to L3 cache either).

Nvidia would have have the exact same thing, expect in Ada, they have a tonne of L2 cache to store information and no L3 cache.

 

[TheRealDeal]

I copied and pasted it straight from a google search. Amd talked about it boosting bandwidth themselves. I know nothing of the technical details just that AMD were excited about it and there 256bit bus and standard gddr 6 was enough to match Nvidia with a bus size of 320 and gddr6x. Seems a bit like a 2 litre turbo hot hatch matching a 5 litre na because it was turbocharged.