*** AMD "Navi 3X" (RDNA 3 discussion thread) ***

[4K8KW10]

320w will never happen on 15360 shaders even on 5nm. Unless they basically cap the power and lose 25%-50% performance doing that depending how power effects it, could be way less too may come down to a 10% loss. Who knows till they come out.. These cards will start at 400-450w for sure.

It sure doesn't work like that. Overclocking results in minor performance improvements. Maybe 10% max going from 320 watts to 450 watts.
They have to stress on the efficiency and sweet spot part of the curve.

Leave the overclock to those who explicitly need it, not for everyone out of the boxes.

 

[4K8KW10]

I don't think I could deal with the noise, or the hassle of plumbing it in to the fridge freezer.

As it is, my 3060-Ti pumps out a surprising amount of heat. I mean, almost scorching hot. I'm glad I didn't get anything more power hungry! There's a limit to what normal air cooling can deal with. I guess that's why you see modern "gamer" cases with literally 20 fans (all RGB, naturally).

Undervolt your poor card, underclock it as well.

 

[4K8KW10]

What? No, completely wrong. I've been mini-ITX for over a decade now and I can confidently say that nearly all good mini-ITX cases will handle such a card and the required PSU.

The other mini-ITX case contains the rest of the system.

How would you fit a long and fat card into a mini-ITX case?

 

[4K8KW10]

AMD Next-Gen RDNA GPUs Might Feature 3D Infinity Cache Technology, MCM GPUs With 3D Stacking (wccftech.com)

The expected performance improvement of Navi 31 over Navi 21 is impressive:


NVIDIA Ada Lovelace Powered GeForce RTX 40 GPUs To Be Twice As Fast & Twice As Power Hungry Than Ampere (wccftech.com)

 

[4K8KW10]

But will the doubled power consumption & double the price be just as impressive? At a minimum.

If that's true, they better give up and cancel the projects.
Because no one will buy the cards

Nowadays, the inflated scalper GPU prices ruin the CPU sales, imagine if that worsens... Bye-bye PC industry

 

[4K8KW10]

Question - does AMD plan to cut the PCI Express bus width on the Navi 3X GPUs like it has been previously infamously done?

You know integrated Vega in mobile Ryzen gets full x16 bus width, while full desktop versions of much faster graphics cards get cut to x4 or x8 bus width?

 

[4K8KW10]

AMD like Nvidia will give you the performance they think is worthy at the price point. if they want to slow a card down by giving it lower Pci-e bandwidth they will do so. Why cut back the chip more when they can simply limit it through the interface if indeed the card has more to give in the first place.

But I won't buy. Because the decision about the purchase comes from me, it is not a process with one direction in which AMD dictates all conditions. It isn't ok and is ugly.

 

[4K8KW10]

Very bad behavior from AMD on their mid-low end cards. They are cutting the bus and other features as much as they can and then they clock the cards as high as they can to make their performance somehow attractive. Instead of giving you a 15-20% overclock potential, you get a card that is pushed to the limits from the factory.

Yeah, this ^^.
AMD is NOT in the position to do shenanigans. AMD offers slower products overall and AMD has lower market share overall.
I'd recommend AMD to start increasing the performance of the products it offers, not the opposite - to decrease that performance.

 

[4K8KW10]

But but but... AMD is going to give them a 7900xt next time with 3 x the performance of a 6900xt ...

Spoiler: But they forgot it will come with...

3x to 4x the price too at the fake msrp if even true, which I doubt as any such card would end up as a server gpu not a gaming gpu and if true again be ready to invest in a £400 psu too that of course at the time will be scalped to double.... related to this thread here :- https://forums.overclockers.co.uk/threads/rx-7900xt-15-360-cores-mcm-tapeout-q4.18933812/

But Nvidia promises supply issues will ease down later this year:

NVIDIA Says GPU Shortages Are Going To Ease Down in 2H 2022, Right Before Next-Gen GeForce RTX 40 Series Hits Market Shelves (wccftech.com)
NVIDIA Says GPU Shortages Are Going To Ease Down in 2H 2022

 

[4K8KW10]

Also next round of cards will start at an all new FE msrp (of course will be fake again and scalped again)..


4080 FE $1000-$1200 MSRP for sure next time

4080Ti FE $1500-$1750 MSRP

4090 FE $2000-$2500 MSRP or a Titan 40RTX at $2500-$3000

4090Ti FE $2500-$3000 MSRP or a Titan 40RTXp at $3000-$3500

But I think 4090 class cards will not exist and will be Titan cards next time, this time because AMD messed up Nvidia's plans is why we got a 3090 type card.


Go read what AMD has done with Epyc cpu pricing recently...

https://www.tomshardware.com/uk/new...0-to-30-intels-sapphire-rapids-delayed-report


They will soon be 50% more expensive.


Also I will link you to this post and what I said there :- https://forums.overclockers.co.uk/posts/35306709/

If AMD, Intel and Nvidia want to bankrupt their graphics businesses, they are free to try infinite prices rises, it is a balloon that will explode and they will not like it

I know that their management teams are stupid and hate sustainability, but we have a society to serve and this society says no.

 

[4K8KW10]

Society said yes and paid the silly prices at 2x and 3x msrp and always out of stock, they sold out so fast as we saw and Nvidia has sold more cards than ever this generation and still people on waiting preorder lists for over a year. As stated before...

Look at the thread "Did you already give up on new graphics card". Anyone just given up on looking for a new GPU? | Page 189 | Overclockers UK Forums

"Out of stock" means nothing if we don't know the real sales numbers.

 

[4K8KW10]

The Radeon RX 7900 XT is going to be a monster.
I don't know if the exact specifications will remain like that, but it's going to be very very fast:


AMD Radeon RX 7900 XT Specs | TechPowerUp GPU Database

 

[4K8KW10]

Those specs don't make it look like a monster, just an expected normal next gen jump.

Infinity cache quadrupled from 128 MB to 512 MB.
Texture rate up from to 648.0 GTexel/s to 1,920 GTexel/s.
Shaders up from 4,608 to 15,360.

 

[4K8KW10]

Certainly a very strong signal that AMD wish to compete at the top of the market.

Prices and availability are the big unknown. But even if wafer supplies improve - even if consoles are being re-spun it should hopefully be for 6nm while these should be on 5nm - that does not mean supply will be plentiful. Simple fact is, at 5nm the same thing will apply: even the cheapest 5nm Zen4 CPU (let's say an Athlon of which they haven't even released any Zen3 variants yet) will have higher margins per wafer area than any GPU.

On the other hand, it seems crazy to design these big RDNA3 cards (and the sunk cost of 5nm GPUs will be huge) and then not allocate any wafers for them. However, AMD have done this (design a chip and then barely ship it) before, so what do I know?

I think that it depends on the crazy bitcoin mining demand. There are already signs that things get slowly improvement, the graphics cards prices are lower.

 

[4K8KW10]

slower clock speeds. Would have expected them to be faster.

Higher clocks destabilise the sweet spot power efficiency metric and curve. You look at the fresh example of the terribly overclocked RX 6500 XT.

Just give us a rest already and clock the cards conservatively.

Everyone knows how to overclock manually..

 

[4K8KW10]

Speculations on the specifications:


AMD RDNA 3 GPUs For Radeon RX 7000 Graphics Cards Confirmed To Feature 5nm & 6nm Process In MCM & Monolithic Designs (wccftech.com)

 

[4K8KW10]

Greymon leak:

first RDNA3 GPU to be available will be Navi33, the 7600XT. It's a monolithic 6nm GPU and it's on par or slightly faster than the 6900XT

https://www.hardwaretimes.com/amd-r...ill-be-the-first-rx-7000-gpu-to-launch-rumor/

Probably fake news. #fakenews

Why?
Because the N7 Navi 23 is 237 sq. mm with 11.06 B transistors. 46.66 MTransistors/sq.mm.
On the other hand, the N6 Navi 24 is 107 sq. mm with 5.4 B transistors. 50.46 MTransistors/sq.mm.

So, actually, the N6 is a slightly improved process.

How will they increase the performance of Navi 33 over Navi 23 with over 80% if they don't make the die prohibitively large for this tier - over 400 sq. mm?


AMD Radeon RX 6600 XT Specs | TechPowerUp GPU Database