RDNA 3 Rumor Discussion

[bill dennison]

[pax]

The navi 33 specs sound more realistic...

[bill dennison]

Originally Posted by pax

The navi 33 specs sound more realistic...

I only care about 31

[pax]

I think 31 will be close to rtx4k but it may beat it handily in TDP. Wouldnt be surprised after all the leaks to see it be ~100-150w less than nvidia's top end.

Last I heard navi 31 tops out at 450w. Thatd be pretty amazing considering they are on the same 5nm node at TSMC. But thats what a solid cpu division gets you. Really good power optimization.

[bill dennison]

Originally Posted by pax

I think 31 will be close to rtx4k but it may beat it handily in TDP. Wouldnt be surprised after all the leaks to see it be ~100-150w less than nvidia's top end.

Last I heard navi 31 tops out at 450w. Thatd be pretty amazing considering they are on the same 5nm node at TSMC. But thats what a solid cpu division gets you. Really good power optimization.

i think it going to smoke RTX 4000 all but the top 4090 ti 3 mile island 800+ watt card that they build only enough for a few chosen reviewers and LN2 benchmarkers

if AMD releases first and is even close to as good as they say one of two things will happen with NV

1.. the 4080 and below will be DOA for all but diehard fans
maybe even the 4080 ti

2 ... or there will be a long delay to Hopper

[pax]

Nvidia could speed up their next gen they have the money for it but itd be summer-fall 20203 at the earliest. TSMC can only do so much. Itd give AMD a good 6-9 months at the top. Its probably due to the chiplet system that they can be that much lower tdp and Nvidia is behind on that note.

[Nunz]

Originally Posted by bill dennison

i think it going to smoke RTX 4000 all but the top 4090 ti 3 mile island 800+ watt card that they build only enough for a few chosen reviewers and LN2 benchmarkers

if AMD releases first and is even close to as good as they say one of two things will happen with NV

1.. the 4080 and below will be DOA for all but diehard fans
maybe even the 4080 ti

2 ... or there will be a long delay to Hopper

yeah you said the same thing about the 6xxx series

[pax]

6000 series was way better than I thought it would be after the mediocre 5000. In pure raster its an amazing boost and beats the 3090. They do that again and it can pass Nvidia easy. Nvidia going large monolithic die is its achilles heel.

[bill dennison]

Originally Posted by Nunz

yeah you said the same thing about the 6xxx series

there are a lot of IF's about both the 7000's and 4000's at this point

but if these rumors are true or even close to true AMD will have a killer card

and NV will have a un-green power sucking card even if the top card does win

[Nunz]

Originally Posted by pax

6000 series was way better than I thought it would be after the mediocre 5000. In pure raster its an amazing boost and beats the 3090. They do that again and it can pass Nvidia easy. Nvidia going large monolithic die is its achilles heel.

No, the Achilles heel was NV going with an inferior node process. NV has corrected that going back to TSMC. I wouldn't be so impressed with the 6900XT beating the 3090 in Raster when the 6900XT was on a refined 7nm node while the 3090 was using Samsung's crap 8nm.

It's the same deal as people being so impressed with Ryzen 3000 and 5000.. it was competing against 14nm+8000. AMD looks good on paper until you realize they are keeping up due to a process advantage, not because of a design advantage. With NV going to TSMC, I wouldn't be so confident anymore; the process advantage is dwindling, especially with the power usage staying on the high-end.

The Samsung process was really .. really bad. We knew Samsung was a poor choice (a lot of us thought that was a mistake) to begin with but I think Ampere was Samsung's last chance to get it right. They didn't.

[acroig]

Originally Posted by bill dennison

but if these rumors are true or even close to true AMD will have a killer card

and NV will have a un-green power sucking card even if the top card does win

If the rumors are true both 7000 and 4000 top end will be "un-green power sucking cards."

When did you become an eco warrior?

[pax]

Originally Posted by Nunz

No, the Achilles heel was NV going with an inferior node process. NV has corrected that going back to TSMC. I wouldn't be so impressed with the 6900XT beating the 3090 in Raster when the 6900XT was on a refined 7nm node while the 3090 was using Samsung's crap 8nm.

It's the same deal as people being so impressed with Ryzen 3000 and 5000.. it was competing against 14nm+8000. AMD looks good on paper until you realize they are keeping up due to a process advantage, not because of a design advantage. With NV going to TSMC, I wouldn't be so confident anymore; the process advantage is dwindling, especially with the power usage staying on the high-end.

The Samsung process was really .. really bad. We knew Samsung was a poor choice (a lot of us thought that was a mistake) to begin with but I think Ampere was Samsung's last chance to get it right. They didn't.

No I mean the monolithic die in RTX 4k. The leaks about 600w TDPs is pretty much related to the fact they stuck to a single large die imo.

[bill dennison]

Originally Posted by acroig

If the rumors are true both 7000 and 4000 top end will be "un-green power sucking cards."

When did you become an eco warrior?

1982

i recycle tons of copper and aluminum and freon every year

and just started a 19.2kw solar system for the house .... well set up for Tesla solar to set up a system design and price about 50k for it one battery
if they won't give me the full 19.2 kw system i'll call someone else
got to get it done before my 2022 ford f150 lighting gets ready in the next year or two

i'm just not a eco fanatic


and yes the 7950 will be bad at 500 watts but the 7900 may well be better

but Lovelace is on a much higher porn level of sucking power

[acroig]

Originally Posted by bill dennison

but Lovelace is on a much higher porn level of sucking power

ROFL!!!

[bill dennison]

Originally Posted by acroig

ROFL!!!

i would love to go to the NVidia launch event and ask the jacket

"Is it true you named it Lovelace because it deepthroats your PSU "

[acroig]

Brutal!

[pax]

https://www.pcgamer.com/nvidia-rtx-4...ign=socialflow

Those 600W+ TDP rumours are looking more and more likely.

Lovelace GPUs are set to be manufactured on TSMC's 4N fabrication process which is tweaked specifically for Nvidia. Despite the significantly smaller node, those rumours of 600W or even higher TDPs just won’t go away. Bear in mind we’re talking about the enthusiast class cards here, where power consumption is typically secondary to performance. We hope that the mid range and entry level cards will end up a lot more power thrifty. Imagine the hassles system integrators will face if they have to accommodate a 300W entry to mid level GPU.

Thats interesting in that I thought it would be on 5nm like Navi 3. To go that far on what is probably a very expensive node shows they have competition in the AMD 7000 gen.

[Megaman]

What about those rumors about 800w? That’s right, full of ****.

[acroig]

Originally Posted by Megaman

What about those rumors about 800w? That’s right, full of ****.

Those were for Ada I think.

[pax]

Originally Posted by Megaman

What about those rumors about 800w? That’s right, full of ****.

Only a few said 800w most said 600w. I think its a dead giveaway, besides the reliable industry leakers like greymon, that for them to go 4nm at the early higher cost that is in any case generally silicon these days, that they have a power

consumption issue.

Igor has another good breakdown here:

https://www.igorslab.de/en/where-the...nd-components/

2x the power of the 3090 for the gpu die alone...

[Nunz]

Why do you think that they would go for a smaller node specifically for a power consumption issue? While it is irregular for NV to go for an advanced node, I'm not sure that it's correlated to the power consumption. If the 4nm process was specifically modified for NV, it could just be learning from the mistakes of going Samsung 8nm for Ampere. Now that there's competition, NV isn't taking the chance of gunning for maximum profit margin. They made money hand-over-fist using Samsung instead of TSMC for Ampere.

I don't think the 4090 will have 12GB of vRAM, though, unless they bring back the TITAN name and relegate the 4090 to "4080TI early access".

[badsykes]

If this 8gb is true for 7700 xt it will one of the most greedy launch i witnessed.

https://wccftech.com/amd-radeon-rx-7...hics-detailed/

AMD made 7700xt as powerfull as a 6900xt and faster in RT and you put .... 8gb ram
I saw GN calling amd greedy with the new cpus like 5500 / 5600 and now they some sheningans with 7700 xt.


Seriously ? Is this a reason so the buyers of 6900 xt should not feel so bad ?
I even thought the 3070/3070ti getting only 8gb was kinda crappy.Now is AMD turn to do a bang with 8gb on a 2022 7700xt ...

AMD is going greedier and greedier.

[LordHawkwind]

Originally Posted by badsykes

If this 8gb is true for 7700 xt it will one of the most greedy launch i witnessed.

https://wccftech.com/amd-radeon-rx-7...hics-detailed/

AMD made 7700xt as powerfull as a 6900xt and faster in RT and you put .... 8gb ram
I saw GN calling amd greedy with the new cpus like 5500 / 5600 and now they some sheningans with 7700 xt.


Seriously ? Is this a reason so the buyers of 6900 xt should not feel so bad ?
I even thought the 3070/3070ti getting only 8gb was kinda crappy.Now is AMD turn to do a bang with 8gb on a 2022 7700xt ...

AMD is going greedier and greedier.

At best a mid range card not an enthusiast card. It is what it is I guess

[pax]

I dont get that, it should be the 7600 not the 7700. The 7700 should be navi 32. It has the basic low 128 bit mem bus. 8 gigs in that case makes sense.

IMO its gonna be a 1080p card for new engines like Unreal 5.

[badsykes]

Originally Posted by LordHawkwind

At best a mid range card not an enthusiast card. It is what it is I guess

It is replacing the 6700xt series.This one has 12gb vram now.I hope it's a typo and as pax said it is about 7600 series.

[Sasquach]

Doesnt a more efficient memory subsystem, internal cache or faster internal cache memory and bus size make up for lack of capacity?

Maybe they're banking on improvements this way?
It should at least have the same amount of memory as the product its succeeding.

[demo]

Perhaps. I do remember a few generations ago (in the 1 or 2GB days?) they were able to somewhat make up for capacity with driver trickery of some sort. I believe 'delta colour compression' was the term and it had verifiable results.

But, unless there's some sort of significant technology like this, I wouldn't hold my breath. We'll know soon enough from the marketing.

[Mangler]

Maybe they plan on making the 79xx series the 68xx replacement as an homage to the 7950 and 7970?

[bill dennison]

AMD RDNA 3 GPUs rumored to feature up to seven chiplets

We already know that AMD is going to tap TSMC for the manufacturing of these new GPUs, but Greymon55 seems to believe the company will go with a mix of process nodes for the various chips involved. For instance, the Navi 31 GPU may be comprised of two 5nm Graphic Complex Dies (GCD), four 6nm Memory Complex Dies (MCD), and an interconnect controller die.

https://www.techspot.com/news/94174-...-up-seven.html


...........

Leak Suggests AMD is Confident It Will ‘Decimate’ Nvidia's Efficiency with RDNA 3

https://gamerant.com/amd-confident-d...ciency-rdna-3/

[pax]

Sounds like they are bringing back HBM.