Official AMD Radeon RX 7000 Series (RDNA 3) Thread

[Nunz]

With the fire sales on the 6000's and the keynote that FSR3 may be enabled on them we might see them go past the 3080's in not just raster but RT in those games that use FSR3. That would be a remarkable change if that happens.

Would make the 6000's the one to reach for as the prices finally drop.

Another reason why this frame gen stuff is horrible for the market. Frame generation being compared to a GPU not using it.. sigh.

[pax]

Its not gonna make sense in every game but unless nvidia puts dlss3 in 3k gen it might not age as well as the 6k's. I dont think its a first gen kind of thing its just a driver issue. We'll see DLSS/FSR3.1 then 3.2 ect...

Another thing is its getting obvious amd doubled down on shaders again. The vast majority of improvements was towards more shading power. At the expense of RT hw. And the reason for that maybe that Unreal 5 and the latest Unity engine have better looking lighting and shadows with their own shading based tech like Lumen for lights and shadows in U5 than raytracing. Without the massive fps hit.

That would leave just RT reflections unless they can improve screen space reflections or other shader based reflections because RT reflections are arguably better... but would one do that one thing for a 50% hit to fps?


https://videocardz.com/newz/alleged-...gram-leaks-out

AMD has thus far confirmed that full Navi 31 GPU features 96MB of L3 Cache (aka Infinity Cache), but the company has not yet confirmed the L0, L1 and L2 caches. Those are mentioned in the new slide:

L0 – 3MB – 240% increase
L1 – 3MB – 300% increase
L2 – 6MB – 50% increase
L3 (Infinity Cache) – 96MB (32MB smaller than Navi 21)

The diagram shows AMD Navi 31 six Shader Engine layout, each with 8 dual Compute Unit. AMD has doubled the compute throughput by adding dual-issue SIMD to their design, but the core count has not been doubled. This is why the Stream Processor count is 6144, but effective single-precision compute performance has increased to 61 TFLOPS.
The slide also mentions that the RDNA3 architecture is to exceed 3 GHz. This might confirm the reports that the architecture is designed to scale up to 3.0 GHz, although the actual reference design is not going above 2.5 GHz boost.

The diagram at the link shows a lot of the changes and its a pretty big jump...

[SSXeon]

Both GA102 and Navi31 are a large increase in caches and FP performance, but Navi will still lag far in raytracing. It should match or lose to the 4080 16GB, but Raster will be closer to the 4090. I do want a 7900XTX, but it will be hard to get one.

AMD and Nvidia still don't care about gamers as they say, when most can't get a card. Bots buying them and scalpers will always win, so i might have to wait after christmas and pray to baby Jesus I can get a new card.

I hope I am wrong.

[bill dennison]

going to wait on a full water cooled card or the 7950 full water cooled

but if i get a 7900 xtx they will announce the 7950 a week after i got it running

[0091/2]

going to wait on a full water cooled card or the 7950 full water cooled

but if i get a 7900 xtx they will announce the 7950 a week after i got it running

Is there a rumor of a higher tier than what they announced?

[pax]

Yea a rumor of a 2x infinity cache model... X3D style.

[bill dennison]

Possible AMD RDNA 3 “Navi 3X” GPU Board IDs For Radeon RX 7000 Graphics Cards Revealed

https://wccftech.com/possible-amd-rd...ards-revealed/

Navi 3X is the 7950 xtx they think

[pax]

Be cool to see a dual GCD card. Maybe they can use the same package by using the smaller navi 32 die.

Thinking the 'designed to run at 3 ghz' thing might have been aimed at that gpu and not navi 31... We see higher clocks on smaller dies on nav2x. Reports coming out that 2.8ghz might be as high as she goes from aibs on navi 31... still a decent uplift but its not 3ghz.

All this time I thought Lumen on Unreal 5 was RT when it was just shader based... Im starting to think with enough increases in raster and leveraging the new 3d engines RT might get displaced for some time as an abandoned feature. If they can do the same image quality and effects for less fps hit and lower transistor budget that is. They certainly can do lighting and shadows... doesnt leave much for RT to do after that.

[Nagorak]

It seems like they're totally set up to have multiple GCDs, which would clearly crush the 4090. Even if it's a super high cost and low production part, it would still be a marketing win, so I'm kind of surprised they didn't come out with that also.

Maybe they still have some bugs to work out with it? Or maybe it needs better memory to not end up being choked by limited memory bandwidth?

[Gandalfthewhite]

yeah ive been thinking they would go for navi 2x32 dies with 6-8 MCDs still a sizeable increase in shader cores and memory bandwidth going to possible 512bit at that point. would get the infinity cache up to 128MB at the same time. the power sweet spot for the 200mm die is going to be solid which should allow for AMD to stay at or below 450w for tbp

the question will be how does it feed work to each die and how does the driver handle everything so that it is seamless and invisible to game engines well having a sizeable uplift.

I think they need to make a dual gcd card NOW so next gen they can have the kinks worked out. Because that is the goal is to have multiple GCDs

[bill dennison]

yeah ive been thinking they would go for navi 2x32 dies with 6-8 MCDs still a sizeable increase in shader cores and memory bandwidth going to possible 512bit at that point. would get the infinity cache up to 128MB at the same time. the power sweet spot for the 200mm die is going to be solid which should allow for AMD to stay at or below 450w for tbp

the question will be how does it feed work to each die and how does the driver handle everything so that it is seamless and invisible to game engines well having a sizeable uplift.

I think they need to make a dual gcd card NOW so next gen they can have the kinks worked out. Because that is the goal is to have multiple GCDs

i think they got the 2 GCD cards ( engineering samples at least ) and it is that they are working on

getting the driver to share the work on 2, 3 or 4 GCD's without the game needing work like crossfire is the hard part

[Trunks0]

There will be MCM with dual GCD cards. As they already did it for the workstation CRNA2 stuff with Radeon Instinct MI250 and MI250x. The only question is if they will do it for a gamer GPU.

[pax]

We might see much higher clocked parts later as seems the first batch of N31 chips had a hw bug that needed a respin that got in too late:

https://twitter.com/XpeaGPU?ref_src=...3037%2Fpage-20

AGF
@XpeaGPU
Replying to
@WreckItROFL
Disclaimer, I'm not AMD leaker, so it's not direct source. This is second hand info so take it with grain of salt.
AIBs were told "easily above 3.2Ghz" with normal OC tweaks (power/voltage) and max frequency in beta BIOS is 3.7GHz
1:05 AM · Nov 6, 2022
·Twitter Web App

AGF
@XpeaGPU
Bad news:
Navi31 needs a respin to fix the issue preventing to reach the 3GHz target at reasonable power level. Hardware team failed this design.
Good news:
The problem has been identified and corrected for Navi32
11:00 PM · Nov 5, 2022
·Twitter Web App

Looks like if N32 hits 3.2 ghz as some think it can itll be ~85% the perf of the N31 at 2.5 ghz... but way lower cost. Now mind you they might throw a ton of power at NM31 to hit 2.8-2.9... so who knows for sure until we see some reviews.

[bill dennison]

We might see much higher clocked parts later as seems the first batch of N31 chips had a hw bug that needed a respin that got in too late:

https://twitter.com/XpeaGPU?ref_src=...3037%2Fpage-20







Looks like if N32 hits 3.2 ghz as some think it can itll be ~85% the perf of the N31 at 2.5 ghz... but way lower cost. Now mind you they might throw a ton of power at NM31 to hit 2.8-2.9... so who knows for sure until we see some reviews.

maybe still less power than the 4090 and i'm sure a AIB will do it
hopefully a full water block card if they do don't want a card the size of a cinder block

[pax]

yea lol they are reusing the same air coolers from the 4090's... seems the future is brick

[SSXeon]

yea lol they are reusing the same air coolers from the 4090's... seems the future is brick

It is pretty sad, I really liked the EVGA next gen card prototype that Steve and Jay showed. I don't like the future size, and I am trying to get used to this new site.

[Nagorak]

Well, at least there is the more svelte reference model. The 6000 series reference cooler was actually pretty good (at least the one on the 6800 XT).

[Napoleonic]

I don't want to start a new thread just for this but what's the difference between AMD CAS and FSR? because I'm trying both of them plus DLSS in Forza Horizon 5 and find that CAS looks better than FSR and also give frame rate bonus, is that supposed to happen? I mean why did AMD create FSR if they already had CAS and works just as well?

speaking of which, FH5 is the first game where CAS + TAA outperform FSR and DLSS for me...

[Kombatant]

I don't want to start a new thread just for this but what's the difference between AMD CAS and FSR? because I'm trying both of them plus DLSS in Forza Horizon 5 and find that CAS looks better than FSR and also give frame rate bonus, is that supposed to happen? I mean why did AMD create FSR if they already had CAS and works just as well?

speaking of which, FH5 is the first game where CAS + TAA outperform FSR and DLSS for me...

This might help https://gpuopen.com/learn/using-fidelityfx-cli/ - I always considered CAS+Upscaling to be FSR 0.5 essentially.

[pax]

https://www.rra.go.kr/ko/license/A_b...17210000326082

Lots of skus here:

ATI TECHNOLOGIES ULC

VIDEO CARD

D74510

D74511 D74512 D74513 D74514 D74515 D74516

R-R-ATI-102-D74510

ATI TECHNOLOGIES ULC

대만

2022-11-09

Wonder if we might see these sooner than later.

[pax]

https://videocardz.com/newz/powercol...llhound-series

Whats interesting is that the Hellhound range in navi2 is the 6700 series. Or the midrange. If the Powercolor Hellhound is in the 7900 series or one of the 7900 skus and is meant as the mid range or upper mid range, it may mean theres a higher sku in the offing at some point than has been admitted to. Kind of gels the rumor that AMD saying the 7900 was meant to go against the 4080 and not the 4090 range. Powercolor's higher end range is the Red Devil series and while it overlapped the 6700 Hellhound range with a Red Devil branded 6700, 6900's were Red Devil only.

[KAC]

When are reviews out?

[acroig]

When are reviews out?

Day before release? 12/12

[pax]

https://wccftech.com/powercolor-rade...in-connectors/

Quad slot giant cooler 7900XTX incoming. Will be interesting to see the uplift vs the small reference model.

[bill dennison]

hope they make a 7900 xtx red devil liquid like last time


i don't need a 50 pound 4 slot card like the AIB 4090's
they must be tossing the 54% power efficiency

[pax]

hope they make a 7900 xtx red devil liquid like last time


i don't need a 50 pound 4 slot card like the AIB 4090's
they must be tossing the 54% power efficiency

Honestly wouldnt an aio version be cheaper to make than those giant air coolers?

[bill dennison]

Honestly wouldnt an aio version be cheaper to make than those giant air coolers?

maybe i think it's a little keeping up with the joneses all the others are going to use 4090 coolers so powercool is going big to
and unless they can realy crank the clocks a lot i don't think the 7900 xtx will need that much cooling
but if they can crank it up it will be a good bit faster

AIO doesn't work for me my case is too big and the AIO hoses too short
i would have to mount the rad fully inside the case like i did with one of my fury x's

and i have full water so just a little repiping and a full waterblock shoudd be cheaper than AIO but never is

[bill dennison]

[pax]

N31 at 3.2 ghz would be amazing if the front end can also clock to the same rate... Lets see what they get with AIBs version of the current version of the chip tho. If its really stuck at 2.4-2.5ghz with the clock bug fixed N32 running to 3.2+ might wait for that one as itll be a smidge under N31 just due to clocks and a few hundred bucks cheaper.

[bill dennison]

We can expect custom Radeon RX 7900 series GPUs on launch on December 13

Read more: https://www.tweaktown.com/news/89666.../index.html

https://www.tweaktown.com/news/89666...-13/index.html

[pax]

https://www.notebookcheck.net/Revise....671239.0.html

3.6 ghz? I can see that on an oced well cooled Navi 32 and the Navi 31 GCD (its not that big at 300 mm2) if the decoupled clock is bigger than the 200 mhz we see now. Perf might be muted in that case. Hard to see the whole Navi 31 gpu doing 3.6 ghz but if true itd be amazing.

Looks like the article is mostly based on the RGT video. Dont remember hearing 3.6 ghz tho...

[pax]

https://videocardz.com/newz/asus-con...g-clock-speeds

ASUS TUF Gaming Radeon RX 7900 XTX OC Edition

• OC mode:
  • up to 2615 MHz (Boost Clock) +4.6%
  • up to 2455 MHz (Game Clock) +6.7%

Quite a bit les than I thought a big cooler like that could get at. Maybe they are keeping better chips for the strix...

[Nascar24]

It appears the respin rumor maybe true. Clocks appear to be gimped because of it.

[pax]

Yea 400 mhz to 3 ghz is worth a respin... looks like its about a year away tho.

[Nascar24]

Should make it a decent refresh.

[Nagorak]

It seems like AMD has really struggled with clocks over they years. Vega also appears to not have clocked as high as they expected, which is why it was so much slower than 1080 Ti despite being a similarly sized die.

It's a bit of a shame in this case though since the difference in clock speed likely makes the difference of the 7900 XTX being decisively faster than the 4090 in raster, and instead being a bit behind.

[Nascar24]

The issue is with the chip, which they said they discovered too late, which will be fixed in the refresh whenever that comes.

[bill dennison]

Revised Navi 31-based RX 7990 XTX, RX 7950 XTX, and RX 7950 XT with up to 3.6 GHz clock speeds and 3D V-Cache could be in the works

https://www.notebookcheck.net/Revise....671239.0.html


they will fix and replace the top 3 cards in 6 months or a little less

https://www.techpowerup.com/gpu-spec...7990-xtx.c3973

[Nascar24]

https://www.notebookcheck.net/Revise....671239.0.html


they will fix and replace the top 3 cards in 6 months or a little less

https://www.techpowerup.com/gpu-spec...7990-xtx.c3973

Billy, it will happen this time next year, somewhere between Oct and Dec. Nothing happens fast on these cards.

[pax]

image.png


Edit: Ah shoulda left at least one JimmyRomaguera in just for laughs...