AMD RDNA3 Specifications Discussion Thread

[DegustatoR]

What do you mean 'spent so much time'?

In marketing materials, during unveil, etc.
It would make sense if all Nv had would be DP 1.4 but they also have HDMI 2.1 which kinda negates this in full.

 

[Bondrewd]

but they also have HDMI 2.1 which kinda negates this in

HDMI2.1 monitors are of mucho noexisto kind so...

 

[troyan]

HDMI2.1 monitors are of mucho noexisto kind so...

Every new monitor supports HDMI2.1. Sure, dont buy one with the latest G-Sync module...

 

[DegustatoR]

HDMI2.1 monitors are of mucho noexisto kind so...

And DP 2.1 monitors are a plenty apparently.

 

[Bondrewd]

And DP 2.1 monitors are a plenty apparently.

That's not an excuse to ship outdated display cores.
Still not as bad as Intel having no HDMI 2.1 support on their mobile SoCs because *reasons*.

 

[DegustatoR]

That's not an excuse to ship outdated display cores.

With zero DP 2.1 products on the market there's nothing "outdated" in DP 1.4. It also provides enough b/w to hit 4K/240Hz - which is enough for the next several years still as we currently have only one product capable of such refresh rates.
Those who will want more can use HDMI 2.1 FRL6 to go up to 8K/120Hz (whenever this will even appear). It really is a non issue.

 

[PSman1700]

On PC you need a higher framerate because mouse causes much faster camera motion. 30fps feels often unplayable, while on console it's more acceptable.

Thats generally true, however it depends on the user, what your used to and what kind of display and control input you use. Those looking to play 30fps single player games are more often using a controller and/or TV than say someone aiming to play CSGO or fortnite at well above 144fps. PC>TV usage has drastically increased since last generation started, so has the usage of controllers. Though kb/m and a monitor are still the most common on the platform, most have the option to connect to a TV.

Also, with similar HW, you'll get worse performance because PC lacks some optimization options (both SW and HW) due to being an open platform spanning variable HW requiring compromised APIs, reducing options and value of low level optimizations.

Thus you need a bit more powerful HW in general on PC to have a good experience with the same games, even if you're not an enthusiast aiming for high gfx settings.

Im sure theres more overhead on the pc, but im also sure its not a huge deflict as has been as seen in previous generations. The mentioned 6600XT is actually somewhat more capable, and seeing comparisons on YT, its atleast a match, more often than not outperforming the PS5. Perhaps due to infinity cache, quite abit higher clocks etc which help out in raster and rt?
Anyway, a 6600XT should suffice for most pc gamers looking to get ballpark premium console experiences. Some settings might have to be lowered perhaps, some allow for higher settings.

Were not talking you need 6800, 3080 or 4080 etc products which arent really for the mainstream i think. I'd say a 3060Ti is a very good mainstream GPU with its current price, you get above baseline in raster and way more capable RT, aswell as dlss. A770 if you arent so much into older games is just as good a alternative aswell.

I REALLY like the design of that card. Looks sexy

I like the design aswell, i hope we see more of that also from NV.

 

[LordEC911]

*snip
For what it's worth Semianalysis's BOM estimates - https://www.semianalysis.com/p/ada-lovelace-gpus-shows-how-desperate
*snip*

Are these the same guys that were suggesting AMD was paying ~$300 for HBM back in the day?
They have some really wacky estimates...
A GPU twice the size of Navi31, on a more expensive custom node (that came with two price adjustments), that they fit <100 die per wafer.... is only 1.5x Navi31?
Nvidia would be lucky if AD102 is less than 3x Navi31.

 

[Bondrewd]

Are these the same guys that were suggesting AMD was paying ~$300 for HBM back in the day?

I don't think that was dolan but yes he's one helluva grifter.

 

[Digidi]

What ist hte New Flexible CP & Geometry Pipe? What does CP means?`

Source: https://videocardz.com/newz/alleged-amd-navi-31-gpu-block-diagram-leaks-out

 

[Kaotik]

What ist hte New Flexible CP & Geometry Pipe? What does CP means?`

Source: https://videocardz.com/newz/alleged-amd-navi-31-gpu-block-diagram-leaks-out

Command Processor?

 

[Remij]

Command Processor?

Yeah, it's right there dead center in the middle of the chip.

 

[Henry swagger]

https://twitter.com/x/status/1588359856926633991

Rdna 3 cu

 

[JoeJ]

https://twitter.com/x/status/1588359856926633991

Rdna 3 cu

If this is accurate, it means LDS has been practically doubled?
So besides 'no more register pressure' due to +50% VGPR, no more occupany drop due to high LDS usage either?
This is very interesting to me, as i could use the extra LDS for much higher quality GI.
So to me, this seems the biggest next gen change since the PS4 days. :O
I always kinda hoped this to happen, but we'll see...

Edit: It also would explain why the uplift from HW traversal is less than expected, becasue RT cores are cut in half.

 

[OlegSH]

Are these the same guys that were suggesting AMD was paying ~$300 for HBM back in the day?

Well, at least they accounted for MCDs and packaging costs, so still looks way more believable in comparison with something like that "Nvidia would be lucky if AD102 is less than 3x Navi 31."

 

[LordEC911]

Well, at least they accounted for MCDs and packaging costs, so still looks way more believable in comparison with something like that "Nvidia would be lucky if AD102 is less than 3x Navi 31."

Packaging costs are a small percentage of a product that has a "relatively" high BOM.
Even if AMD's packaging costs are 3-4x a monolithic GPU, it doesn't change the math very much.

Just look at N33 vs AD106... roughly the same size but one is on N6 and the other is on a custom 4nm.
AD106 is only 1.5x when they are paying at least 2x more a wafer? How?

Edit- Forgot to mention that GDDR6x still has ~20-30% price premium over GDDR6.

 

[OlegSH]

AD106 is only 1.5x when they are paying at least 2x more a wafer? How?

Where did you get that? Any proofs?

Just look at N33 vs AD106... roughly the same size but one is on N6 and the other is on a custom 4nm.

Again, not confirmed info on N33, so still highly inaccurate and speculative stuff

Even if AMD's packaging costs are 3-4x a monolithic GPU, it doesn't change the math very much.

5$ turned into 20$ would change a lot

Forgot to mention that GDDR6x still has ~20-30% price premium over GDDR6.

Are there any real info on prices rather than just rumors?

 

[pjbliverpool]

I guess I just struggle to relate to this hypothetical person that would always use the RT mode on console games (which in many titles, can mean you're locked at 30fps or an unlocked framerate below 60), yet will also never use it on a card that will give them ~3X the performance with the same settings.

Like sure, I haven't taken an extensive survey, and perhaps it's a very biased example as it's based on people that use Internet forums. But when the topic of RT comes up, far more often than not, the biggest pushback against its value that I see comes from console gamers. I expect more often than not, when given the choice, it's just not enabled due to the resolution/performance hit.

So in reality, I don't think this individual who thinks in this binary between "RT is worthless/RT means the best graphics" is really that common. I think most make their judgment on a title by title basis, for some the more physically accurate lighting/shadows is worth it for the performance/resolution hit. For others, the scenes where these enhancements are actually noticeable are too rare to justify the drawbacks to the rest of the presentation that is ever-present. The person that would disable RT on their 7900 is sure as hell not going to enable it on their console tiles if given the choice, so it's not 'they're getting worse graphics' - they're getting far better graphics and performance vs. the mode they would also use on the consoles.

Don't forget too, that where the Ampere/ADA architecture really shines with RT is when the higher precision settings are used. When the lower settings that consoles employ are also used on RT titles, RDNA naturally doesn't suffer quite as catastrophically by comparison. Whether those lowered settings make RT pointless though of course, will be a matter of debate - but if they do, then the "but consoles will use RT" argument makes even less sense.

I guess the contention here is always going to be what 'into play' means, as it's such a variable impact depending on the title and the particular precision level used. The main argument is that in the future, this impact will be far more significant, as much more of the rendering pipeline will be enhanced, or fully based on RT. Ok then, but you can't expect people to fully embrace a hypothetical future - they're going to be buy the hardware they feel can enhance the games they're playing now.

I mean the extra outlay you spend on a PC over a console is justified for many reasons, but one of those is certainly that all of your existing library is enhanced immediately rather than hopes and prayers for it to be fully taken advantage of in an indeterminate future.

I don't really disagree with anything you've said there. There's not really a right or wrong to this discussion as we're just representing 2 different, but equally valid points of view. Essentially it boils down to my earlier statement:

The main thrust of my argument was that if we are to say RT performance doesn't matter because we'll just turn it off anyway, then you are accepting that on your $900 GPU, you are getting a lesser experience in some respects than even an Xbox Series S gamer. On the other hand if you turn RT on, then the AMD GPU is much slower than it's competing Nvidia GPU. Both options are bad IMO.

You're simply saying that the first option above is not a bad thing in your opinion because (perhaps more rationally than my argument) you don't care that the Series S might have better core graphics because you value the PC trade off of much higher frame rates and resolution more. For my part the minimum baseline would always be what I could get from a much cheaper piece of hardware, and then the extra value from the more expensive/powerful PC comes from building on that. That almost always translates as turning all settings to max (with the potential occasional exception of some invisible Ultra settings), and then adjusting my resolution accordingly to hit a suitable frame rate for that particular genre - which in many cases I define as solid and well frame paced 30fps or slightly above on my 1070 but would be looking to up that to a minimum of 60 or thereabouts on a modern GPU.

And on the other-other-hand, for the equivalent price bracket, the Nvidia GPU could end up being much slower than the 7900 in rasterization, which is still important for a massive number of games, especially when you value a high frame rate experience.

While this is also true, I think it's fair to consider which aspect of performance matters more at this overall price/performance segment. Are you really going to encounter many situations where a 4080 isn't fast enough in pure raster performance at your desired resolution target? And how numerous are those likely to be compared to the number of games that a 7900XTX with 3080-3090 level RT performance is unable to run fast enough at that same resolution target with RT enabled?

 

[Newguy]

If N32 hits those much higher clocks (3GHz+) while staying reasonably efficient does that put it uncomfortably close to the reference 7900XT, assuming no bandwidth/other perf constraints? Say 7900XT hits 2.5GHz in game average because claimed clocks underrate it, N32 hits 3-3.25GHz. 60 CUs @3-3.25GHz vs 84 @2.5GHz, that's 85.7-92.9% compute perf (bandwidth is 80% VRAM, 66.7% cache), compared to 6700XT vs 6800 it's roughly 40 CUs @2.6GHz vs 60 CUs @2.2GHz, 78.8% compute, bandwidth is 75% for VRAM and cache with actual performance about 80-85% (1440p not 4k)

Will AMD allow that? If you have AIBs pushing it and getting 90-95% of the 7900XT's perf with the greatly increased clocks at what I assume will be a much lower price it becomes a very pointless card

 

[Bondrewd]

Will AMD allow that?

Yea.

it becomes a very pointless card

Kinda the point, really.