AMD RDNA3 Specifications Discussion Thread

yes , they did and it din´t help Vega to become better gaming GPU, because despite higher clock, the limitations due architectural design choices still remained :) which actually proves my point, that RDNA3 clock is not limited by process node .-) RDNA2 clock up to 2,8GHz on 7nm , I don´t see any valid point from you, why it´should be worse on 5nm node beside some architectural glitches or design flaw....

Hmmmm, since we only have the official information released by AMD thus far, let's compare shall we? Clocks are Base, Game, Boost in MHz using AMDs officially released numbers for each (taken off of Techpowerup's GPU database)...
  • 7900 XTX
    • 1900, 2300, 2505
  • 7900 XT
    • 1500, 2000, 2400
  • 6950 XT
    • 1860, 2100, 2310
  • 6900 XT
    • 1825, 2015, 2250
  • 6800 XT
    • 1825, 2015, 2250
  • 6800
    • 1700, 1815, 2105
  • 6750 XT
    • 2150, 2495, 2600
  • 6700 XT
    • 2321, 2424, 2581
  • 6650 XT
    • 2055, 2410, 2635
  • 6600 XT
    • 1968, 2359, 2589
  • 6500 XT
    • 2310, 2610, 2815
  • 6400
    • 1923, 2039, 2321
So, yeah, we all want the 7900 XTX to be like the 6500 XT because that was such a wonderful card. :p

But if we compare it to the cards it's actually replacing, the 6900 XT and/or 6950 XT, then the base clock is slightly higher but game clock and boost clock are significantly higher.

The 7900 XT certainly is a bit disappointing at first glance (especially considering MSRP is only 100 USD less), but while the base clock is pretty low, the game clock is relatively equal to the 6900 XT while the boost clock is 150 MHz higher.

I'll wait for independent benchmarks before I even start to think about whether this is disappointing or not. :p

Regards,
SB
 
What do we know about wave size for compute shaders? Some people assume it's 32, others 64.
Afaik the compiler decides arbitraily in a black box, but not sure to which shader stages this applies.
64 for VS/GS/PS as indicated by Mesa RADV
Controllable via VK_EXT_subgroup_size_control apparently.
 
64 for VS/GS/PS as indicated by Mesa RADV
Controllable via VK_EXT_subgroup_size_control apparently.

Wonder if this is what's killing the performance per ALU. 2.2x ALU for 1.6x the performance. Maybe the dual issue just doesn't work for a lot of instructions, the waves don't get filled, and vaunted "density improvement" from this design amounts to little. Could even be killing any RT improvement, 32 wave size is already large there. Depending on the ALU involvement still you'd get killed by low occupancy on heavy divergence.

Anyway, at least the new infinity cache looks good. More selectivity and bandwidth seems to have paid off, less cache for more improvement. And the chiplet arch as well, combined that's saved them like $80 on BOM, around 28% of the cost.

Edit "7900 designed for clockspeeds up to 3ghz" supposedly straight from AMD, err. 24gbps, 450 watts, 3ghz card next year?
 
Last edited:
No.
The reality is a lot funnier than that.

Letsa call it 7970XTX 3GHz edition and throw nostalgiabuxx at the screen.
OMG, are you just suggesting another soon to be realeased part with more clock/more WGPs? Or some kind of 295X2 style multicore solution?
 
So AMD confirmed the 7900XTX is competitive with the 4080, not the 4090.

I started looking closely at their claims, and this is how they derived their 54% perf per watt claim, they set both a 6900XT and 7900XTX to a 300W TBP, and tested both in select titles.

RX-816 – Based on AMD internal analysis November 2022, on a system configured with a Radeon RX 7900 XTX GPU, driver 31.0.14000.24040, AMD Ryzen 9 5900X CPU, 32 GBDDR4-7200MHz, ROG CROSSHAIR VIII HERO (WI-FI) motherboard, set to 300W TBP, on Win10 Pro, versus a similarly configured test system with a 300W Radeon 6900 XT GPU and driver 31.0.12019.16007, measuring FPS performance in select titles. Performance per watt calculated using the total board power (TBP) of the AMD GPUs listed herein. System manufacturers may vary configurations, yielding different results.

So this means, that in those titles and at a lower TBP of 300W (instead of 355W), the 7900XTX was 54% faster than than the 6900XT (@300W). That already puts it at the 4080's performance level (which is 35%~40% over the 3090Ti). So the 7900XTX will beat the 4080 by however much it gains going from 300W to 355W.

It also means that 54% claim also includes some low hanging ray tracing gains as well.
 
OMG, are you just suggesting another soon to be realeased part with more clock/more WGPs? Or some kind of 295X2 style multicore solution?
The annotated GCD die shot (provided by AMD tho) doesn't show any large interface for GCD-GCD communication.

We need a real shot to check whether TSVs are present on the MCDs. If so we could still expect a 3D stacked premium' version at some point.
 
For those of us with more limited $ & W budgets is there any indication when we will see the mid-range stuff?

I pretty much only play stuff that would class as legacy these days & my 5600XT copes with most of that fine at 4K but I could do with a bit more horsepower without having to spend $$$ & I really don't want a space-heater so I've been hoping mid-range + the crash of the crypto GPU bubble should mean I can look forward to a decent performance upgrade at affordable $ & similar W.
 
I find the lack of ray tracing specific performance boosts weird considering AMD talks about a 50% increase in performance per CU and the 7900 XTX has 20% more CUs. So at least in some titles we should be seeing an 80% increase in RT performance.
 
OMG, are you just suggesting another soon to be realeased part with more clock/more WGPs? Or some kind of 295X2 style multicore solution?
Consider that at best he's the amd version of eastman, at worst just an attention seeker making things on the spot.
 
I find the lack of ray tracing specific performance boosts weird considering AMD talks about a 50% increase in performance per CU and the 7900 XTX has 20% more CUs. So at least in some titles we should be seeing an 80% increase in RT performance.
In some titles that are completely RT limited, probably (MInecraft RT?). In the others the pipeline has likely other bottlenecking factors, and you will see the improvement leaning more towards the raster factor increase.
 
Last edited:
AMD-RDNA3-NAVI31-BLOCK-DIAGRAM.jpg



architecture "to exceed 3GHz" confirmed
 
In some titles that are completely RT limited, probably (MInecraft RT?). In the others the pipeline has likely other bottlenecking factors, and you will see the improvement leanong more towards the raster factor increase.
I'd expect the opposite - bigger gains in "lite RT" titles than in heavy path traced ones.
 
I'm not sure what that means - could you give an example of a game where this could possibly be the case?

Like the 7900 is not a non-RT GPU, it's just a GPU who's RT performance will likely only match a 2-year old Ampere 3080. That's definitely disappointing, but it's far, far above the RT performance of the S, X and PS5. The main advantage of the PC is choice, you're not restricted into what the developer personally felt were the right compromises for a particular framerate. There is not going to be something released within this console generation that will only run well on Ada, as that would be financial suicide for the developer.

The argument isn't that "RT is worthless and has no future" - the argument is that the sacrifices it brings to performance and resolution are too great for the small improvements in brings in many of the titles currently - at least at the framerate standards that have been raised considerably in the past 5 years. You can mitigate those sure - but the actual cost in $ is far too high for the majority of PC gamers.

If the 4070-whatever comes out in 2023, and is in the same price ballpark of the 7900, has a 20% raster disadvantage but a 100% RT advantage, then you can definitely make the argument that it's a shortsighted decision to go with Radeon - like it would have been when Ampere was introduced and the 6800/6900 series were almost priced identically, which explains why Nvidia completely took over in marketshare (that and in many games, you could argue it was also much faster in raster due to DLSS). We'll see with actual benchmarks I guess, I don't think the 4080 will necessarily put a cork in this argument though.



You get that with a 7900 series card though, in spades. And you can get that without RT as well - ports like God of War, Days Gone etc have significant raster improvements even outside of framerate/resolution. Again I don't think anyone is arguing that there is no point to ray tracing at all, just that the cost to get a product that can run full-RT games at the resolutions/framerates many gamers want is too high atm, so the sacrifice to that area of performance may be more acceptable if the other 2 can be met at a more reasonable price point.

Yeah I was a probably a bit too literal in my argument. I do of course understand a 7900XT/X can easily achieve better graphics, frame rate and image quality than an Series S or Series X all at the same time. As can a 6900 and probably even a 6800 to a marginal degree in each area. However...

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.

The 7900 (and 6xxx series for that matter) is a beast for rasterization and great value when compared to the $1100 4080 in that respect. But when RT comes into play - as it does in many big titles and will in many more moving forwards - then it's potentially only competing with a 3080Ti which can be picked up for much less or it's giving you an experience that falls short of the consoles in some respects.
 
Still wondering what's the difference between 2nd gen RT and 1st gen.
Mentioning larger caches is off topic, imo.
'RT featuers for perf. & eff.' does not mention what's the features.
1.8x @ x GHz sounds like comparing per CU vs. the old gen at same frequency, which would be the speedup i would have expected in the best case. But benches suggest it's comparing the whole GPUs.

Well, ISA docs will tell...
 
That might be ok if people are aware of it in relation to PSUs and before they specified this in detail in 2018. The R9 is from 2014 so during the wild west period.
True, though it makes you wonder about corsairs 600W 12vhpwr connector as it only has 2 8pins.

Not saying AMD should be trying to pull 500W from 2 8pins just pointing out that the spec actually has room for that much amperage to be pulled if one was willing to.


I am also curious about N31 clocks compared to N21. Are there any N21 cards that only match the reference clocks? My reference 6900xt clocks higher even at the OOB 250W power limit.
 
Back
Top