RDNA4

[no-X]

That's exactly what it means doesn't it? Kepler pushed that 40% faster pretty much as a fact right until the very end and it's obvious he got it wrong.

Core for core is not IPC. According to the recent leak, Epyc Turin seems to be even ~60 %* faster core for core than Epyc Genoa in CPU-Z benchmark, because of higher effective clocks.

*and ~30 % faster core for core than Zen 4 Threadripper PRO.

https://www.tomshardware.com/pc-com...the-multi-threaded-performance-champ-in-cpu-z

 

[Erinyes]

It means the performance per clock, usually measured on single core but not necessarily.
So chip/core A at X GHz vs chip/core Y at X GHz, the actual clock is irrelevant, it just needs to be the same for both)

But perf/clock for single core at the same clock is IPC isn't it?

Core for core is not IPC. According to the recent leak, Epyc Turin seems to be even ~60 %* faster core for core than Epyc Genoa in CPU-Z benchmark, because of higher effective clocks.

*and ~30 % faster core for core than Zen 4 Threadripper PRO.

https://www.tomshardware.com/pc-com...the-multi-threaded-performance-champ-in-cpu-z

Keep in mind that's comparing a 500W 128C Zen5 Turin vs ~350W 128C Zen 4c Bergamo and 96C Zen 4 Threadripper so basis TDP alone it should be ~43% faster (discounting the fact that IOD power is relatively static and the additional TDP is entirely used by the cores). So it's not exactly an apples to apples comparison. As we've seen with the Ryzen parts, the clocks are expected to be similar or maybe slightly higher as the process is pretty much the same and the number of cores are increasing.

I had read the referred Anandtech thread and Kepler, along with a few others were heavily pushing the 40% faster in SPEC INT 1t hype. Anyway this is an RDNA 4 thread so I would not like to derail the thread even more with CPU talk, but I would simply take anything that Kepler says (or any other internet poster/leaker for that matter) with a large grain of salt.

On topic, if the RDNA 3.5 Perf/W and/or clock improvements are anything to go by, RDNA4 should offer measurably better performance at the same TDPs which should be quite competitive for midrange parts (If one considers ~$500 as midrange these days).

 

[no-X]

Keep in mind that's comparing a 500W 128C Zen5 Turin vs ~350W 128C Zen 4c Bergamo…

That's why my previous post doesn't contain even a single mention of Bergamo. I compared Zen 4 (not Zen 4c) to Zen 5. 128 cores at 500W TDP compared to 96 cores at 350W TDP result in close per-core power (3.9 vs. 3.65 W = just 6.8% difference).

 

[Frenetic Pony]

https://twitter.com/x/status/1821117215703216290

 

[Erinyes]

https://twitter.com/x/status/1821117215703216290

Probably just a repeat of Zen 5 parts and current gen graphics cards. They might tease Zen 5 X3D I guess? Seems a bit early for RDNA 4.

 

[eastmen]

Probably just a repeat of Zen 5 parts and current gen graphics cards. They might tease Ryzen 5 X3D I guess? Seems a bit early for RDNA 4.

well with rumors of blackwell maybe being pushed back until 2026 for consumers it could be a really good time for amd to launch something new and make some waves.

 

[DegustatoR]

well with rumors of blackwell maybe being pushed back until 2026 for consumers

Why not 2029 while we're at it? It's still 4Q24 for now.

 

[Erinyes]

well with rumors of blackwell maybe being pushed back until 2026 for consumers it could be a really good time for amd to launch something new and make some waves.

I assume that's a typo and you meant 2025? That's the only rumour so far, that it might be delayed to CES'25 but even if it happens that's not much of a delay. I'd still expect both Nvidia and AMD to make Q4’24 for their next gen, at least one SKU if not more.

 

[eastmen]

I assume that's a typo and you meant 2025? That's the only rumour so far, that it might be delayed to CES'25 but even if it happens that's not much of a delay. I'd still expect both Nvidia and AMD to make Q4’24 for their next gen, at least one SKU if not more.

no 26' jeff gurtsman mentioned it in his latest pod cast. So unless he misspoke

 

[LordEC911]

Probably just a repeat of Zen 5 parts and current gen graphics cards. They might tease Zen 5 X3D I guess? Seems a bit early for RDNA 4.

New rumor is RDNA4 went into mass production last month. Release dates are supposedly Navi48 in Oct and Navi44 in Nov.
If true, then they will definitely show/tease something at GamesCom.

 

[Erinyes]

New rumor is RDNA4 went into mass production last month. Release dates are supposedly Navi48 in Oct and Navi44 in Nov.
If true, then they will definitely show/tease something at GamesCom.

You're referring to the MLID video? That only mentioned Navi 48 but Navi 44 shouldn't be far behind. Even he mentioned we might just get a teaser tomorrow at Gamescom. Let's see.

 

[LordEC911]

You're referring to the MLID video? That only mentioned Navi 48 but Navi 44 shouldn't be far behind. Even he mentioned we might just get a teaser tomorrow at Gamescom. Let's see.

Not sure, it could be the same rumor or just based off of MLID's video.
Ugh, makes me feel dirty thinking I might have been repeating info from MLID.

 

[Erinyes]

Not sure, it could be the same rumor or just based off of MLID's video.
Ugh, makes me feel dirty thinking I might have been repeating info from MLID.

Seems likely. Link -

Pretty much a repeat of older info, i.e. 64CU, 256 bit GDDR6, ~2.9-3.2 Ghz clock. Expected to be around or slightly faster than 7900XT in raster and ~4070 Ti Super in RT. MSRP of $499-599, volume release in Oct-Nov.

 

[Frenetic Pony]

Early geekbench testing. Looks like the lower bin version of the big one has 28WGP/56CUs

https://twitter.com/x/status/1827937117328007293

 

[Qesa]

Early geekbench testing. Looks like the lower bin version of the big one has 28WGP/56CUs

https://twitter.com/x/status/1827937117328007293

It's similar to 7800 xt in most of the subtests.

Face detection gives 0 score for some reason despite images/s actually being slightly improved? That is odd and also seems to hurt the overall score to a huge degree.

The other outlier is particle physics only being half as performant on the RDNA4 chip. Geekbench give very few details about that benchmark unfortunately so I'm gonna leave interpretation of that one to the floor.

 

[Frenetic Pony]

It's similar to 7800 xt in most of the subtests.

Face detection gives 0 score for some reason despite images/s actually being slightly improved? That is odd and also seems to hurt the overall score to a huge degree.

The other outlier is particle physics only being half as performant on the RDNA4 chip. Geekbench give very few details about that benchmark unfortunately so I'm gonna leave interpretation of that one to the floor.

These early benchmark tests are never a good indication of performance. Stuff isn't clocked to final spec, other tests don't work at all (such as face detection), it's mostly interesting to note because it's a confirmation that the thing exists and maybe isn't super far off.

 

[Frenetic Pony]

https://twitter.com/x/status/1828223002586050648

 

[Erinyes]

https://twitter.com/x/status/1828223002586050648

So 3 versions for the top Navi 48, just like 6xxx series and 7xxx series.

64 CU 256 bit 20 Gbps 16 GB
56 CU 256 bit 18 Gbps 16 GB?
48 CU 192 bit 19 Gbps 12 GB?

But given its a much smaller GPU, yields should be fairly good. Would such a heavily cut down part make sense?

 

[Newguy]

Depends how much of a product stack they'd want. Maybe 8800XT, 8700XT, 8700 for N48 then 86XT to 85XT for N44? The perf/price/value gap may be far too great without it and if they can get more money for cheap parts why not from their perspective

 

[LordEC911]

So 3 versions for the top Navi 48, just like 6xxx series and 7xxx series.

64 CU 256 bit 20 Gbps 16 GB
56 CU 256 bit 18 Gbps 16 GB?
48 CU 192 bit 19 Gbps 12 GB?

But given its a much smaller GPU, yields should be fairly good. Would such a heavily cut down part make sense?

Yes, you are always going to have asome percentage of dies that need to be salvaged, if yields are that good then you want a big cut to salvage the most into a single product.
They really need that 192bit N48 to fall inbetween the two N32, which seems likely if they are giving it 19Gbps GDDR6.

The main issue is how high they can push N44 and not be bandwidth limited? A 20% bump over N33 seems doable but how much farther...
If they can push N44 to ~3.7ghz that could net them ~1.35x of N33 and be within ~10% of 7700XT, but that might need 20Gbps DDR6.
The other issue is, could they do a clam shell mode with both 1GB and 2GB GDDR6 ICs and give N44 12GB of VRAM.

I have been tweaking this lineup for quite awhile but this seems to make the most sense to me, though maybe a bit optimistic in some places.
Prices are obviously up in the air, Just depends on final design/performance and a bit on the branding/model name.

N48 16GB 64CU/128ROPs @ 3.4ghz 250w ~7900XT (+50%7700XT,+30%7800XT)
N48bin 16GB 56CU/128ROPs @ 3.2ghz 220w ~7900GRE (+30%7700XT)
N48cut 12GB 48CU/96ROPs @ 3.0ghz 180w ~RTX4070 (+60%7600XT,+10%7700XT)
N44 12GB 32CU/64ROPs @ 3.7ghz 160w ~3070Ti (+35%7600XT, +10%4060Ti)
N44bin 12GB 32CU/64ROPs @ 3.2ghz 140w ~6700XT (+15%7600XT)
N44cut 8GB 28CU/64ROPs @ 2.8ghz 100w ~7600XT/RTX4060 (OEM?)

Edit- Removed my naming and pricing estimates since they were likely way off.