TESKATLIPOKA
Newcomer
Meaning Ada102 will be faster, right?
- Status
Leoneazzurro5
Regular
That or it would come just short of 90 TF, meaning a 3,5-3,6GHz peak clock.
Bondrewd
Veteran
Vewy smart.
Gotta keep that power in check.
TESKATLIPOKA
Newcomer
Even 3.5-3.6GHz would be >1.5x higher clock speed than N21, but then N33 or N32 should be the one with clocks close to 4GHz.
hurleybird
Newcomer
N33 is on 6nm while N32 and N31 are on 5nm. Looking at Zen 4, 5nm probably clocks better, or at the very least, AMD has no issues attaining very high frequencies on 5nm. Monolithic vs. chiplets throws another wrench into the comparison, but chiplets should improve binning a bit. Smart money is on N32 being able to reach the highest clocks.
For RDNA2, N22 is 1/2 of N21 and boosts about 15% higher for the x00 models or 12% for the x50 refreshes. In comparison N32 is 2/3rds of N31 in terms of memory bus and 5/8ths in terms of compute, so maybe halve those figures if the same trend is appropriate (eg. you could expect N32 to clock 6-8% higher than N31).
That's still pretty simplistic though. The relationship between binning and die size is not linear. Going from 50->100mm2 will not affect binning the same way that going from 250->500mm2 will. The N31 and N32 GCDS are smaller than monolithic N21 and 22, so it may be reasonable to shrink our estimate further. Also, If there's sufficiently more need for a high clocked N31 than for a N32, everything may go out the window: GA102 is significantly larger than GA104, yet across SKUs, GA102 and GA104 boost similarly, and there are two GA102 SKUs that boost higher than the fastest GA104.
hurleybird
Newcomer
If true, AMD should send feelers out to Elon to partner for a super-binned 4.069 GHz N31 "Musk edition" that hits the 100 TFLOP mark exactly
If true, AMD should send feelers out to Elon to partner for a super-binned 4.069 GHz N31 "Musk edition" that hits the 100 TFLOP mark exactly
Musk would insist on a 4.20GHz version. Because he's an edge lord.
DavidGraham
Veteran
Right before the reveal of Ada, AMD is doubling down on their performance per watt leadership in a new blogpost, citing RDNA2 and RDNA3.
community.amd.com
Advancing Performance-Per-Watt to Benefit Gamers
By: Sam Naffziger, Senior Vice President, Corporate Fellow, and Product Technology Architect The demand for immersive, realistic gaming experiences is constantly pushing the boundaries of technology, driving enhancements to support features like raytracing, variable rate shading, and advanced...
Right before the reveal of Ada, AMD is doubling down on their performance per watt leadership in a new blogpost, citing RDNA2 and RDNA3.
Advancing Performance-Per-Watt to Benefit Gamers
By: Sam Naffziger, Senior Vice President, Corporate Fellow, and Product Technology Architect The demand for immersive, realistic gaming experiences is constantly pushing the boundaries of technology, driving enhancements to support features like raytracing, variable rate shading, and advanced...
Hmmm talking about performance-per-watt without talking about absolute performance usually isn’t a great sign. Presumably they would use that greater efficiency to deliver best in class performance too but they aren’t saying it.
Right before the reveal of Ada, AMD is doubling down on their performance per watt leadership in a new blogpost, citing RDNA2 and RDNA3.
Advancing Performance-Per-Watt to Benefit Gamers
By: Sam Naffziger, Senior Vice President, Corporate Fellow, and Product Technology Architect The demand for immersive, realistic gaming experiences is constantly pushing the boundaries of technology, driving enhancements to support features like raytracing, variable rate shading, and advanced...
So they don't expect to be the outright performance leader, but it sounds like they do expect to take the performance/watt crown.
Problem is that the overall performance crown doesn't seem that relevant this generation given the 4090's likely astronomical price, and the probable huge performance gulf between it and the affordable cards. If AMD can offer better performance/$ than Nvidia in the <$800 in both RT and raster then they could do very, very well this gen. Of course their RT performance is a massive question mark but after the relative disaster of RDNA2 in that regard I struggle to believe they haven't gone all out to fix that issue on RDNA3.
Even if that comes to pass, DLSS still offers a clear value add for Nvidia but FSR2 has at least blunted that - a little.
DegustatoR
Legend
They can say different things. They even continue saying that RDNA2 is somehow more power efficient than a process step behind it Ampere.
Launch the thing and lets see the actual benchmarks.
Launch the thing and lets see the actual benchmarks.
hurleybird
Newcomer
You're reading far too much into it. What is AMD supposed to say? That their unannounced products will beat Nvidia's unannounced products in performance? That's something neither company knows for sure, although Nvidia seemingly expects it to be close or they wouldn't be pushing power consumption so much. No, the only thing AMD knows for sure right now is that they're going to have a significant efficiency advantage, and that Nvidia is taking power consumption to unprecedented levels. Banging on the "Nvidia is hot and loud" narrative isn't just an obvious play, it's basically the only play at this point.
DavidGraham
Veteran
They could just say nothing, like always, like all the years before that, and wait for the actual reveal, before claiming to have perf/w leadership before any solid specs are shown.
DegustatoR
Legend
They don't know that "for sure" yet.
Up to 450W isn't "unprecedented", and that's for a top dog which can be whatever really if it ends up being the fastest.
A more "mainstream" solution will seemingly run at the same wattage as 6900XT/3090/etc.
It's also not true. Nvidia isn't getting any hotter or louder.
DegustatoR
Legend
Let's not pretend that Nvidia release only 4090 type GPUs shall we?
If they know they will beat Nvidia in performance/watt then they must know both Nvidia's performance and watts.
Exactly.
- Status
