Wednesday, August 28th 2024
AMD RDNA 4 GPU Memory and Infinity Cache Configurations Surface
AMD's next generation RDNA 4 graphics architecture will see the company focus on the performance segment of the market. The company is rumored to not be making a successor to the enthusiast-segment "Navi 21" and "Navi 31" chips based on RDNA 4, and will instead focus on improving performance and efficiency in the most high-volume segments, just like the original RDNA-powered generation, the Radeon RX 5000 series. There are two chips in the new RDNA 4 generation that have hit the rumor mill, the "Navi 48" and the "Navi 44." The "Navi 48" is the faster of the two, powering the top SKUs in this generation, while the "Navi 44" is expected to be the mid-tier chip.
According to Kepler_L2, a reliable source with GPU leaks, and VideoCardz, which connected the tweet to the RDNA 4 generation, the top "Navi 48" silicon is expected to feature a 256-bit wide GDDR6 memory interface—so there's no upgrade to GDDR7. The top SKU based on this chip, the "Navi 48 XTX," will feature a memory speed of 20 Gbps, for 640 GB/s of memory bandwidth. The next-best SKU, codenamed "Navi 48 XT," will feature a slightly lower 18 Gbps memory speed at the same bus-width, for 576 GB/s of memory bandwidth. The "Navi 44" chip has a respectable 192-bit wide memory bus, and its top SKU will feature a 19 Gbps speed, for 456 GB/s of bandwidth on tap.
Another set of rumors from the same sources also point to the Infinity Cache sizes of these chips. "Navi 48" comes with 64 MB of it, which will be available on both the "Navi 48 XTX" and "Navi 48 XT," while the "Navi 44" silicon comes with 48 MB of it. We are hearing from multiple sources that the "Navi 4x" GPU family will stick to traditional monolithic silicon designs, and not venture out into chiplet disaggregation like the company did with the "Navi 31" and the "Navi 32."
Yet another set of rumors, these from Moore's Law is Dead, talk about how AMD's design focus with RDNA 4 will be to ace performance, performance-per-Watt, and performance cost of ray tracing, in the segments of the market that NVIDIA makes the most volumes in, if not the most margins in. MLID points to the likelihood of the ray tracing performance improvements riding on there being not one, but two ray accelerators per compute unit, with a greater degree of fixed-function acceleration for the ray tracing workflow (i.e. less of it will be delegated to the programmable shaders).
Sources:
Kepler_L2 (memory speeds), Wccftech, VideoCardz (memory speeds), Kepler_L2 (cache size), VideoCardz (cache size), Moore's Law is Dead (YouTube)
According to Kepler_L2, a reliable source with GPU leaks, and VideoCardz, which connected the tweet to the RDNA 4 generation, the top "Navi 48" silicon is expected to feature a 256-bit wide GDDR6 memory interface—so there's no upgrade to GDDR7. The top SKU based on this chip, the "Navi 48 XTX," will feature a memory speed of 20 Gbps, for 640 GB/s of memory bandwidth. The next-best SKU, codenamed "Navi 48 XT," will feature a slightly lower 18 Gbps memory speed at the same bus-width, for 576 GB/s of memory bandwidth. The "Navi 44" chip has a respectable 192-bit wide memory bus, and its top SKU will feature a 19 Gbps speed, for 456 GB/s of bandwidth on tap.
Yet another set of rumors, these from Moore's Law is Dead, talk about how AMD's design focus with RDNA 4 will be to ace performance, performance-per-Watt, and performance cost of ray tracing, in the segments of the market that NVIDIA makes the most volumes in, if not the most margins in. MLID points to the likelihood of the ray tracing performance improvements riding on there being not one, but two ray accelerators per compute unit, with a greater degree of fixed-function acceleration for the ray tracing workflow (i.e. less of it will be delegated to the programmable shaders).
104 Comments on AMD RDNA 4 GPU Memory and Infinity Cache Configurations Surface
The RX 6900 XT has 5120 shaders
The RX 7800 XT has 3840 shaders.
The RX 7800 XT performs nearly the same as the RX 6900 XT but with just 75% of the shaders and roughly 80% of the power consumption.
RT, it’s between the 3090 and 3090Ti, which have an RT gap of about 12% between them.
256 bit vs 384.
64 mb vs 96mb l2 vs l3
+ amd has a big l2 of 6mb (you know... that's the last big cache of a 3090ti...)
and large L1 L0, so they're using bandwidth terribly bad, it's the architecture being extremely narrow and overfed and badly utilized.
Maybe getting tile based renderer working would help some scenarios, get the double fp working better and in more situations, drivers, matrix to help with certain tasks.
So many things, they're lagging behind so bad now and yet their trying desperately to make a WGP\Dual CU do everything without adding things to them, it's like they're trying to push and hammer down data(memory bandwidth through cache and bus) into a CU and magically thinking it'll work.
www.techpowerup.com/gpu-specs/radeon-rx-7800-xt.c3839
www.techpowerup.com/gpu-specs/radeon-rx-6900-xt.c3481
It's not an earth-shattering increase in RT performance per core but it's certainly noteworthy.
I think you're confused with regard the shader counts of the RDNA3 lineup but there hasn't been a massive increase on that front.
Also regarding Mindfactory... Single store in the EU, AMD friendly market. I maintain it.Yet fake frames with fake reflections are some of the star features they've chosen to copy...The RTX 4090 came out first and it was clear they could not match it. This "it's positioned against the 4080" together with a price reduction is literally the only thing AMD could do in that situation.
The fact that 7900 XTX is a much larger processor with a clearly higher bill of materials remains unchanged.The 7900 XTX does not come even close to the 4090 in raster performance. It's 1-4% vs. 4080, 1-2% vs 4080 Super.
RDNA3 has been the most boring release AMD has ever done, not broken like RDNA1, but offering no plus over RDNA2 and only increasing prices enough to not make everyone angry.
Focusing on the mid and low end (the market they literally abandoned right now) when it's obvious they are in no place to match XX90 NVIDIA products in hardware or software at the moment is the right move.
2. Best selling GPU model of the generation. Outsold the entire Radeon stack. Demand is so extreme, there are shortages for some of its board components
3. The RX 6800 is 4 years old, not two. So you got a previous generation card on a discount - that doesn't really count. MSRP by MSRP (they launched at roughly the same cost) even with all the cuts in the 4070's configuration, the 4070 comes far ahead as a product - it is faster, its drivers are of superior quality (a studio branch for productivity is also offered at no extra cost), it is more power efficient - and in 2 years, it'll have the same age and be just as devalued as your RX 6800 is today - in other words, there is nothing special or praiseworthy about your graphics card, it's just an earlier generation card past its prime that obviously still does its job, as it always hasSupply and demand
Steam Hardware Survey:
It's quite the scroll to find it but damn.
And of course any time I have some kind of graphics problem and need to default to something that just works: BAM!
This thing is everywhere. It's so everywhere that it became the new religion on the Chinese chopping block of chips for a number of years and is now so long in the tooth that we're just finally starting to see it happen to other cards either because the performance now sucks or DX feature lockout is kicking people out of the gaming hobby or the crap encoder is finally starting to force everyone out of service.This is the hard reality. At the end of the day, people don't give a damn about GPUs. They just want a display adapter that can play the game. That means not playing around with voltage settings for weeks trying to find the sweet spot where it doesn't crash (if there is one), not struggling with balancing fans and noise when tesselation breaks and turns one squirrely mesh into pure nightmare fuel until reboot, not dicking around with a bunch of Bitcoin miner market bullshit and playing momentum monkey with drops just trying to acquire ONE of these cards and that's all assuming they're not fighting to diagnose one dud after another like I had to deal with on 7000 series. What I'm saying is there isn't a good price.
The same fiasco has just kicked off with 5000 series and will happen to 6000 very soon. I'm not saying it hasn't already, I'm talking MASS SCALE. This is the only thing keeping me in a loop of second guessing, which keeps me out of the 2nd hand market for chips. It's bad enough that every delivery of 7000 series seems to be a bomb. I can only hope that 8000 series arrives as expected (functional) and delivers on good enough quantity and price because it took several months to find anything good for previous generations and I still don't have a card.
The performance should be worlds better than what I have now. That's the bar. Don't really care where they fit between 7000 series SKUs, just that enough of them hit the shelves before all the scalpers figure out AMD pulled a massive disinfo campaign to finally get these cards in the hands of real customers.
I also tend to play some Warzone nowadays and for whatever reason RDNA3 is stupid fast in that game, faster even than the 4090 at most resolutions.
I would say being 43% faster is a plus..I think they will at least match the 7900xtx but most seem to think otherwise. I guess we'll see soon enough
MantleVulkan, shared memory, Hairworks et al?The real competitor for RTX 4080 is the RX 7900 GRE.
www.techpowerup.com/gpu-specs/radeon-rx-7900-gre.c4166
www.techpowerup.com/gpu-specs/geforce-rtx-4080.c3888You must compare shader-to-shader. 6144 vs 5120.
XFX Radeon RX 7900 XTX Magnetic Air Review - Relative Performance | TechPowerUp
4080 is 23%, 28% and 35% faster than 7900GRE at 1080p, 1440p and 2160p respectively.
On the other hand, 7900XTX is 2%, 3% and 5% faster than 4080 at 1080p, 1440p and 2160p respectively.
Edit: 4090 is 13%, 19% and 23% faster than 7900XTX in the same graphs.
DLSS - or FSR for that matter - works on top of that.
What complicates things is the double ALU thing although it seems to have helped even less than same in Nvidia's case.
Less Infinity Cache but given evolution/optimization of the size of that on both AMD and Nvidia newer generations this has negligible impact.
Other than that - clocks are up 6-15% and VRAM bandwidth up 22%.
Based on the last TPU GPU review 7800XT is overall 22-23% faster than 6800 which basically matches the expectations based on specs.
In RT, 27-34% with gap increasing with resolution. There is a nice little jump there - AMD clearly did improve the RT performance.
Performance is equal.
www.techpowerup.com/gpu-specs/radeon-rx-6650-xt.c3898
www.techpowerup.com/gpu-specs/radeon-rx-7600.c4153