Tuesday, August 31st 2021
AMD Reportedly Readying RX 6900 XTX, Bringing the Battle to NVIDIA RTX 3090
Graphics cards may be on their way to becoming unicorns that you can only pay for after finding the proverbial pot of gold from under a rainbow, but that doesn't mean AMD and NVIDIA will slow down their competition any time soon - especially in this market, there's a huge profit to be made. And AMD may just be finally readying their true halo product - a graphics card that aims to beat NVIDIA's RTX 3090 across the board. Twitter user CyberPunkCat shared an alleged AMD slide showcasing a new, overpowered RX 6900 XTX graphics card. AMD's naming scheme for their RX 6900 series may be slightly confusing nowadays: the original RX 6900 XT carries the Navi 21 XTX die, and AMD has recently released a higher-performance version of that Navi 21 chip in the form of the Navi 21 XTXH - which power the liquid-cooled versions of the RX 6900 XT, with higher overall clocks than the original GPU release. However, there hasn't been a change in the RX 6900 XT nomenclature - but this new slide suggests otherwise.
If the leaked slide is real (keep your NaCl ready, as always), it appears that the RX 6900 XTX might pair both the higher-performance Navi 21 XTXH chip with higher memory speeds. While both Navi 21 XT and Navi 21 XTXH both make use of 16 Gbps GDDR6 memory, the slide indicates that the RX 6900 XTX will feature 18 Gbps memory speeds, exploring another avenue for increased performance. This decision would bring an increase in maximum theoretical memory subsystem bandwidth from the 512 Gbps in the RX 6900 XT up to 576 Gbps - a 13% increase, which would not translate into a proportional increase in final performance. However, considering how our own reviews show that AMD's RX 6900 XT with the Navi 21 XTXH silicon is already between one and three percent faster than NVIDIA's RTX 3090, even a slight, 5% performance increase over that cards' performance means that AMD might be able to claim the performance crown for the mainstream market. It's been a while since that happened, hasn't it?
Sources:
CyberPunkCat @ Twitter, via Tom's Hardware
If the leaked slide is real (keep your NaCl ready, as always), it appears that the RX 6900 XTX might pair both the higher-performance Navi 21 XTXH chip with higher memory speeds. While both Navi 21 XT and Navi 21 XTXH both make use of 16 Gbps GDDR6 memory, the slide indicates that the RX 6900 XTX will feature 18 Gbps memory speeds, exploring another avenue for increased performance. This decision would bring an increase in maximum theoretical memory subsystem bandwidth from the 512 Gbps in the RX 6900 XT up to 576 Gbps - a 13% increase, which would not translate into a proportional increase in final performance. However, considering how our own reviews show that AMD's RX 6900 XT with the Navi 21 XTXH silicon is already between one and three percent faster than NVIDIA's RTX 3090, even a slight, 5% performance increase over that cards' performance means that AMD might be able to claim the performance crown for the mainstream market. It's been a while since that happened, hasn't it?
107 Comments on AMD Reportedly Readying RX 6900 XTX, Bringing the Battle to NVIDIA RTX 3090
TSMC themselves state 91.2 MTr/mm² for N7 (high density variation). A good rule of thumb is that high performance variation is 2/3 of that which would be ~60 MTr/mm². With logic being less dense than whatever the official numbers come from the 51 MTr/mm² for Navi21 is a respectable and expected result.
Ahh well, small sample of us clearly flogging a dead horse at this point, can certainly see from the post reactions how a few of you feel and that I should seek the answers I want from a much wider and more balanced audience. We are in the thread about a 6900XTX after all.
The poll sums up like this: 5 degrees of 'want' for emphasis on RT effects relative to the emphasis on rasterized performance.
And the 79% is the total of the poll outcomes that lean towards prioritizing raster performance - or at the very least that raster perf should never be LESS important than RT perf. Another way to say: RT is a nice to have, but not a must have.
And like I tried to explain earlier, a big part of that result is simply price and availability. Sure, if all things are equal you would pick the GPU with the bigger feature set. But they never really are even just on the basis of local differences. But also hardware wise; RDNA2 has unique selling points there, and so does Ampere. Its not a clear cut 'Nvidia is better' like it has been for quite some time.
Another element is the proprietary approach and how much trust you place in continued support. Nvidia has developed a track record there and it provides no guarantees.
Ill take the Car instead of a Stupid GPU:toast: