AMD Radeon RX 7900 XT and XTX unboxing

Published by

Click here to post a comment for AMD Radeon RX 7900 XT and XTX unboxing on our message forum
https://forums.guru3d.com/data/avatars/m/181/181063.jpg
geogan:

Looking at these specs I compiled, this 7900XTX should beat the crap out of 4080 in raw raster performance... assuming equal shader processor performance... although ray-tracing performance unknown... and if it sells for that price 😱
7900 XTX vs RTX 4080

Architecture:       5nm vs 4nm
Shader Processors   12288 vs 9728
Memory            24GB G6 vs 16GB GDDR6X
Memory Bus        384-bit vs 256-bit
Memory Speed        20 Gbps vs ? Gbps
Infinity Cache        96 MB vs 64 MB (L2 cache)
Bandwidth        960 GB/s vs 716.8 GB/s
TBP            355W vs 320W
Pixel Rate       ? GPixel/s vs 280.6 GPixel/s
Texture Rate       ? GTexel/s vs 761.5 GTexel/s
Price           €999 (assuming 1:1 with dollar) vs €1500
It is a theoretical comparison but let's just wait for reviews; just an example why in practice there is more to AMD vs Nvidia battle than specs: the number of shader units doesn't necessarily translate in more performance (they were never clocked the same so no chance for same performance): 8800gt had 112 shader units and the 4830 had 640, still they were in the same performance league. I think that it will be exactly like last gen: AMD will win some, lose some (by very small margins) in rasterization and lose on RT, the incentive being the price IF Nvidia doesn't slash prices on the 4080 at the 999 barrier as rumored and AMD will not have the also "rumored" supply issues.
https://forums.guru3d.com/data/avatars/m/220/220214.jpg
barbacot:

It is a theoretical comparison but let's just wait for reviews; just an example why in practice there is more to AMD vs Nvidia battle than specs: the number of shader units doesn't necessarily translate in more performance (they were never clocked the same so no chance for same performance): 8800gt had 112 shader units and the 4830 had 640, still they were in the same performance league. I think that it will be exactly like last gen: AMD will win some, lose some (by very small margins) in rasterization and lose on RT, the incentive being the price IF Nvidia doesn't slash prices on the 4080 at the 999 barrier as rumored and AMD will not have the also "rumored" supply issues.
Yeah, that's why I said "assuming equal shader processor performance" - but as we know the SMs are NOT equal.
https://forums.guru3d.com/data/avatars/m/181/181063.jpg
geogan:

Yeah, that's why I said "assuming equal shader processor performance" - but as we know the SMs are NOT equal.
Yes, I saw - you were right to specify. But again I think that it will be like last gen... To tell the truth I can't wait for next week - new hardware real performance reviews are always exciting - the rumors, the naked truth, the drama, the intrigue 😎
https://forums.guru3d.com/data/avatars/m/271/271684.jpg
H83:

Shouldn't the back plate be vented or perfurated to allow the card to breathe better??? Anyway, looking forward for the review.
It is quite unusual that there isn't even a cutout in the area of the die. But I guess it makes little difference in thermal performance; all the heat generating components are on the other side after all. Though we might see some thermal pads between the backplate and PCB.
https://forums.guru3d.com/data/avatars/m/265/265988.jpg
WOW 🙂
Undying:

Hyped for the review. 🙂
Me Too 😀
https://forums.guru3d.com/data/avatars/m/260/260103.jpg
It's a fantastic looking card except for the backplate. C'mon AMD, that's the part that is usually seen the most. You can do better there.
data/avatar/default/avatar14.webp
Looks like a tight package, barely over two slots. I am sure these will jump quite a bit in price, but I appreciate the "somewhat" reasonable MSRP.
https://forums.guru3d.com/data/avatars/m/237/237478.jpg
@Hilbert Hagedoorn Can you confirm if a USB C to DP adapter comes with the card?
https://forums.guru3d.com/data/avatars/m/165/165326.jpg
Looking good 🙂
data/avatar/default/avatar28.webp
This is cool on paper, but for the past 2 or 3 gens high end AMD cards are woefully understocked for the first 6-9 months (which inflates the price) and I don't expect this launch to be any different.
https://forums.guru3d.com/data/avatars/m/220/220214.jpg
barbacot:

Yes, I saw - you were right to specify. But again I think that it will be like last gen... To tell the truth I can't wait for next week - new hardware real performance reviews are always exciting - the rumors, the naked truth, the drama, the intrigue 😎
Yep, but too late for me... I have made my bed already this generation and stuck with it. TBH I would have chosen this anyway for RTX reasons, only price is not great.
https://forums.guru3d.com/data/avatars/m/181/181063.jpg
tty8k:

7900xt - 1.35x faster than 6900xt and about 5% faster than 4080 on raster - 1250-1350e shop price 7900xtx -1.5x faster than 6900xt - 1400-1500e shop price There are only two problems. Stock and scalping (from aib/shop too)
IF you are nailing these numbers precisely to the point in reviews I'm gonna have to ask you about next week lottery numbers. After the expenses that I made this year on pc hardware and my baby (my car) it is winning the lottery or move with the whole family to my parents basement accepting that I am a failure and disappointment.
https://forums.guru3d.com/data/avatars/m/181/181063.jpg
geogan:

Yep, but too late for me... I have made my bed already this generation and stuck with it. TBH I would have chosen this anyway for RTX reasons, only price is not great.
I bought one for my wife as a birthday gift - I am impressed- runs cool, great power efficiency and great performance - it is (IMHO) really the best GPU from Nvidia since Pascal except, as you said, for the price.
https://forums.guru3d.com/data/avatars/m/79/79740.jpg
Thats the way .. uh-huh uh-uh .. I like it.
amd.png
data/avatar/default/avatar35.webp
CPC_RedDawn:

960GB/s of memory bandwidth and 5.3TB/s MCM bandwidth??? How is this bottlenecked by memory? If anything AMD normally bottleneck their cards by forcing hard power and OC limits like they did with the 6000 series only being able to push memory from 2000MHz to 2150MHz, unless this is what you mean and therefore we more than likely won't be able to push the memory any further. But 960GB/s should be plenty.
MCM is just another bottleneck that will ad extra latency issues, as raytracing and gddr wasn't already a problem. raytracing hates latency since it needs random access most of time outside cache. A huge L3 cache would brought much better benefits for sure.. AMD put 2000 limit on 6xx0 series for a reason: those mem chips were already pushed on limits on first release.
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
Alessio1989:

MCM is just another bottleneck that will ad extra latency issues, as raytracing and gddr wasn't already a problem. raytracing hates latency since it needs random access most of time outside cache. A huge L3 cache would brought much better benefits for sure.. AMD put 2000 limit on 6xx0 series for a reason: those mem chips were already pushed on limits on first release.
l3 cache is actually smaller on rdna3 than rdna2, and on separate chiplets.clocks are lower on n31 than n21. wonder how all that plays out in the end.
https://forums.guru3d.com/data/avatars/m/156/156348.jpg
imo by far the best looking reference card made by AMD. I actually like the look a lot and they would fit my build very well. Cooler looks good now it remains to be seen if it will perform good. Ray tracing is slowly starting to be relevant though and from the look of it (own AMD charts) RT performance will again be underwhelming 🙁 FSR (2.2) has been improving more than RT performance if AMD charts are accurate.
https://forums.guru3d.com/data/avatars/m/266/266726.jpg
Alessio1989:

MCM is just another bottleneck that will ad extra latency issues, as raytracing and gddr wasn't already a problem. raytracing hates latency since it needs random access most of time outside cache. A huge L3 cache would brought much better benefits for sure.. AMD put 2000 limit on 6xx0 series for a reason: those mem chips were already pushed on limits on first release.
latency is lower on rdna3 according to amds slides , they increased the fabric clock speed to compensate for the increased latency that come with chiplets. edit:slide in question https://cdn.mos.cms.futurecdn.net/gjVHASzBujhMHvVwtGdTib-970-80.jpg.webp
https://forums.guru3d.com/data/avatars/m/274/274577.jpg
im thinking of getting rx7900xt
https://forums.guru3d.com/data/avatars/m/216/216349.jpg
Alessio1989:

MCM is just another bottleneck that will ad extra latency issues, as raytracing and gddr wasn't already a problem. raytracing hates latency since it needs random access most of time outside cache. A huge L3 cache would brought much better benefits for sure.. AMD put 2000 limit on 6xx0 series for a reason: those mem chips were already pushed on limits on first release.
Maybe you are right about the extra latency but MCM is the future of GPUs in one way or another. After all, there`s a limit to how big monolithic GPUs can be.