AMD Big Navi Captured on photo, die could be 536mm2 in size

Published by

Click here to post a comment for AMD Big Navi Captured on photo, die could be 536mm2 in size on our message forum
https://forums.guru3d.com/data/avatars/m/260/260103.jpg
Thats a big biach right there. Gonna be a beast me hopes. I know it will perform amazingly in rasterized graphics, I just want it to be great at Ray Tracing.
https://forums.guru3d.com/data/avatars/m/189/189980.jpg
It will be a hot product!o_O What?
data/avatar/default/avatar02.webp
Maddness:

Thats a big biach right there. Gonna be a beast me hopes. I know it will perform amazingly in rasterized graphics, I just want it to be great at Ray Tracing.
I am skeptical for the size. Not the first time someone leaks a picture out of scale and then run a wild calculation. Back in January everyone and his dog was adamant that the Xbox X Series will be 405mm2 because of some leaked image. It was proven it was 10% smaller at 360.45mm2 in real. A 10%-15% variation of the image scale is small on bad quality image, but huge in real as is the difference between expected 466-470mm2 maximum and rumored 536mm2 based on that image.
https://forums.guru3d.com/data/avatars/m/164/164033.jpg
IF it ends up being 26.3 billion transistors clocked at 2.2ghz. The card will be powerful indeed. IF.
data/avatar/default/avatar36.webp
Little over three weeks to see what they got. πŸ™‚
data/avatar/default/avatar09.webp
Fediuld:

I am skeptical for the size. Not the first time someone leaks a picture out of scale and then run a wild calculation. Back in January everyone and his dog was adamant that the Xbox X Series will be 405mm2 because of some leaked image. It was proven it was 10% smaller at 360.45mm2 in real. A 10%-15% variation of the image scale is small on bad quality image, but huge in real as is the difference between expected 466-470mm2 maximum and rumored 536mm2 based on that image.
10% approximation on a blurry image seems good to me. Is indicative unless you want to know the exact transistor count
https://forums.guru3d.com/data/avatars/m/175/175902.jpg
"Big" on it doesn't mean anything, german tank "mauss" is bigger than "tiger"... it's just a picture, that is nice of course, and have it's place in the news too... dut don't jump too fast on the train πŸ˜‰ Just wait until it will on Hilbert's desk.
data/avatar/default/avatar38.webp
rl66:

"Big" on it doesn't mean anything, german tank "mauss" is bigger than "tiger"... it's just a picture, that is nice of course, and have it's place in the news too... dut don't jump too fast on the train πŸ˜‰ Just wait until it will on Hilbert's desk.
Depends are we talking about tracked combat vehicles overall, or only those with turret, or those who actually seen in combat. So if overall Jagdtiger is been the biggest vehicle saw actual combat, hands down. If we are talking about turreted tanks, M1A2 Abrams is the biggest tank ever used in combat to this day followed by the KΓΆnigstiger (King Tiger). Maus was one prototype like the GTX Volta, never saw usage πŸ˜›
https://forums.guru3d.com/data/avatars/m/256/256969.jpg
For those following everything nowadays: it's inline with rumors regarding 80CU and expected performance around 3080-3090, right? AMD HQ before both CPU and GPU October reveal: [SPOILER] https://i.imgur.com/urbKwmg.jpg [/SPOILER]
https://forums.guru3d.com/data/avatars/m/270/270008.jpg
Ryu5uzaku:

IF it ends up being 26.3 billion transistors clocked at 2.2ghz. The card will be powerful indeed. IF.
This sits in the middle of the 3070 and 3080 die sizes if true. Also there are rumors of some type of cache so I wouldn't be surprised if 30-40mm2 is dedicated to cache. Going to be interesting, popcorn ready.
https://forums.guru3d.com/data/avatars/m/164/164033.jpg
JamesSneed:

This sits in the middle of the 3070 and 3080 die sizes if true. Also there are rumors of some type of cache so I wouldn't be surprised if 30-40mm2 is dedicated to cache. Going to be interesting, popcorn ready.
When you take into account how well 5700 xt (the older 7nm) did against a way bigger 2070 (12nm). And considering that 7nm AMD uses now is good bit more dense than samsung 8nm which nvidia uses.
https://forums.guru3d.com/data/avatars/m/175/175902.jpg
Just the size of the Mauss compared to the size of the Tiger, a mouse bigger than a tiger πŸ™‚ I have seen the one in russia the remnant turret of the one that was destroyed mounted on the second chassis that was turret less.
https://forums.guru3d.com/data/avatars/m/196/196426.jpg
^ That's the most off-topic discussion I have seen in a while πŸ˜‰ Back on topic. Unless AMD really screws it up badly with the architecture (which I doubt they will), there's a good chance this giant chip will be faster than even 3090 ! - More than twice the size vs 5700!! - More clocks than 5000 series - Improvements in architecture (IPC, more efficient use of the shader cores) Just the size alone would be enough to equal NV's 3080... the rest are just the cherry on top. --- I also think this will be a 512-bit bus chip, literally doubled from RDNA1 (5700XT), not just in core counts, but EVERYTHING. It's been a long, long while since we've seen a 512-bit bus GPU ( Radeon HD 2900XT ). They could do it then, they can do it now. The leaks we've been seeing are from the mid-range chip, which will be a 256-bit one, and NOT from this BIG Navi. I also expect it to be priced accordingly, but probably not exceeding $999. In any case, a smack in the face of RTX 3090... making that chip's price even more silly than it already is.
data/avatar/default/avatar11.webp
wavetrex:

^ That's the most off-topic discussion I have seen in a while πŸ˜‰ Back on topic. Unless AMD really screws it up badly with the architecture (which I doubt they will), there's a good chance this giant chip will be faster than even 3090 ! - More than twice the size vs 5700!! - More clocks than 5000 series - Improvements in architecture (IPC, more efficient use of the shader cores) Just the size alone would be enough to equal NV's 3080... the rest are just the cherry on top. --- I also think this will be a 512-bit bus chip, literally doubled from RDNA1 (5700XT), not just in core counts, but EVERYTHING. It's been a long, long while since we've seen a 512-bit bus GPU ( Radeon HD 2900XT ). They could do it then, they can do it now. The leaks we've been seeing are from the mid-range chip, which will be a 256-bit one, and NOT from this BIG Navi. I also expect it to be priced accordingly, but probably not exceeding $999. In any case, a smack in the face of RTX 3090... making that chip's price even more silly than it already is.
Don't forget something equivalent to tensor cores takes space too. It can't be 2 5700 in the space of 2 5700 if you have to add also some other kind of processing area.
data/avatar/default/avatar13.webp
Just waiting on this AMD release to come out to be able to select the components of my next personal build. 3070 is the one on my mind at this moment. Let's hope AMD gets my attention too. Options are great! Of course the CPU will be AMD.
https://forums.guru3d.com/data/avatars/m/270/270008.jpg
@Hilbert Hagedoorn AMD Just trademarked "AMD Infinity Cache". Things are going to get interesting on Oct. 28th.
https://forums.guru3d.com/data/avatars/m/270/270008.jpg
asturur:

Don't forget something equivalent to tensor cores takes space too. It can't be 2 5700 in the space of 2 5700 if you have to add also some other kind of processing area.
Actually it can and would. You would not replicate everything 2x. There are some components of the GPU you one need one of no matter how many CU's. A 80CU version of the 5700XT, if AMD had made one back on RDNA1, would be a smaller die than 2 5700XT dies.
https://forums.guru3d.com/data/avatars/m/281/281256.jpg
Ahh heres hoping I can go all Team Red shortly am itching for new Hardware fix!
https://forums.guru3d.com/data/avatars/m/196/196426.jpg
We don't even know if NV's "tensor cores" and "RT cores" are real, and not just some extra transistors attached to the floating point units inside SMs, to support new instructions. And nobody outside nvidia really knows what are the actual instructions that go to the GPU, since everything is abstracted by the driver. The entire thing is a black box and we get some nice green squares on the "floor plan" of the GPU. If those RT cores were truly independent units, they wouldn't slow down the FP32 calculations so much, massively reducing FPS for raytracing games. My wild guess is that the FP32 units do the raytracing too, but "better" in Turing/Ampere due to extra instructions specific for this purpose. --- What has all got to do with AMD's chip ? Well, simple, AMD can most likely do raytracing as well in their FP32 units, by adding certain instructions which help with the line to triangle intersection calculation. So yes, the FP32 units might get a bit more complex and bigger, but not THAT much bigger. No fancy "mumbo-jumbo cores" needed.
data/avatar/default/avatar20.webp
JamesSneed:

@Hilbert Hagedoorn AMD Just trademarked "AMD Infinity Cache". Things are going to get interesting on Oct. 28th.
We know from the AMD Linux kernel signatures both CDNA and Navi2 support Infinity Fabric 2 interconnect. Now if Navi2 will have the same abilities with the CDNA card for directly linking 4 GPUs on 1 CPU we shall see. But we know CDNA-EPYC combo can do that and probably TR4 CPUs will do also. (IF not to be confused with CF)