NVIDIA Ampere GA102-300-A1 GPU (GeForce RTX 3090) Caught on Camera

Published by

Click here to post a comment for NVIDIA Ampere GA102-300-A1 GPU (GeForce RTX 3090) Caught on Camera on our message forum
https://forums.guru3d.com/data/avatars/m/183/183694.jpg
@Hilbert Hagedoorn: Where did you find it's made on 7nm? the reports are that it's Samsung's 8nm (10nm+). Suddenly you report it's 7nm without a valid source. I would like the see the source for this.
https://forums.guru3d.com/data/avatars/m/273/273678.jpg
Cidious:

@Hilbert Hagedoorn: Where did you find it's made on 7nm? the reports are that it's Samsung's 8nm (10nm+). Suddenly you report it's 7nm without a valid source. I would like the see the source for this.
the made in TW stamp on it.......
https://forums.guru3d.com/data/avatars/m/256/256969.jpg
TW= Taiwan/TSMC 7nm ?
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
Yeah, it's usually stamped where the package is assembled and not where the die is fabricated.
data/avatar/default/avatar25.webp
Rumors about these things are still all over the map even a week before launch. The only thing I believe is true at this point is that all of the cards at the current price points are going up by $100 USD. The 2080Ti replacement goes to $1400, the 2080 to $800, and the 2070 to $600. There is also a rumored card at the $2000 price point, and there's a lot of speculation that the card pictured above (with a fully/nearly fully unlocked die and maximum RAM) will sit at, with a version with half as much RAM and a few more blocks turned off sitting at the $1400 point, which matches the current 2080Ti and RTX Titan. The other consistent rumor is the performance, stating a one step increase plus 35% (so a 3070 will be 35% faster than a 2080, a 3080 will be 35% faster than a 2080Ti, and a 3090 will be 65% faster than a 2080Ti), but those numbers are REALLY high and I will believe them when I see some reviews. If true, that $100 step up will probably be worthwhile, since it makes the 3070 and 3060 a bargain at their price points compared to the current cards (which isn't too hard, since the 2000 series were never a good deal for the money).
https://forums.guru3d.com/data/avatars/m/197/197287.jpg
Cidious:

@Hilbert Hagedoorn: Where did you find it's made on 7nm? the reports are that it's Samsung's 8nm (10nm+). Suddenly you report it's 7nm without a valid source. I would like the see the source for this.
You're questioning a rumor that is not another rumor asking why the new rumor is not in line with the other rumor when rumors are rumors no matter which way you spin it? Don't understand your motivation for the post. 7nm TSMC, 8nm Samsung, doesn't matter and products haven't been released.
data/avatar/default/avatar15.webp
illrigger:

Rumors about these things are still all over the map even a week before launch. The only thing I believe is true at this point is that all of the cards at the current price points are going up by $100 USD. The 2080Ti replacement goes to $1400, the 2080 to $800, and the 2070 to $600. There is also a rumored card at the $2000 price point, and there's a lot of speculation that the card pictured above (with a fully/nearly fully unlocked die and maximum RAM) will sit at, with a version with half as much RAM and a few more blocks turned off sitting at the $1400 point, which matches the current 2080Ti and RTX Titan. The other consistent rumor is the performance, stating a one step increase plus 35% (so a 3070 will be 35% faster than a 2080, a 3080 will be 35% faster than a 2080Ti, and a 3090 will be 65% faster than a 2080Ti), but those numbers are REALLY high and I will believe them when I see some reviews. If true, that $100 step up will probably be worthwhile, since it makes the 3070 and 3060 a bargain at their price points compared to the current cards (which isn't too hard, since the 2000 series were never a good deal for the money).
The performance rumor is the easiest one to believe. With a combination of smaller node and massive TDP increase, I honestly feel like those numbers are a conservative/underwhelming.
https://forums.guru3d.com/data/avatars/m/16/16662.jpg
Administrator
Cidious:

@Hilbert Hagedoorn: Where did you find it's made on 7nm? the reports are that it's Samsung's 8nm (10nm+). Suddenly you report it's 7nm without a valid source. I would like the see the source for this.
One twitter user 'reports' 8nm Samsung (8LPP actually) and it's a fact all of a sudden? Nobody outside the NV bubble really knows, but I tend to stick to logic. Remember, Ampere already has been released in the form of the A100, that's factual a 7nm chip fabricated at TSMC, indicating NVIDIA has targeted 7nm for its design. The photo you see above is listing TW, that's again TSMC. It's a big chip, NVIDIA will want as small as possible wafer fabrication. So logic dictates it's going to be 7nm TSMC. However, if there isn't enough production capacity at TSMC or prices are too high, surely the Samsung route is an option. So could I be wrong, heck yeah? Does it matter? Heck no. In fact, NVIDIA might be sourcing both in the end to keep up demand.
data/avatar/default/avatar04.webp
Hilbert Hagedoorn:

One twitter user 'reports' 8nm Samsung (8LPP actually) and it's a fact all of a sudden? Nobody outside the NV bubble really knows, but I tend to stick to logic. Remember, Ampere already has been released in the form of the A100, that's factual a 7nm chip fabricated at TSMC, indicating NVIDIA has targeted 7nm for its design. The photo you see above is listing TW, that's again TSMC. It's a big chip, NVIDIA will want as small as possible wafer fabrication. So logic dictates it's going to be 7nm TSMC. However, if there isn't enough production capacity at TSMC, surely the Samsung route is an option. So could I be wrong, heck yeah? Does it matter? Heck no. In fact, NVIDIA might be sourcing both in the end to keep up demand.
The thing that makes me doubt it's TSMC, is the massive disparity between the GA100 TDP and the rumored TDP of these upcoming GPU's. If they can hit 300w with 55 billion transistors, you think they could do better than 350 on a cut-down design.
data/avatar/default/avatar40.webp
Cidious:

the reports are that it's Samsung's 8nm (10nm+).
Is this just wishful thinking outside what someone tweeted? Jensen already said most Ampere capacity would be TSMC with a little going to Samsung.
https://forums.guru3d.com/data/avatars/m/63/63170.jpg
This could be TSMC, and 7nm. It could also be a Limited Edition 3090 Founders Edition 24GB, with the Freaky cooler. Put Limited Edition on something, and it will sell, even at $1400 w/o Tax. GA104 could be Samsung, ie : The Majority of the Sales.
https://forums.guru3d.com/data/avatars/m/248/248627.jpg
with all the rumours flying all over the place i would not be surprised if they use a variety of nodes to make the next gen cards maybe even still 12nm for the low end cards 8nm for the mid-range and 7nm for the top tier cards. In the end its the performance vs cost that matters.
https://forums.guru3d.com/data/avatars/m/265/265988.jpg
Anybody have a guess on die size? 680mm² to 700mm² ? I'm not sure why they need that new power connector (if this model has it) , i'm seeing this as a 300W TDP card. Maybe when the RT cores are active they use an extra 100W?
data/avatar/default/avatar40.webp
A100 is 826mm².
https://forums.guru3d.com/data/avatars/m/224/224952.jpg
I dont think the image is real, the white writing looks superimposed.
https://forums.guru3d.com/data/avatars/m/224/224952.jpg
Aura89:

Not entirely certain why people think this. There's nothing funky about it at all, and gpu model numbers have been white in the past. https://forums.macrumors.com/attachments/dsc_0085-jpg.765636/
Its the first thing that stood out to me as soon as I saw the image, the white text doesnt look like it belongs. I'm not saying they dont have white text on them normally, just that white text in that image doesnt look right.