Nvidia is moving to 7 nm only for HPC GPUs in 2020. Late 2020-late H1 2021 is when next gen gaming GPUs we will get on 7 nm node, from Nvidia.
Nvidia is moving to 7 nm only for HPC GPUs in 2020. Late 2020-late H1 2021 is when next gen gaming GPUs we will get on 7 nm node, from Nvidia.The info points very much to next year being the year of Nvidia - they move onto 7nm so AMD looses it's node advantage which being as AMD still can't match Nvidia for efficiency with the node advantage is pretty bleak. In addition it'll be all about ray tracing and Nvidia will have second gen ray tracing cards out, where as AMD will have just released their first cards so are likely to get destroyed in ray tracing performance.
I still get the feeling AMD aren't massively interested in discrete gpu's - they'll take some console and apple wins, keep some presence in the descrete market and be happy with that. In the mean time all the money and focus goes on the cpu market, which tbh is understandable - it's the larger market and they are much better placed.
lol. you kidding right? Where do you think all the Zen 2 chiplets are coming from on 7nm? Hint: it's not GF.
Don't forget AMD's 7870's role in this which allowed mid range dies to come in at near flagship prices.
The point was comparing Intel designs with their 14nm vs intel 10nm.
AND Amd designs at TSMC 7nm vs TSMC 16nm and you can't do that because there never was an high performance x86 cpu at tsmc before zen2.
Because if there was you could have an idea what gains could Nvidia have there and you can't, you don't know what power and clocks will Nvidia get, could be much better, the same or could be even worst.
This image from Anandtech shows that high clock GPUs chips are very few, really few, I know it's on 5nm but 7nm could be similar and explain why Nvidia didn't do it, unless you think they don't want to be the leader and extent that lead:
More likely, Nvidia looked at TSMC as the only game in time for 7nm at the time, and took an educated bet that they didn't need process parity with AMD
More likely, Nvidia looked at TSMC as the only game in time for 7nm at the time, and took an educated bet that they didn't need process parity with AMD (I.E fighting over Huawei, Qualcomm & Apple for wafers and fab capacity) in order to be competitive with AMD, and could afford to wait until it was a more mature process with more supply before jumping in.
In hindsight that turned out to be a pretty good idea.
Nope. Until the cookies are out of the oven you are incapable of getting any information, how well your opponent will perform in terms of efficiency.More likely, Nvidia looked at TSMC as the only game in time for 7nm at the time, and took an educated bet that they didn't need process parity with AMD (I.E fighting over Huawei, Qualcomm & Apple for wafers and fab capacity) in order to be competitive with AMD, and could afford to wait until it was a more mature process with more supply before jumping in.
In hindsight that turned out to be a pretty good idea.
No, AMD/ATi always will be ahead of Nvidia in terms of process advantage.Maybe, Nvidia is skipping 7nm and going to 5nm. RDNA2 is also 7nm. That gives Nvidia room to fit 5nm GPUs in early w/o competition in HPC-GPU.
source?Maybe, Nvidia is skipping 7nm and going to 5nm. RDNA2 is also 7nm. That gives Nvidia room to fit 5nm GPUs in early w/o competition in HPC-GPU.
Nothing hard yet, but here is some Nvidia softies.source?
Just that? That's very pale. How about to add more:
- Too expensive for them
- Performance wouldn't increase much
- No better performance/watt
- Bad yields
- Much smaller dies, no high end cards initially
.....
If it was beneficial for them Nvidia would do it, I don't see how they would like Amd at performance (and power) parity with them, like I said they want to be the leader and extent that lead.
Staying at 12nm it wasnt NVIDIAs first choice. With 500-700mm2 dies you need 2x or 3x times more wafers for the same die capacity vs 7nm.
Also, because NV would not get any fab advantages (speed, lower power etc) at 12nm , they had to spend way more in to R&D for the new Turing Architecture unlike AMD that they got free speed and power reduction from using 7nm fabrication.
They both took a strategic decision early on, one bet on 7nm and spend less in R&D and the other opted for the more mature, cheaper and higher volume 12nm but had to spend more on R&D and had to live with very big dies.
Nope. Until the cookies are out of the oven you are incapable of getting any information, how well your opponent will perform in terms of efficiency.
So Nvidia was not able to draw any conclusions on 7 nm AMD GPUs directly before the tape out of those GPUs.
Which would be way past of any RND, of any design, from Nvidia.
There is way too many other reasons, however, why Nvidia hasn't got to 7 nm before AMD.
The point was comparing Intel designs with their 14nm vs intel 10nm.
AND Amd designs at TSMC 7nm vs TSMC 16nm and you can't do that because there never was an high performance x86 cpu at tsmc before zen2.
Why would they even bother and waste all the time, money, resources with the 7nm Vega?AMD wouldn't have readily adopted 7nm over their old 14nm process if it was as bad as you described.
Ok let's go out of x86.SPARC64 XII - Fujitsu - WikiChip
SPARC64 XII is a high-performance 64-bit dodeca-core SPARC microprocessor designed by Fujitsu and introduced in April 2017.en.wikichip.org
heres a 4ghz 12core cpu on tmsc 20nm, its effectively an 8 wide SMT4 core. Why limit to x64? everythings uops after decode.
Both of them were spending R&D on their respective new architectures anyway (Turing for NV, RDNA for AMD) so it is questionable if the transition to 7nm is a significant money-saver for AMD on R&D spending.
In a vacuum of course Nvidia would take 7nm over 12nm, but given the cost and supply situation for 7nm at the time, they made a trade-off to live with Tesla-sized dies on mature process, something they could do because their architecture was likely still good enough to compete in performance with an older process.
You cant compare Navi with big Turing. Turing supports a lot more features for a lot more markets Navi doesnt exist. With Raytracing Turing is 2x+ faster, the same with AI.
TU116 is 30-40% more efficient than Navi. AMD has designed a new gaming only architecture on a brand new process and cant beat a 12nm gaming product. Small Navi is DoA because Turing is such a beast on 12nm.
They are competing with small chips running at ruinously high clocks which does nothing for power efficiency.You cant compare Navi with big Turing. Turing supports a lot more features for a lot more markets Navi doesnt exist. With Raytracing Turing is 2x+ faster, the same with AI.
TU116 is 30-40% more efficient than Navi. AMD has designed a new gaming only architecture on a brand new process and cant beat a 12nm gaming product. Small Navi is DoA because Turing is such a beast on 12nm.
Could you imagine the cost of a 400mm2 7nm GPU, let alone the heat dissipation issues?They are competing with small chips running at ruinously high clocks which does nothing for power efficiency.
The delay in releasing Navi 10 also makes me think it has serious problems with that new uArch on a new process - not exactly a giant surprise when combining 2 new endeavors together in compute tech doubles the risk of significant errata in the first products.
I would expect the next iteration to show more improvement than would be expected of just 7nm+ process changes.
Also bare in mind nVidia have not jumped on TSMC 7nm despite AMD using it since the Vega 20 at the start of the year.
It probably doesn't work so well for larger chips yet (and Turing is huge) - the Matisse CCD's are only 74mm2, so Zen2's success doesn't necessarily translate to much larger chips, even Navi 14 is more than twice that.
I can - I doubt it will happen from AMD, at least on standard N7.Could you imagine the cost of a 400mm2 7nm GPU, let alone the heat dissipation issues?