NVIDIA ‘Ampere’ 8nm Graphics Cards

[Troezar]

The info I have is basically the other way around to what is in his video.

Should be interesting to see which is correct. Is there enough capacity in TSMC 7nm to supply huge numbers of Nvidia cards? I thought most of their output had been bought up by other companies?

 

[Rroff]

Should be interesting to see which is correct. Is there enough capacity in TSMC 7nm to supply huge numbers of Nvidia cards? I thought most of their output had been bought up by other companies?

Some customers are transitioning to 5nm so some capacity is opening up.

 

[Richdog]

Sucks.

Definitely need to keep them backed up :s

Had that happen with my custom build Clevo gaming laptop - can't remember how long I'd had it but heat (I'm pretty sure) got the GPU and RAM in one go - one day it was fine then within a week both were showing signs of data corruption.

All of my sensitive docs are backed up to my NAS and then cloud so no worries there Rroff... the Nvme drive inside will still be fine so I can recover my game data and copy it over. It's more the inconvenience and knowledge that it is likely never going to function again as a whole that irks me. With a desktop I could have just replaced a component myself.

Should be interesting to see which is correct. Is there enough capacity in TSMC 7nm to supply huge numbers of Nvidia cards? I thought most of their output had been bought up by other companies?

A quick google shows that AMD and Nvidia reportedly bought the excess capacity https://wccftech.com/nvidia-amd-ramped-orders-tsmc-2020/

 

[Troezar]

A quick google shows that AMD and Nvidia reportedly bought the excess capacity https://wccftech.com/nvidia-amd-ramped-orders-tsmc-2020/

It definitely looks like they are using both TSMC and Samsung then as Samsung was confirmed some time ago. Which is the better process and which has more volume I've no idea? I presume the cheaper higher volume one will be for the mainstream cards.

 

[BigBANGtheory]

We’ve known that for ages though what isn’t clear and still isn’t is how Nvidia will divide the manufacturing

 

[Troezar]

We’ve known that for ages though what isn’t clear and still isn’t is how Nvidia will divide the manufacturing

I'm assuming that can be deduced if we know how those two processes compare. I don't have that information so hoping someone with more in depth knowledge of chip manufacturing can apply some logic to it and enlighten us

 

[JediFragger]

For anyone who hasn't seen;

 

[Grim5]

For anyone who hasn't seen;

* Ampere is an evolution of Turing
* Ampere will be used for both HPC and Gaming cards, but manufacturing process differs
* Most of the gaming cards are on Samsung's 8nm process, all the HPC and the 3080ti/Titan are on TSMC's 7nm+ process.
* Launches with new utility called NVCache - this leverages your PCIE4 SSD or DDR4 RAM to "enhance gaming load times and help with VRAM data storage"
* RTX Voice will be a launch feature, targeting no performance hit.
* New overhauled Nvidia control panel available at launch
* DLSS 3.0 available at launch - works in any game that uses TAA
* Cuda version 11 is supported and available at launch
* 7th gen NVENC released, support for 8k 60fps and H264/H265
* IPC has increased 15%
* L2 cache has been doubled
* All cards to have boost clock around 1900mhz
* Double the Tensor cores per SM unit
* Ray Tracing - targeting up to 4 times performance improvement compared to Turing (Expect RTX3060 to have ray tracing performance of a RTX 2080ti or better and so forth)
* Ray Tracing - In most games, the target is to not have any performance hit from using ray tracing or if there is, just a small one
* New memory compression algorithm.
* VRAM sizes won't change much - offset by new compression tech and NVCache - RTX3080ti comes in at 12GB VRAM
* Memory bandwidth won't change much - still limited by what GDDR6 16Gbps is capable of
* Higher power draw than Turing - expect cards to draw power like Pascal but not as efficient as Turing - i.e RTX3080ti will have 280w TDP compared to 250w on the RTX2080ti
* All cards are called RTX, GTX has been burried for good
* Launch date is in September, final date still to be confirmed

 

[JediFragger]

* New memory compression algorithm.

Uses the Tensor cores too apparently, sounds like it could be pretty exciting.

 

[Adam123]

Are we expecting these to be pcie 4?

 

[JediFragger]

Are we expecting these to be pcie 4?

4 or a jump to 5, can't imagine them being 3 still.

 

[TNA]

Sounds good to me. I have PCIE 4 already and happy to hear (if true) that the RT performance will be even better than the minimum 2x improvement I expected. Also if true happy to hear they will have a new control panel as it currently is from windows xp days, slow and ugly. Good to see them catching up with AMD in that regard

The RTX 3070 will be a stonking card if priced right for me, as all I ever need is 60fps at 4K which will be easy to achieve on all my current steam library and any future titles I can use DLSS if need be

 

[Perfect_Chaos]

Was very tempted to get a 2000 series card as a stop gap but stopped myself from that, those rumours sound good to me.

Cheaper than expected is also a good sign, but being cheaper may not be by much anyway. We'll see.

 

[danlightbulb]

One of the best tech you tube channels ive seen.

He seems excited for the shootout between AMD and Nvidia and is expecting performance to be close at the top level.

Just hope we see a proper launch this year.

 

[TNA]

Was very tempted to get a 2000 series card as a stop gap but stopped myself from that, those rumours sound good to me.

Cheaper than expected is also a good sign, but being cheaper may not be by much anyway. We'll see.

Yeah. I had a strong feeling the 3000 series would be very good so I skipped the 2000 series. Only thing is I expected them out in Q1 this year, not Q3, but oh well, sounds like they will be worth the wait

Funny how in the video he says Turing will not age well and if you got it all you was was a guinea pig. Haha.

 

[JediFragger]

so I skipped the 2000 series.

Good call tbh, and Pascal was still a very good architecture. I got my Ti at a very good price so won't lose on it at all when I come to sell just before the new ones launch (more than likely, maybe £50 tops)

Good that nVidia are finally revamping their software stack, AMD have been embarrassing them for years on this point.

 

[P.B]

* Ampere is an evolution of Turing
* Ampere will be used for both HPC and Gaming cards, but manufacturing process differs
* Most of the gaming cards are on Samsung's 8nm process, all the HPC and the 3080ti/Titan are on TSMC's 7nm+ process.
* Launches with new utility called NVCache - this leverages your PCIE4 SSD or DDR4 RAM to "enhance gaming load times and help with VRAM data storage"
* RTX Voice will be a launch feature, targeting no performance hit.
* New overhauled Nvidia control panel available at launch
* DLSS 3.0 available at launch - works in any game that uses TAA
* Cuda version 11 is supported and available at launch
* 7th gen NVENC released, support for 8k 60fps and H264/H265
* IPC has increased 15%
* L2 cache has been doubled
* All cards to have boost clock around 1900mhz
* Double the Tensor cores per SM unit
* Ray Tracing - targeting up to 4 times performance improvement compared to Turing (Expect RTX3060 to have ray tracing performance of a RTX 2080ti or better and so forth)
* Ray Tracing - In most games, the target is to not have any performance hit from using ray tracing or if there is, just a small one
* New memory compression algorithm.
* VRAM sizes won't change much - offset by new compression tech and NVCache - RTX3080ti comes in at 12GB VRAM
* Memory bandwidth won't change much - still limited by what GDDR6 16Gbps is capable of
* Higher power draw than Turing - expect cards to draw power like Pascal but not as efficient as Turing - i.e RTX3080ti will have 280w TDP compared to 250w on the RTX2080ti
* All cards are called RTX, GTX has been burried for good
* Launch date is in September, final date still to be confirmed

Thank you for condensing that

 

[TNA]

Good call tbh, and Pascal was still a very good architecture. I got my Ti at a very good price so won't lose on it at all when I come to sell just before the new ones launch (more than likely, maybe £50 tops)

Good that nVidia are finally revamping their software stack, AMD have been embarrassing them for years on this point.

Lucky you, if I could have got a 2080Ti and knew I would even just lose £100 I may have gone for it, but majority who got one will likely lose more like £500+. Trouble is there were not many must play games that came out for me during the Turing life cycle.

This year there will be quite a few and if I feel the RTX 3070 will not be enough I may just grab a 3080TI for a few months, play them (take a small hit) then sell on. One of the things that concern me with the 3070 is it may still have 8GB RAM which would be disappointing. I want 12GB going forward.

 

[Perfect_Chaos]

Yeah. I had a strong feeling the 3000 series would be very good so I skipped the 2000 series. Only thing is I expected them out in Q1 this year, not Q3, but oh well, sounds like they will be worth the wait

Funny how in the video he says Turing will not age well and if you got it all you was was a guinea pig. Haha.

to many tech companies

 

[Random Guy Esq.]

Thanks for that summary @Grim5.

It would be interesting to compare NVCache to the systems the next gen consoles are going to use. It's one of the things that genuinely intrigued me about the PS5 for example.