AMD Navi 23 ‘NVIDIA Killer’ GPU Rumored to Support Hardware Ray Tracing, Coming Next Year

[Rroff]

I know they did this with the 5700 but think about how upset it will make those early adopters of the 3800/3900 who have suffered with the upward trending prices and poor availability. I think they'd be shooting themselves in the foot.

nVidia have done it a few times in the past and I wouldn't be surprised if they do it again.

 

[shankly1985]

I would be surprised if Amd showed their best hand yesterday. I think it clearly saying 6000 series tells us all that GPU wasn't full big Navi.

 

[Wrinkly]

3090 is still 2 SMs short of the full die - but that isn't a huge uplift. Lots of potential if they move Ampere (gaming) to 7nm though (from what I've heard Ampere exists on 7nm not just GA100 and is likely going to be used on mobile parts).

Really? I posted on one of AdoredTV's videos that I wouldn't be surprised to see a node shrink with Ampere given the movement in AI and silicon compilers.

 

[Soapysage]

3090 is still 2 SMs short of the full die - but that isn't a huge uplift. Lots of potential if they move Ampere (gaming) to 7nm though (from what I've heard Ampere exists on 7nm not just GA100 and is likely going to be used on mobile parts).

And yet the 7nm capacity at TSMC is fully booked so no idea where they'll suddenly get the wafers to make 7nm Ampere. Plus you can't just plop an architecture on a different node, each have very different silicon levels qualities that you need to account for and due to NDAs between companies, those at Nvidia that worked on making it work on Samsungs 8nm can't work on making it work at TSMCs 7nm, Samsung won't allow their node details to be taken to TSMC and vice versa for that matter.

 

[z10m]

And yet the 7nm capacity at TSMC is fully booked so no idea where they'll suddenly get the wafers to make 7nm Ampere. Plus you can't just plop an architecture on a different node, each have very different silicon levels qualities that you need to account for and due to NDAs between companies, those at Nvidia that worked on making it work on Samsungs 8nm can't work on making it work at TSMCs 7nm, Samsung won't allow their node details to be taken to TSMC and vice versa for that matter.

Oh really.? Then how come iPhone soc was manufactured in the past by both TSMC and Samsung.? Same iPhone, same soc but two different chip manufacturers.

https://arstechnica.com/gadgets/2015/10/samsung-vs-tsmc-comparing-the-battery-life-of-two-apple-a9s/

 

[Rroff]

And yet the 7nm capacity at TSMC is fully booked so no idea where they'll suddenly get the wafers to make 7nm Ampere.

That assume nVidia hasn't already scheduled it in with the allocation they've already bought - these things go into action months in advance.

Plus you can't just plop an architecture on a different node, each have very different silicon levels qualities that you need to account for

This is less of an issue for nVidia than people often say - their design labs and failure analysis labs, etc. are way way more advanced than the industry average. (They can literally turn around silicon issues in 24 hours that take a month or more for other companies).

and due to NDAs between companies, those at Nvidia that worked on making it work on Samsungs 8nm can't work on making it work at TSMCs 7nm, Samsung won't allow their node details to be taken to TSMC and vice versa for that matter.

I take it this is based on the comments by Daniel Nenni? it isn't quite like that and neither have a confidentiality agreement quite like he was saying - I'm not even quite sure what he was basing his comments on as no one in the industry work quite like he was saying. As can be evidenced by Pascal existing on both Samsung and TSMC. EDIT: And above there are other examples where it happens like Apple.

 

[elroberto]

Will there be a card for my price range of 200 at a push 250, i got the vega 56 last year when it was reduced from near 300 to 249 and thought was a bargain for its specs, had a 670 then a 780ti for a short while before the vega 56 pulse. Im wondering if i sold the 780ti and the vega 56 pulse i might be able to get something new when the new cards come out, not looking for the best but if theres something near 2080 range for cheap i might bite as might be nice to try some ray tracing games or dx12 ultimate compatible ones later on. I mean if there isnt, im not fussed, i can play games ok.

Tho tbh i probs should upgrade the mb ram and cpu first lol but i really dont want to upgrade that stuff as its boring and takes effort, rather just keep gettin stuff i can just swap easily in 2 mins.

I imagine there'll be some sort of Nvidia Sku at about 2060 power and I would imagine AMD will finally produce 480/580 replacement at about 5700/XT levels

 

[elroberto]

Oh really.? Then how come iPhone soc was manufactured in the past by both TSMC and Samsung.? Same iPhone, same soc but two different chip manufacturers.

There's way more profit in the GA100 pro cards though, they won't pass that up and they haven't got that many 7nm wafers to go around. AMD do have a lot more capacity there.

 

[Soapysage]

There's way more profit in the GA100 pro cards though, they won't pass that up and they haven't got that many 7nm wafers to go around. AMD do have a lot more capacity there.

Exactly this, even if they do have capacity at TSMC already, they'll use it for the GA100 cards, same with AMD and keeping their best chiplets for Epyc and giving Ryzen hand me downs.

 

[AMDPower]

I really hate to speculate but indication are, based on AMD preview and rumours , Big Navi to be appx 3080 performance. But with possibly smaller die and more power efficient so this leaves room for AIB to overclock the GPU, I can really see AIB overclocked version with 10-15% more performance bring its performance close to 3090. I wonder if AMD would release a AIO type cooler like the VEGA before. leather jacket man will cry if he looses his performance crown.

the price is now what decides if AMD takes this to the bank or not. NVIDIA ampere has flopped with lack of availability bad price.

Lets see what happens but it would seem this year we get amazing performance for aboutn£600-700 range bad not £1300 madness of 2080TI.

don't you love competition.

My First all AMD system in almost 10 years, looking forward to this day.

 

[Nablaoperator]

Found this in rogame's twitter fee:
https://patents.justia.com/patent/10795825

Its supposed to be cache compression tech broadly implemented as:
L1: Compress Twice
L2: Compress Once
L3: Uncompressed

• Is this a known factor
• Will these result in larger or smaller cache designs
• Can this explain Navis low bandwidth? (I still cant see this happening)

Edit: Someone pls read it.. i just had a 5 min glance

 

[morta]

I know they did this with the 5700 but think about how upset it will make those early adopters of the 3800/3900 who have suffered with the upward trending prices and poor availability. I think they'd be shooting themselves in the foot.

lol yea im one of them ..still got my 3080 on preorder.lol can still cancel anytime though. but i dont think i will.

 

[ThunderLicker]

Don't know if anyone has touched on this but what i find interesting as someone at 3440x1440 and couldn't give a crap about 4k at the moment.

If big navi is within spitting distance of the 3080 at 4k, ampere doesn't seem to scale as well as the resolution drops compared to previous gens. If navi doesn't have this issue it could look very good as a 1440p card.

 

[Rroff]

There's way more profit in the GA100 pro cards though, they won't pass that up and they haven't got that many 7nm wafers to go around. AMD do have a lot more capacity there.

You don't normally build a whole wafer with entirely GA100 size dies (even with the kind of money they make from them) - nVidia have quite a lot of stuff in production on TSMC 7nm not just GA100 anyhow - they bought up a ton of production awhile back at the same time AMD increased their orders.

 

[w00dy85]

Don't know if anyone has touched on this but what i find interesting as someone at 3440x1440 and couldn't give a crap about 4k at the moment.

If big navi is within spitting distance of the 3080 at 4k, ampere doesn't seem to scale as well as the resolution drops compared to previous gens. If navi doesn't have this issue it could look very good as a 1440p card.

I have been thinking about this to, I also have a freesync monitor and find my self looking at Navi 21 more.

 

[Soapysage]

I have been thinking about this to, I also have a freesync monitor and find my self looking at Navi 21 more.

Because RDNA2 is designed around purely gaming whereas Ampere has compute in mind, RDNA2 should be a lot better at 1080p and 1440p.

 

[gpuerrilla]

Don't know if anyone has touched on this but what i find interesting as someone at 3440x1440 and couldn't give a crap about 4k at the moment.

If big navi is within spitting distance of the 3080 at 4k, ampere doesn't seem to scale as well as the resolution drops compared to previous gens. If navi doesn't have this issue it could look very good as a 1440p card.

I have been thinking about this to, I also have a freesync monitor and find my self looking at Navi 21 more.

Well its good news as it absolutely seems to be the better option for overall 1440p. However AMD presentation states is delivering the 4k experience, so with some fine wine and console port love, you could be in for a nice card. I also have freesync monitor, I mean if it was garbage I would look a the 3070 or 3080; but from what i am seeing, the navi21 option is looking good. Just need availability.

 

[KentMan]

Because RDNA2 is designed around purely gaming whereas Ampere has compute in mind, RDNA2 should be a lot better at 1080p and 1440p.

I am hoping this is the case, 3440x1440p 100hz Freesync here, 32" Ultrawide, love my monitor, upgraded from the BenQ 1440p 144hz 27" Screen, which my kids have now and is also a cracking screen. But the extra screen estate going wide is awesome, and i thought i would hate curved but i actually love it.

Im thinking the high end Navi will allow me to play modern stuff with bells and whistles at my res and mhz quite comfortably for a while, but i am also no opposed to turning down some settings either.

 

[Scougar]

If the pricing of the cpus is anything to go by, expect the prices to be higher. Especially if they now have RT ability.

 

[Chuk_Chuk]

If the pricing of the cpus is anything to go by, expect the prices to be higher. Especially if they now have RT ability.

The pricing of the CPUs is based on the fact that they are better than Intel in every way. If AMD is better than Nvidia in everyway they will be priced as so.

The only thing we can gleam is that AMD are less likely to low ball as they have done in the past.