Expect High-end Navi: AMD CEO

[Razbojnik]

I expect intel will make amd regret not stepping up their game earlier, and the Big Na...N word...cause it hurts me to hear about it already, so it's just a big N word...is gonna flop cause Ampere is looking too hot imo

 

[kings]

I believe when I see it.

AMD has been saying this since 2018, we are in 2020 and nothing...

 

[medi01]

..is gonna flop cause Ampere is looking too hot

Messages 2 (0.67/day)

Oh, great, Totally Not The Leather Man Shill One, where can we see the the hot Ampere?

 

[ixi]

Messages 2 (0.67/day)

Oh, great, Totally Not The Leather Man Shill One, where can we see the the hot Ampere?

Haha.

I need to see performance, noise and price performance for AMD and Nvidia upcoming... I can sense/feel full pc upgrade for me this year.

 

[Timbaloo]

Furthermore, if Nvidia is wedded to the idea of dedicated units for RT rather than divisible FP\INT, Nvidia's going to have to continue to maintain very large die sizes.

Navi 10 has 10.3x10^9 transistors while TU106 has 10.8x10^9. Both have very comparable performance (according to the TPU RX 5700 XT review). A simple shrink to 7nm (or similar) would clearly boost TU106 past Navi 10 while only needing ~5% more transistors. That RT units need a lot of transistors / die space is a myth.

Also Turing did introduce concurrent data paths for FP and INT. Actually iirc Volta did already.

 

[TheinsanegamerN]

As long as AMD refuses to compete on the high end, they will continue to fail at GPU penetration. The market speaks wonders, even with dramatic market improvement Nvidia still controlled 75% of dGPU shipments. But, I am guessing that much like big polaris, and late vega, that big navi will only release; if it even exists, after the 3080ti comes out, leaving AMD in second place once again, forced to cut margins on their GPUs to keep sales up.

It's unfortunate, I enjoyed my vega 64 quite a bit, as well as my freesync monitor, but I'll have to grab a 3080ti for an upgrade at this rate since AMD isnt interested in users like myself who want (and are willing to pay for) big GPUs.

They stated that years ago but RDNA is a new architecture targeting the high and low end market segments.

Is it? So far Navi has given us two low end chips with royally screwed up pricing structures and one upper mid range chip that, when OCed ballz to the wallz, barely scraped up enough power to match the over 3 year old 1080ti. "high end" performance has been solidly left in nvidia camp for years now, despite polaris being capable of far more, as radeon VII demonstrated, and rDNA seems to be following the same path, with big navi likely being a rDNA 2 product now.

Hmmm..... Still waiting for that 1080ti killer..... After my poor very very old gtx1080 died on me, went with a 5700XT, mainly because I can not and will not be part of feeding Ngreedia. But in the end I almost got 1080ti performance at a pretty decent price, so ya......

Ngreedia? Really? You do see AMD overpricing their low and mid range GPUs to make more cash off of RTG fans right, and AMDjacking the price of threadripper up as soon as they had a major performance edge, pushign a new platform after just two CPU generations, right? I mean, AMD finally graced you with performance you could have had 3 years ago, so ya......

 

[efikkan]

What happened to Navi 12? Abandoned in favor of 7nm+ probably because yields were too low? If yes, then Navi 21, on 7nm+ won't have any new architectural features, it's just a pilot card for the new manufacturing node not unlike RX 590 was compared to RX 580. VRS/RT will be in RDNA2 which AFAIK is debuting with Arcturus. 7nm+ w/ VRS/RT is too much just to call "Navi 21." That implies the lower Navi cards could also do VRS/RT but...it can't?

I've not been able to find anything substantive supporting Navi 12 being "big Navi", there have been recent reports of it having only 36 CUs. If there ever was a plan for "big Navi 1x", it was canceled long before tapeout. But this pattern we saw with Vega and Polaris, many refused to believe that there wasn't a bigger one right around the corner, especially with "Vega 11" being "bigger" and a "killer".

There are three known Navi 2x GPUs so far from drivers; Navi 21, Navi 22 and Navi 23. AMD may very well be changing the CU count for the second batch and get some benefits from 7nm EUV, but Navi 2x will still be Navi, and fundamentally scale like Navi. A doubling of CU count is unlikely, as it will result in an incredible power consumption.

AMD has scope to make a high end card, per nm, the 5700 XT is pretty good. Furthermore, if Nvidia is wedded to the idea of dedicated units for RT rather than divisible FP\INT, Nvidia's going to have to continue to maintain very large die sizes.

At best, Navi is the bare minimum to stay present in the market.
Even with the substantial node advantage, the tiny Navi 10 GPU manages to consume 225 W (board power), this is simply not scalable.
When Nvidia's next gen comes around, it will be on the same node as AMD, and Nvidia will have a new architecture, that will be rough unless AMD have a surprise new architecture in their back pocket.

 

[HD64G]

I would release the 60CU chip 1st, clearly winning 2080S for less money and once nVidia announces their next big chip, I would land the 80CU mammoth for $800 . That would make a mess in the nVidia marketing and get all the positive reviews 3 times more sound for customers. But who am I to teach AMD's marketeers...

 

[Fabio]

A competitor to the RTX 3070. If the rumors are true about the 50% perf boost, then the RTX 3060.

so for the nvidia price politics x more perf= x more price the new 2060 will cost kinda 800USD LOL

 

[yeeeeman]

We expect it from 2017 )

 

[CoolRonZ]

Ngreedia? Really? You do see AMD overpricing their low and mid range GPUs to make more cash off of RTG fans right, and AMDjacking the price of threadripper up as soon as they had a major performance edge, pushign a new platform after just two CPU generations, right? I mean, AMD finally graced you with performance you could have had 3 years ago, so ya......

Yup! Just like Apple and guess what Samsung their main competitor also price hiked. No doubt AMD followed suit, they definitely dropped the ball, but the 2080 was pretty much on par with a 1080ti for the exact same price too, but my 5700xt is still 90% the card for less than 60% the price. Its pretty much $/frame now. There is no bottom end pricing anymore, starts in the midrange. The tiering was skewed with Turing, but then again thats normal in the PC space, same with AMD charging pretty much $62USD per core with the 3990, I pretty much got my 3600X for half that per core on boxing day and I got a free game to boot. Can't wait for DRAM and Flash memory prices to go up again in 2020.

 

[Mouth of Sauron]

What are you even talking about???

You expect that bleeding edge fastest server chip ever be the proportional price as your mid-range consumer CPU? Seriously? What else do you expect - to get 20 games free with 3990x?

And you WISH that price of parts everyone needs rise? It will probably happen, but to wish it...

Bad Sir, you're a troll...

 

[CoolRonZ]

What are you even talking about???

You expect that bleeding edge fastest server chip ever be the proportional price as your mid-range consumer CPU? Seriously? What else do you expect - to get 20 games free with 3990x?

And you WISH that price of parts everyone needs rise? It will probably happen, but to wish it...

Bad Sir, you're a troll...

I cant compare a 3990X to my measly 3600X, it was more or a fact that the 3600X is a killer deal, as for a 3990X, thats an impressive chip, its still faster than dual XEONs costing $20,000($357/core), and lets face it the 9900K is $60usd per core. My sarcasm about memory and NAND prices being fixed, the giants dont like selling them with such small margins. ALL tech giants want to charge us more, thats a fact. And with the new consoles coming out, theres a very high demand for memory and NAND, thats definitely going to affect supply. But its company leaders like Apple, Nvidia, Intel and Samsung who set the bar and everyone else including AMD to a point follow suit. trolling 101, did I strike a nerve?

 

[SIGSEGV]

RX 5800 variant? RX 5900 variant?
so f*kin late. lame.

I expect nothing to the high-end Navi.
I prefer to build a new rig for my wife with RX 5700XT (cheaper than the second hand of 1080Ti)

 

[Mouth of Sauron]

trolling 101, did I strike a nerve?

No, you just made a confusing post... It's much clearer now what you meant to say...

 

[purecain]

AMD should release a card which equals the 2080ti at half the price and make it crossfire compatible. Using the pcie4 bus this should be an improved implementation. I would buy two straight away.
Ive been using my V100 for what feels like a lifetime. it averages around 45-60fps @4k on the latest titles with every setting on maximum quality. heres a stream from the end of the star wars game. Dont watch if you dont want to see the Darth Vader scenes. Spoiler alert!!!

A navi card or two that could offer that kind of performance would be amazing.
AMD could even set it up to dedicate a card to ray tracing and AA. there are unlimited possibilities… I'd love to be working in AMD's graphics department right about now.

 

[Casecutter]

IDK, Why... Big Navi isn't justifiable until AMD/RGT has excess wafer starts, which most would agree they don't. AMD has all their 7nm production that they can get now, they don't need to have some Big low yield part sucking wafer starts. AMD is smart as long a CPU's are selling, they've console parts to deliver, and as long as 85% of the discrete GPU is covered and selling robustly they're in no rush to compete.

Look at it like this... AMD not having a Big GPU is like saying Nvidia isn't making CPU's. AMD is their own company and only have to do what makes sense for them and their stock price. Sorry if your the "way less than 10%" of gamers who want AMD to come in shake things up just because Nvidia is charging whatever they like. I took a long 7+ years for AMD CPU's to be competitive, and it was like 2015-2016 that they walked away from Ultra-High End Enthusiast market. This too will pass...

 

[Razbojnik]

Messages 2 (0.67/day)

Oh, great, Totally Not The Leather Man Shill One, where can we see the the hot Ampere?

Hey nvidia has no need of putting it out cause amd has lost the battle with rtx...no point in beating a dead horse imo.

 

[efikkan]

IDK, Why... Big Navi isn't justifiable until AMD/RGT has excess wafer starts, which most would agree they don't. AMD has all their 7nm production that they can get now, they don't need to have some Big low yield part sucking wafer starts.

Considering how much greater the profit margin per chip would be, a "big Navi" would make very much sense. The yields are plenty good that they could in theory make a 50% bigger Navi without any major yield issues, it would be just a little bigger than Radeon VII. If there is anything "wasting" wafers it would be making low-end GPUs for OEMs, those are high volume and low margin products.

There are two primary reasons why they don't have big Navi (yet);
1) Their focus is custom chips for consoles, Sony and Microsoft fund a lot of their R&D budget. The PC market is always going to come second, until they shift focus.
2) Big Navi isn't feasible due to power issues. A 50% bigger Navi would consume ~300-350W to compete with RTX 2080 Ti (assuming it would even scale at all). Even if they could match the performance, it would be so hot that few would buy it.

AMD is smart as long a CPU's are selling, they've console parts to deliver, and as long as 85% of the discrete GPU is covered and selling robustly they're in no rush to compete.

In the long-term they are facing competition by Intel for the consoles, to the extent the consoles will continue beyond the upcoming generation.

I took a long 7+ years for AMD CPU's to be competitive, and it was like 2015-2016 that they walked away from Ultra-High End Enthusiast market. This too will pass...

AMD(ATI) used to be competitive, but has fallen behind.
For CPUs it took Intel failing for over 3 years, allowing for AMD to catch up. If Intel hadn't failed with their nodes, we would have Ice Lake in 2017/2018 and Sapphire Rapids now, just for perspective.
I hope you're not relying on Nvidia failing for AMD to become competitive in the GPU market too.

 

[Casecutter]

If there is anything "wasting" wafers it would be making low-end GPUs for OEMs, those are high volume and low margin products.

Their focus is custom chips for consoles, Sony and Microsoft fund a lot of their R&D budget. The PC market is always going to come second, until they shift focus.

On such points I'm in agreement. Although, I would say AMD and Lisa Sue had to hold onto the console business for funding the R&D, but also compel and inflict base game architecture and in some way the companies make the games.

In the long-term they are facing competition by Intel for the consoles, to the extent the consoles will continue beyond the upcoming generation.

Very true, I don't think Sony and Microsoft are yet truly "talking" to Intel, but Xe Graphics Technology is going to be perhaps compelling in 2 year when both would be eyeing next gen platforms. Sure, for Intel they would be Designer-Fab so they would really be a better spot to be a viable AIO supplier so it make sense for them, but for the next 5-7 years AMD has locked in revenue.

AMD(ATI) used to be competitive, but has fallen behind.

That's a story that has has many unfortunate wrong turns from 2013 till 2017, but I think AMD/RGT has been building back the last 3yr's given the R&D/funding constrains against what the Green Giant has at it's disposal. I'd would rather deliver the stuff that's competitive, than spend resources on something that would just harvest more bad press if just going to come up short like Vega and Fury.

I hope you're not relying on Nvidia failing for AMD to become competitive in the GPU market too.

Well, as you stated for Intel, things can go bad in lot's of ways . Nvidia is both moving on a supposed new architecture, while also on a new node process at Samsung. Nvidia has to really hope all things go as planed.

Lastly it seems people can't recall those Bulldozer or Fermi years, as said...

This too will pass.

 

[efikkan]

Nvidia is both moving on a supposed new architecture, while also on a new node process at Samsung. Nvidia has to really hope all things go as planed.

I believe Nvidia has stated they will use both TSMC and Samsung, with TSMC supplying the highest volume. I haven't been able to locate the original source though.

Everything is shaping up to be great, well except for things being a bit "too quiet". Nvidia is either keeping things even more secret than before, or there could potentially be delays. Usually there are telltale signs of where they are in the development process, and so far we haven't seen anything to indicate whether they are 3 months, 6 months or 9 months away. I think it's >3 months, since there are usually a lot of leaks when AiBs get their hands on cards.

Lastly it seems people can't recall those Bulldozer or Fermi years, as said...

At least Fermi was outperforming the competition, at the price of energy efficiency. That can't be said for Bulldozer.