[Rumor, Tweaktown] AMD to launch next-gen Navi graphics cards at E3

[DisEnchantment]

https://videocardz.com/65521/amd-vega-10-and-vega-20-slides-revealed

The last of the roadmaps. WhyCry needs to find new leaks about Next-Gen


The strange thing was that bridgman in Phoronix mentioned that going forward SW enablement will be done before product launches. But for Navi there is not a single commit on freedesktop. Nothing.
Instead something in LLVM shows up which is a very obscure patch.
In the past, looking at the gfx9 code you can tell whether a new PCI id was added, you can tell dpm clocks and so on. Looking at the code I had discovered earlier that RR was gonna use normal RAM and I was able to find out Vega 20 max dpm clocks.
This time nothing, even though the product was gonna launch with three months or so as per Earnings call. For Vega , enablement code in Linux was available almost 7-8 months before launch.

 

[Krteq]

....But for Navi there is not a single commit on freedesktop. Nothing.
Instead something in LLVM shows up which is a very obscure patch.
In the past, looking at the gfx9 code you can tell whether a new PCI id was added, you can tell dpm clocks and so on. Looking at the code I had discovered earlier that RR was gonna use normal RAM and I was able to find out Vega 20 max dpm clocks.
This time nothing, even though the product was gonna launch with three months or so as per Earnings call. For Vega , enablement code in Linux was available almost 7-8 months before launch.

Hmm, over 11K+ lines of code only for LLVM compiler... I'm pretty sure they will submit AMDGPU updates in merging window for Linux kernel 5.3 which will be released in the beginning of autumn.

 

[DisEnchantment]

Right.
When you check the activity on AMDKFD/AMDGPU and mesa it is strangely very quiet. They were fixing bugs mainly. Even the new SMU framework suddenly went quiet
They are not merging anything to the staging repo.

They are not very keen to share things. In contrast bridgman just mentioned a year ago that the strategy is to make the SW ready and merged much in advance, (see VegaM, Vega20, Raven2 etc) but for Navi it is the opposite.

For sure their internal work is much advanced, I mean I saw the git logs that most of the Vega enablement was done 10 months before Vega Launch but was pushed to staging branch about 6 months or so before launch.

 

[Krteq]

Well, we've seen Vega prototypes running half a year before actual release. With Navi we have nothing except some old photos of ES board.

And yep, they are suspiciously quiet about Navi. It reminds me RV770 launch.

 

[Yotsugi]

Instead something in LLVM shows up which is a very obscure patch.

It does say planty about things to come.

This time nothing, even though the product was gonna launch with three months or so as per Earnings call. For Vega , enablement code in Linux was available almost 7-8 months before launch.

Silence is golden.

And yep, they are suspiciously quiet about Navi. It reminds me RV770 launch

Or any other ATi/AMD launch before RTG came to be.

 

[caswow]

Compare new Intel PR for their upcoming who knows when coming gpu on twitter talking every day about their gpu and amds current approach . AMDs old PR coincidentally at intel now

 

[coercitiv]

Compare new Intel PR for their upcoming who knows when coming gpu on twitter talking every day about their gpu and amds current approach . AMDs old PR coincidentally at intel now

About that, don't forget how talkative AMDs old PR was when it came to Fury, Polaris, Vega. Half of that talk turned out to be wishful thinking... to use a crude euphemism.

 

[caswow]

AMDs current approach is way better. hated their old pr. Nobody knew about r7 for consumer. imaging them talking up r7 months prior its release

 

[maddie]

It does say planty about things to come.

Silence is golden.

Or any other ATi/AMD launch before RTG came to be.

One thing about Raja Koduri, he really really couldn't stop talking. I'm actually shocked that he's been so quiet at Intel.

Anyone remember the attempted whisky trap in the PC Perspective interview?

Edit: Quiet here is a relative term.

 

[Yotsugi]

I'm actually shocked that he's been so quiet at Intel.

I dunno, the Architecture Day happened, and they're also very loud about their upcoming dGPUs on the internet.

 

[Head1985]

AdoredTV rumors:
Worse IPC than vega
Its power hungry
Way worse prices

 

[Yotsugi]

AdoredTV rumors

These are not rumors anymore, just his schizofrenia.
I guess his Polaris self-humilation wasn't enough.
The dies don't make sense either, somebody, make me a more believable table.

 

[Muhammed]

These are not rumors anymore, just his schizofrenia.
I guess his Polaris self-humilation wasn't enough.
The dies don't make sense either, somebody, make me a more believable table.

Stop spinning the situation, Navi was prone to fail and AMD's silence and delay on it is "telling", the fact they released a Radeon VII is also telling ..

 

[DisEnchantment]

Lately I grew very weary of AdoredTV.

1. According to AMD's roadmap there will be no Navi Instinct, I posted the roadmap few posts before. Vega 20/MI60 will take that spot until MI-NEXT. David Wang also showed this in his roadmap recently.

Andy Parma of AMD mentioned some weeks ago, about the MI-NEXT Instinct product succeeding the MI60

David Wang

2. How can MI Instinct be delayed to Q1 2020 when AMD's roadmap from the Earnings call few days ago mentioned NEXT-GEN for 2020.

http://ir.amd.com/static-files/9c985e84-bbb6-4e23-99bd-dcbb21f18592

3. Navi cannot fail, it will go into the PS5 and will stay there for the next 7 years. AMD will throw everything at it to make it work.

I agree to some extent on the TDP projection but thats about it.

 

[maddie]

These are not rumors anymore, just his schizofrenia.
I guess his Polaris self-humilation wasn't enough.
The dies don't make sense either, somebody, make me a more believable table.

What is it? 2 die designs with 64, 60, 56 , 52, 48 compute units.

Sounds very strange at the least.

 

[Yotsugi]

I agree to some extent on the TDP projection but thats about it.

The entire thing just makes so little sense it's hilarious.
Mediocre FUD/10.

2 die designs with 64, 60, 56 , 52, 48 compute units.

AMD paying $300m for a N7 tapeout to add ~8 CUs is like, hilarious, fails reality check so hard I don't even.
The #2 hilarious thing is a 20CU test die that appeared in CompuBench being magically absent from here.

Sounds very strange at the least.

Because it's FUD, good old FUD at this point.
Not a very well done FUD to boot.

 

[DooKey]

These are not rumors anymore, just his schizofrenia.
I guess his Polaris self-humilation wasn't enough.
The dies don't make sense either, somebody, make me a more believable table.

And yet he is the piped piper for so many...

 

[ozzy702]

And yet he is the piped piper for so many...

Only for rabid AMD fanbois that haven't been paying attention. Jim hasn't been credible for quite some time and what little "insider" connections he had in the past appear to have dried up and now it's just a 100% speculation station. I watched that video earlier today and as always I'm amazed at how much time he can kill with so little information (even information that's wrong). Good on him for getting views.

 

[Yotsugi]

And yet he is the piped piper for so many...

Well, can't cure the stupid.

Only for rabid AMD fanbois that haven't been paying attention

I don't even anymore.
Doing a basic reality check isn't hard.
Using google to find any previous metions of gfx10 isn't hard either.

 

[NostaSeronx]

As of recent, AMD needs a revitalized upper low-end stack for consumers.

Vega Enthusiast => February 7, 2019 // 7nm(N7)
Polaris Performance => November 15, 2018 // 12nm(12LP)
Polaris Mainstream => May 2017 // 14nm(14LPP)
Polaris Budget => April 3, 2018 // 14nm(14LPP)

Not within the mainline stack;
Vega M = Polaris Performance-Mainstream => Q1 2018 // 14nm
Vega Pro = Vega Performance-Mainstream => November 2018 // 14nm


RX 560 needs the most replacing. Since, a competitor got Vega M and a customer got an exclusive run of a product.

Greater performance than RX560, HBM2e 1024-bit at 8 GB at 2.4 GHz through 3.2 GHz, small die at high volume;
From this => 1024:64:16 @ 1175-1275 w/ 128-bit GDDR5 7 GHz
Two options on HBM2e; 16 Gigabit stack(Pleb model)//8 GB or 24 Gigabit stack(Pro model)//12 GB

Higher D0 should mean smaller dies, as D0 gets lower then bigger dies.
Navi-S (RX550/RX560 replacement), 7nm ||-> Navi-P (RX570-RX590), 7nm+ ||-> Navi-HE (Vega xx), 6nm

 

[tamz_msc]

How can MI Instinct be delayed to Q1 2020 when AMD's roadmap from the Earnings call few days ago mentioned NEXT-GEN for 2020.

Instinct roadmap is separate from consumer-GPU roadmap.

 

[tamz_msc]

Why the disbelief surrounding these rumors that Navi could end up being a turd?

 

[DiogoDX]

Why the disbelief surrounding these rumors that Navi could end up being a turd?

Is the most realistic rumor until now based on AMD GPU history in the last five years.

Last good GPU for AMD was Hawaii in freaking 2013.

 

[Head1985]

Is the most realistic rumor until now based on AMD GPU history in the last five years.

Last good GPU for AMD was Hawaii in freaking 2013.

Yes last good performance cards was hawaii, but those cards was late and with no AIB cards at launch and they just launched 2 cards and rest just rebranded.So i think last good cards from low-end to high-end was in 6970 era on terascale(remember 6950 unlocking?).And last better cards than nv was 5870/5850 era.
So more like since 2009/2010
I am sure it was because Rory Read destroyed AMD GPU division at that time and most good engineers left AMD.

 

[AtenRa]

AdoredTV rumors:
Worse IPC than vega
Its power hungry
Way worse prices

View attachment 5883

If the above chart is legit, I dont see why NAVI is a fail.