Caporegime
Purveyors of leaks are wrong all the time. It's just a bit of fun and more about who's the 'least wrong'.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
** The AMD Navi Thread **
- Thread starter subbytna
- Start date
Purveyors of leaks are wrong all the time. It's just a bit of fun and more about who's the 'least wrong'.
Associate
- Joined
- 17 Sep 2018
- Posts
- 1,474
Yea, I remember him doing a video defending why his RTX leaks weren't really wrong.
It's kind of intelligent guess play. Although the prices are where it has to be well and truly way off.
Soldato
And take out all the general/enterprise compute stuff and you're still left with the good old 4 shader engines that have been holding GCN back for a long time.
A 7nm-centric design will help performance of course, and if those shader engines have been beefed up then great, but it's still going to be a GCN derivative with all the trappings that come from it.
So I don't understand what you're getting at: gaming-focussed GCN design without the enterprise compute hardware is suddenly going to bang out Radeon VII and above performance? Um...
Errr, I don't know about you but most leaks that have been most true.
What has been hit and miss are speculations, ones not based on leaks or road maps, just assumptions what the companies would take such as how to counter X product.
But the leaks themselves have been true for the most part.
GCN is a failed architecture. Don't know why you keep insisting on it and promoting Navi as being GCN.
Failed? No,
Just needs a rework on the front end to improve scheduling throughput. It's a perfect time to go 6x wide Geometry engine on the 7nm now though.
Soldato
Oh wind your neck in. There is significantly more discussion, rumour and leaked information that states Navi is the last iteration of GCN than it being anything else. In fact, I've only ever seen WCCFTech say Navi is anything other than GCN, and we know what those guys are like. And like everything else with Navi, we literally have nothing other than rumour.
I'm not "promoting" anything, sunshine, I'm merely discussing used the common consensus as my baseline.
Associate
- Joined
- 17 Sep 2018
- Posts
- 1,474
Going back to the benchmark find/leak. Now it could be fake but the card was near Vega 56/1660 ti levels with only 20 compute units. For comparison the RX 580 had 36 compute units, the 570 32 and the Vega cards are 56 and 64 units respectively. It makes you question 'if that is a real card and not a fake, what will the performance be with 64 compute units?'
Ofcourse the benchmark leak and AdoredTV can't both be true because of the compute units performance.
The thing is if Sony/Microsoft have been investing in AMDs new arch, then surely, they demand just that. The new arch, new tech instead of last gen on 7nm. But maybe there's middle ground where GCN can be improved.
Ofcourse they could be putting some of that into PS5 Pro to keep PS5 costs down.
Also looking back to PS4 launch the Radeon card that went in the PS4 was a big jump on previous gen.
Ofcourse the benchmark leak and AdoredTV can't both be true because of the compute units performance.
The thing is if Sony/Microsoft have been investing in AMDs new arch, then surely, they demand just that. The new arch, new tech instead of last gen on 7nm. But maybe there's middle ground where GCN can be improved.
Ofcourse they could be putting some of that into PS5 Pro to keep PS5 costs down.
Also looking back to PS4 launch the Radeon card that went in the PS4 was a big jump on previous gen.
Really........really.......I mean really.
The mind boggles, how can you even consider thinking that let alone typing that out.
Soldato
This is what I've suggested before when the "Navi 20 beats 2080 Ti with ray tracing" rumours started, WCCFTech started saying Navi wasn't GCN and 4K8K waded in with his absolutes.
AMD have been in severe financial turmoil for a long time, there is no way they could've afforded to fun R&D for Zen, Zen 2 AND post-GCN Navi all at the same time. Unless Sony paid for Navi.
Associate
- Joined
- 17 Sep 2018
- Posts
- 1,474
Well we know the Navi project has been high priority with many shifting onto it. So whether GCN or not, there is going to be plenty of new tech taking it up a level.
The most exciting rumour is that Navi 20 won't be that much faster than Navi 10. So we can hope there is a full fat Navi 10 competing with either the 2070 or 2080
Soldato
It all depends on how accurate Adored Jim's latest video is. Ever since the initial leak way back when, I always felt there'd be a card above the RX 3080 would push into RTX2080/Radeon VII territory. His latest video says the same thing, as well as another one even further up.
Although I'm not sure about this "special edition" nonsense; either AMD have made a Navi 20 or they haven't, I can't see an explicitly designed product being a limited run.
Time will tell...
Although I'm not sure about this "special edition" nonsense; either AMD have made a Navi 20 or they haven't, I can't see an explicitly designed product being a limited run.
Time will tell...
Last edited:
Soldato
Isn't the rumour that Navi "light" powers the PS5 at 2080 performance, it would go logically that a non "light" version would be more like 2080ti or somewhere between the 2, but what i dont understand is that Radeon vii sits there in the spectrum, so eith navi is everything below R7 or it is one down and below and one above?
Soldato
I would ignore the R7 when thinking about this, it's a repurposed data centre part (Radeon Instinct) used to keep AMDs foot in the door at the high end until an alternative shows up
Soldato
So you think they will just slip a new line up over the top of it?
And that the navi non "light" would be 2080ti performance?
Soldato
You're using Navi "light" as a phrase? This all gets some wonderfully mashed up now, can any of us keep track on all the rumours and phrases any more?
It seems every month the goal posts for Navi performance keep shifting up and up and up. The optimist in me would like to think those leaking information are getting more up-to-date stuff and Navi is looking better than originally anticipated. The cynic (or realist?) in me thinks somebody is trying to sabotage the entire thing, pushing expectations even higher than a hype train could manage so when Navi does land it's going to be a crushing disappointment and everybody buys a GTX 1600 series.
The PS5 stuff muddies the waters even more in that I don't think the Navi going in that console is representative of Navi on the desktop. Certainly, if the PS5 does have ray tracing acceleration, it sure as hell won't be a standard component, at least not the mid-range Navi 10 and 12 we all thought was the product coming.
As for the Radeon VII, that thing was never anything more than a PR stunt made possible by a perfect storm of Nvidia failures. It is not a real product, it is not a sustainable product and the second AMD have something that can replace it they will. Radeon VII performance is for all intents and purposes RTX 2080/GTX 1080 Ti performance. That means you need a Navi card that is Vega 64 +25% (ish) to get there. I long suggested that there would be such a card based on a full Navi 10, and the latest AdoredTV information (for whatever the hell it's worth) actually has such a card.
Even before that video came out and complemented my crazy theories, I still think this:
RX 3070 (full Navi 12) = Vega 56, 8GB GDDR6
RX 3080 (cut down Navi 10) = Vega 64 +15%, 8GB GDDR6
RX 3090 (full Navi 10) = Vega 64 +25% (i.e. a tiny bit under Radeon VII/RTX 2080/GTX 1080 Ti), 11GB GDDR6
Radeon VII = RTX 2080/GTX 1080 Ti, 16GB HBM2
So the RX 3090 nicely fills the performance and price gap underneath Radeon VII without treading on its toes: it gives you almost the same performance but doesn't give you as much VRAM capacity or speed. Radeon VII is still for those who want balls-deep on Radeon gaming, or want the compute power.
Radeon VII is replaced by the alleged Navi 20 with ray tracing and whatnot. The only reason I don't see the RX 3090 (as I call it) from being the direct Radeon VII replacement is because of the 16GB VRAM precedent set. AMD make a big deal of 16GB in 4K gaming and the 1TB/s throughput, but 16GB GDDR6 would burn a lot of power and PCB space and won't operate anywhere near as fast, so could clearly be interpreted as a backward step for AMD.
Soldato
The radeon vii must have been thought of way before the rtx kerfuffle though, you can't just magic up A gpu on the spot. But I mostly agree with you on the navi lineup except i think there will be a 3090x with your naming scheme that will be avoid radeonvii
Soldato
Well, you can if you have cancelled projects you can reopen. Remember when Mike Rayfield "left" RTG? Amongst the disgruntled chatter from unnamed "employees", one of the things they were annoyed about was the "unviable" consumer Vega 20 project which was binned off. Now, you can't call something "unviable" if you've not done any work on it. It stands to reason that RTG did start work on a consumer Vega 20 with the PCB to match, but somewhere down the line a decision was made to cancel the project (that's probably where the $700 build figure came from in the first place).
It is entirely possible that Radeon VII is largely based on that cancelled consumer Vega 20 project. Dig out the PCB design, make a few tweaks to support the Instinct packages (assuming there ever was a difference between Instinct and consumer Vega 20 in the first place), bin some failed MI50 packages to see which ones can have the clocks ramped up, solder the whole thing together, slap a beautiful but half-assed cooler on it and release with appallingly unfinished drivers.
3 month turnaround, boom, done. Because they didn't actually have to design anything. And the tiny amount of cards available at launch really does smack of "we literally have 30 days to make 1,000 PCBs and coolers".
Good shout on differentiating Navi from Vega with the RX moniker, for all my fevered typing that never even entered my head, even with RTX and GTX staring me right in the face
Yeah, because by using GCN, the Radeons lost all the advantages they had enjoyed prior to that - performance per square mm, performance per watt, absolute performance crown or at least healthy competition in that, etc.
If you argue, at least try to defend your point.
This is probably just a stupid theoretical assumption to what would have happened to the Vega 64 if the compute shaders are removed, and then this performance slapped onto the poor Navi.