Rage3D Discussion Area

Rage3D Discussion Area (http://www.rage3d.com/board/index.php)
-   AMD Radeon Discussion and Support (http://www.rage3d.com/board/forumdisplay.php?f=45)
-   -   RDNA 3 Rumor Discussion (http://www.rage3d.com/board/showthread.php?t=34052049)

pax Jan 23, 2021 07:53 PM

Quote:

Originally Posted by LordHawkwind (Post 1338263119)
Supergeil whilst the 5700XT was a "high end" AMD card it was basically a mid range card when you took Nvidia cards into account.

That card was never meant to compete against Nvidia's enthusiast card the 2080ti, the 5700XT competed against the 2070/Super which was Nvidia's mid range card.

The MSRP of the 6800XT is £649 so about 50% more expensive than a 5700XT. The 6800XT competes against the 3080 which is one of Nvidia's enthusiast cards and the MSRP of the 3080 is £649.

AMD released the 6800XT/6900XT to compete in the enthusiast segment against the 3080/3090 which is the first time they've done that in over a decade.

There was a recent info release of an engineering sample of a 5900xt that was never put out due to excessive power issues. Problems that only got fixed in the 6000 gen. It wouldve been out about same time frame as the 5700xt. RDNA is a on a roughly yearly schedule. Also mass volume of 5nm that apple is already using will be had fall 2021. Makes no sense to wait once that is out.

bill dennison Jan 23, 2021 09:06 PM

Quote:

Originally Posted by pax (Post 1338263146)
There was a recent info release of an engineering sample of a 5900xt that was never put out due to excessive power issues. Problems that only got fixed in the 6000 gen. It wouldve been out about same time frame as the 5700xt. RDNA is a on a roughly yearly schedule. Also mass volume of 5nm that apple is already using will be had fall 2021. Makes no sense to wait once that is out.

https://tekdeeps.com/amd-radeon-rx-5...from-big-navi/

https://videocardz.com/101494/asrock...mer-rx-6800-xt

it was still better than the 3090/80 :lol:

LordHawkwind Jan 25, 2021 03:06 PM

Quote:

Originally Posted by bill dennison (Post 1338263155)

That videocardz article is brilliant TBH. So one guy out of 9bn people on the planet manages to procure a 5900XT engineering sample, an unreleased Rocket Lake S CPU and an unreleased Z590 board and even provides benchmarks :lol:

Really, come on. I think the Donald would be correct in calling this fake news. I haven't laughed so much for ages.

Shapeshifter Jan 26, 2021 09:52 AM

https://www.techpowerup.com/277717/a...arity-database

Quote:

AMD's mysterious Nashira Summit GPU has been spotted in Ashes of the Singularity database. A similarly named Nashira Point GPU has appeared some time ago on the USB-IF website, which was also a mysterious product in AMD's Radeon graphics processors lineup. The Nashira Summit and Nashira Point seem to be a part of the common Nashira GPU family, which is presumably a codename for a lower-end Navi 22 or Navi 23 GPU models. Today, we managed to get a Nashira Summit score in the Ashes of the Singularity database. The GPU has been put through a set of AotS benchmarks and we have the scores. Unfortunately, tests have been run using all-custom settings, so it is impossible to compare it to some other GPU as a reference. The test was probably performed by AMD or some AIB. So far it is impossible to distinct whatever this is a mobile or a desktop product as both mobile and desktop GPUs are tested in the same manner. It remains a question what the mysterious Nashira Summit GPU is, so we have to wait for more information to find out.
tested on a ryzen 3600x system, ashes thinks it's a dual GPU (so rdna3?)

pax Jan 26, 2021 11:01 AM

Dang if this thing is just a hair under the 6800 for a third the price I might grab it... then oc it for the hell of it.

12 gb should be good enough for 1440p anyways.

bill dennison Jan 26, 2021 11:57 AM

Quote:

Originally Posted by Shapeshifter (Post 1338263712)
https://www.techpowerup.com/277717/a...arity-database



tested on a ryzen 3600x system, ashes thinks it's a dual GPU (so rdna3?)

:yep:
2 chiplet RDNA 3 ES

wonder what 3 or 4 chiplet will do
7800 xt by Q4 this year

pax Jan 26, 2021 12:28 PM

Art says its navi 22-23...

If its navi 3 well thats a ways away then...

bill dennison Jan 26, 2021 12:43 PM

Quote:

Originally Posted by pax (Post 1338263765)
Art says its navi 22-23...

If its navi 3 well thats a ways away then...

Quote:

AMD Nashira Summit GPU Spotted, Possibly RDNA3 with a Multi-Chip Design
https://www.tomshardware.com/news/na...rdna2-or-rdna3

it could be a bug why it shows up as 2 GPU's

but why is it so close to a 6800 then
what would be the point of two sku's that close on the same gen

LordHawkwind Jan 26, 2021 02:37 PM

Quote:

Originally Posted by bill dennison (Post 1338263750)
:yep:
2 chiplet RDNA 3 ES

wonder what 3 or 4 chiplet will do
7800 xt by Q4 this year

When I saw that click bait story I thought oh no Bill is going to be trumping this like it's going out of fashion. :lol:

Most probably a 6700 or lower otherwise why bother testing it with a Ryzen 3600? Budget CPU really. If that was RDNA3 they would have used a Zen 3 5900/5950 CPU for sure.

Lisa Su has emphasised many times that they are bringing the 6xxx down the stack so it makes more sense for it to be a Navi 22-23 as Pax said.

bill dennison Jan 26, 2021 02:52 PM

Quote:

AMD Navi 31 rumored to be dual 80CU chiplet design, up to 10240 cores?
https://videocardz.com/newz/amd-navi...to-10240-cores

LordHawkwind Jan 26, 2021 03:15 PM

Quote:

Originally Posted by bill dennison (Post 1338263821)

Another click bait story. So you think Q4 2021 and the article says..

"AMD has never confirmed when exactly could RDNA 3 architecture be announced, but based on the roadmaps, AMD is expected to talk about the architecture by 2022."

Don't think they'll launch it Q4 2021 and then talk about it in 2022 :lol:

bill dennison Jan 26, 2021 03:31 PM

Quote:

Originally Posted by LordHawkwind (Post 1338263839)
Another click bait story. So you think Q4 2021 and the article says..

"AMD has never confirmed when exactly could RDNA 3 architecture be announced, but based on the roadmaps, AMD is expected to talk about the architecture by 2022."

Don't think they'll launch it Q4 2021 and then talk about it in 2022 :lol:

you are :nuts: .

AMD roadmaps say RDNA 3 will be out before 2022
not that they will talk about it

and that is the way they have always worked it in the past



..........
but since you bought a 3090 i'm sure they will wait till you are ready

LordHawkwind Jan 26, 2021 04:00 PM

Quote:

Originally Posted by bill dennison (Post 1338263847)
you are :nuts: .

AMD roadmaps say RDNA 3 will be out before 2022
not that they will talk about it

and that is the way they have always worked it in the past

..........
but since you bought a 3090 i'm sure they will wait till you are ready

Bill look at the image. A big tick for RDNA1 nothing for the other two. Wonder why that is? They haven't launched the 6800XT/6900XT yet properly or the 6700XT/6700 and below which are RDNA 2. Mid range cards are where the money is and Nvidia have already announced 3060 for late February launch.

So you think AMD will release RDNA 3 before completing the stack of RDNA2 cards don't think so.

As Cream said yesterday we're loving our 3090's in gaming. First time in a long time I've owned an enthusiast card. Surely you can understand that.

bill dennison Jan 26, 2021 04:29 PM

Quote:

Originally Posted by LordHawkwind (Post 1338263855)
Bill look at the image. A big tick for RDNA1 nothing for the other two. Wonder why that is? They haven't launched the 6800XT/6900XT yet properly or the 6700XT/6700 and below which are RDNA 2. Mid range cards are where the money is and Nvidia have already announced 3060 for late February launch.

So you think AMD will release RDNA 3 before completing the stack of RDNA2 cards don't think so.

As Cream said yesterday we're loving our 3090's in gaming. First time in a long time I've owned an enthusiast card. Surely you can understand that.

no
i never buy anything but the top gaming card at the time and used to buy two each of the top cards from both NV and AMD
been on 4k for 6 years now and 2560x1600 resolution before that back to 2007
it's a waist of time and money not going with the top gaming card at the time at high res

but the Titans and 3090 are not gaming cards and have always been nuts .
2x or more the price for 10% :nuts:

...

and AMD roadmaps are always like that and they are not late very damn often i think it will be Q4 or at most Q1

the 6900 does not OC well as well as the 6800 AND both missed the 4k mark that AMD said Big Navi would be
they are great at 1440p but run out of infinity cache at 4k .

and you don't think AMD knows that ?
and wants to fix it ASAP

option 1 = move on to RDNA 3 and a new 5nm node
and make as few 6000's as they can with the game consoles and CPU's on the same node

option 2 = keep making as many 6800 XT's as they can with the consoles and CPU's a slower 4k card than a 3080 with slower RT that few will buy over the 3080 for the same prices

they need a killer 4k enthusiast card with as good of RT as NV to win back market share and THIS ISN'T IT and I am sure they know it .

a MCM chiplet card before NV has one would be a killer

SuperGeil Jan 26, 2021 06:44 PM

There is no evidence to suggest that RDNA2 is running out of infinity cache or memory bandwidth. I think a couple of reviewers made some premature statements. Memory overclocking Big Navi gives negligible gains, even non-existent gains in some cases.

As for summit and point, I am completely loss. :sherlock:
Could be RDNA2 MCM. This would be most shocking.

SuperGeil Jan 26, 2021 07:08 PM

I'm also not convinced there is anything wrong with AMD's ray tracing. Virtually all the ray tracing supported games out right now except for two were made with Nvidia in mind. CP77 did not even bother coding for AMD hardware for ray tracing in the Xbox and PS are coming later. When more games come out for next gen platforms with ray tracing, I think we will get a much better look at what AMD can do as far as RDNA2 goes.

Nunz Jan 26, 2021 09:30 PM

Quote:

Originally Posted by SuperGeil (Post 1338263891)
There is no evidence to suggest that RDNA2 is running out of infinity cache or memory bandwidth. I think a couple of reviewers made some premature statements. Memory overclocking Big Navi gives negligible gains, even non-existent gains in some cases.

As for summit and point, I am completely loss. :sherlock:
Could be RDNA2 MCM. This would be most shocking.

Overclocking the memory won't fix an issue with the bus width not being wide enough to handle more throughput. The card is absolutely stuck due to the 256-bit bus. The infinity cache gimmick has done nothing thus far; if it worked as AMD claimed then there wouldn't be any bandwidth limitations.. but looking at how the 6900XT performs, it's definitely not. The card has all the horsepower, which is why it performs so well at 1080p/1440p, where that 256-bit bus isn't being overloaded.

Crank up to 4K and the card falls off dramatically. The only weak point of the entire card is it's memory bandwidth; specifically the bus width.

bill dennison Jan 27, 2021 01:18 AM

the only thing that would fix RDNA 2 for 4k is HBM2 at this point if they have a HBM version

Quote:

Rumor: AMD might be prepping new NAVI RDNA 2-based GPUs, “XTXH” and “Nashira Point” codenames spotted
https://www.dsogaming.com/news/rumor...names-spotted/

but that seem a lot of work on a RDNA 2 upgrade when they could as easy go to RDNA 3 that is most likely is MCM with HBM2e

Nunz Jan 27, 2021 05:26 AM

HBM is just another money sink that has nose-dived AMD before. They don't need it. GDDR6 puts out very good bandwidth, especially if they use Samsung memory chips instead of Hynix/Micron. If they had used a 384-bit bus, or maybe a 352-bit bus, we wouldn't be talking about HBM. Look at GDDR6X and the bandwidth it's putting out; I don't care for HBM at all.

HBM seemed genius when GDDR4 AND GDDR5 were slow as piss, but GDDR6 and 6X dramatically increased memory throughput and I think it shows that the memory bandwidth is not the limiting factor right now. Combine the raster performance of the 6900XT with a good sized bus instead of the gimped garbage AMD put on the card and it likely changes much of what we see in 4K performance.

bill dennison Jan 27, 2021 11:39 AM

Quote:

Originally Posted by Nunz (Post 1338263961)
HBM is just another money sink that has nose-dived AMD before. They don't need it. GDDR6 puts out very good bandwidth, especially if they use Samsung memory chips instead of Hynix/Micron. If they had used a 384-bit bus, or maybe a 352-bit bus, we wouldn't be talking about HBM. Look at GDDR6X and the bandwidth it's putting out; I don't care for HBM at all.

HBM seemed genius when GDDR4 AND GDDR5 were slow as piss, but GDDR6 and 6X dramatically increased memory throughput and I think it shows that the memory bandwidth is not the limiting factor right now. Combine the raster performance of the 6900XT with a good sized bus instead of the gimped garbage AMD put on the card and it likely changes much of what we see in 4K performance.

i think MCM chiplets will require HBM .

acroig Jan 27, 2021 11:47 AM

Quote:

Originally Posted by bill dennison (Post 1338264043)
i think MCM chiplets will require HBM .

How come?

bill dennison Jan 27, 2021 12:16 PM

Quote:

Originally Posted by acroig (Post 1338264046)
How come?

because no infinity cache will be big enough for MCM chiplets

i think they will need memory as close as posable and that means on die HBM, and where I think they have been going with HBM and infinity cache from the start steeping stones to MCM

think 2 or 4 or more chiplets all working on a frame at the same time off die memory will be too slow add too much latency

Nunz Jan 27, 2021 12:18 PM

I truly hope they make infinity cache something that works then, because right now it's useless.

Bill you're obsessed with MCM :lol:

bill dennison Jan 27, 2021 12:24 PM

Quote:

Originally Posted by Nunz (Post 1338264051)
I truly hope they make infinity cache something that works then, because right now it's useless.

Bill you're obsessed with MCM :lol:

yep

we all should be

we are at 5nm and GPU's chips are getting bigger every gen
how much longer before they run out of nodes that work ?

or GPU's that are so big they cost 2000+ bucks

Nunz Jan 27, 2021 01:17 PM

I don't think MCM will make costs go down. If there's one thing that's been a constant in the world, it's that things don't get cheaper. AMD/NV will just take the extra profit margin. Look at the costs of the 5000 series; AMD seen that Intel doesn't have a 12-core or 16-core direct comparison chip, so they jacked the prices up by a nice chunk. The only thing that keeps prices reasonable is competition, but if both companies jack it up, then we're all screwed.

I don't think it'll get cheaper. We've seen prices on the low and mid-range chips go up every year, to the point where people think the 3070 and 6800 are good values.

bill dennison Jan 27, 2021 01:52 PM

Quote:

Originally Posted by Nunz (Post 1338264073)
I don't think MCM will make costs go down. If there's one thing that's been a constant in the world, it's that things don't get cheaper. AMD/NV will just take the extra profit margin. Look at the costs of the 5000 series; AMD seen that Intel doesn't have a 12-core or 16-core direct comparison chip, so they jacked the prices up by a nice chunk. The only thing that keeps prices reasonable is competition, but if both companies jack it up, then we're all screwed.

I don't think it'll get cheaper. We've seen prices on the low and mid-range chips go up every year, to the point where people think the 3070 and 6800 are good values.

50 bucks :lol: is "jacked the prices up by a nice chunk"
16-core from 3950x for 749 to 799 for a 5950X


before Ryzen Intel was charging 1000+ for 4, 6 and 8 core CPU's
and if Intel had Ryzen instead of AMD that 5950X would be 1500 to 2000+


I think it will get a little cheaper but it will more slow the rise in prices

bill dennison Jan 27, 2021 02:11 PM

Quote:

AMD made a billion dollars more in Q4, and multiple billions in 2020
Quote:

All of that can translate to eye-popping dollar signs: the chipmaker just announced its Q4 and full-year 2020 earnings, and it’s adding billions of dollars wherever you look:

$3.2 billion in revenue this quarter, up 53 percent from $2.1 billion last quarter
$1.78 billion in profit this quarter, up 948 percent from $170 million last quarter*
$9.76 billion in revenue this past year, up 45 percent from $6.7 billion in 2019
$2.49 billion in profit this past year, up 630 percent from $341 million in 2019*
*Those profits include “a fourth quarter income tax benefit of $1.30 billion associated with a valuation allowance release,” according to the company.
https://www.theverge.com/2021/1/26/2...-earnings-2021

yea i think we will see RDNA 3 in 10 to 14 months
they got the cash to do it now and they need to get the jump on NV

LordHawkwind Jan 27, 2021 03:06 PM

Quote:

Originally Posted by bill dennison (Post 1338264086)
https://www.theverge.com/2021/1/26/2...-earnings-2021

yea i think we will see RDNA 3 in 10 to 14 months
they got the cash to do it now and they need to get the jump on NV

Bill this a comment from the article you posted.

And while AMD CEO Lisa Su did claim that the new Radeon 6000 cards had “launch quarter shipments three times larger than any prior AMD gaming GPU priced above $549,” that’s not much of a claim. The company’s Vega cards were also insanely hard to find, and to our knowledge AMD never previously sold a single gaming GPU above $550. The Radeon HD 7970 launched at $549, excluding it from AMD’s claim, and though the 6990 and 7990 cost $700 and $1,000 respectively, they were niche cards that had two GPUs each.

And before anyone shouts Radeon VII that was a failed Instinct card that only got launched because of the stupid prices of the RTX20 series. All the profit has come from console sales, EPYC ,Ryzen and mobile chips.

It's also interesting to note that the ASP on desktop processors dropped due to the mix of sales. Which translates to people like me who ended up buying a 3800X Zen 2 because I couldn't buy a 5800X. Also don't forget the fiasco surrounding Zen 3 not working on X470 M/B's and they needed a new bios which is only now being released.

If AMD want to keep making money all they need to do is release the 6700XT, 6700 and any mid-range card below because that is where the big sales and money are. They don't need an enthusiast 4K card. Lisa Su is a very intelligent business person and not a 4K enthusiast like you I'm afraid.

bill dennison Jan 27, 2021 04:53 PM

fact of life the top high end cards sell the midrange cards

if AMD can't come out with a winning high end card to beat NV their midrange won't sell well


flat out they need to beat the 3090 and smoke the 3080 soon or what ever NV's top cards are at the time
to get market buzz and sell the midrange money makers


take last time as AMD doesn't have 6000 midrange out yet

most people didn't say to themselves hay NV has the best fastest card in the 2080 ti but i can't afford that 1200 bucks monster so i'll buy a midrange RX 5700 that is slower than NV's last top card the 1080 ti
no they went for the 2060/70/80 that they could afford and why AMD lost a crap ton of market share

the masses may not buy the wining top card but the tend to buy the winning brand name of that top card .

and AMD needs to get there soon

SuperGeil Jan 27, 2021 07:29 PM

Quote:

Originally Posted by Nunz (Post 1338263925)
Overclocking the memory won't fix an issue with the bus width not being wide enough to handle more throughput.

If the buswidth is 128bit and the memory clock is 10ghz, it's still the same memory bandwidth as a 256bit bus with 5ghz memory, so I'm not sure what you mean by handling throughput better. If RDNA2 is bandwidth constrained, it would reflect it with near linear performance increase at 4k by overclocking the memory. But that is not what is happening. A 7% boost in memory clocks is fetching one or two frames per second.


The better way to look at it is that the 6800 is scaling better than the 3000 series at lower resolutions because of the ''garbage'' infinity cache. Not because it is memory bandwidth constrained at 4k, but because at resolutions below 4k, much less data has to leave the package. If the 3080 had a 128mb cache, it would probably show a similar delta between 1440p and 4k.

A deeper dive and investigation is needed. But there is no evidence to suggest RDNA2 is bandwidth bottlenecked.


All times are GMT -5. The time now is 02:37 AM.

Powered by vBulletin® Version 3.6.5
Copyright ©2000 - 2022, Jelsoft Enterprises Ltd.
All trademarks used are properties of their respective owners. Copyright ©1998-2011 Rage3D.com