RDNA 3 Rumor Discussion

[LordHawkwind]

Originally Posted by pax

Yeah thats what I mean in getting an oled 43-48" 8k tv as a monitor. Or maybe a qled not sure yet with that screen burn thing. I plan an 8k mon upgrade in a year or so.

Pax, as I've said to Bill, to run that monitor you'll need to spend $2,000 on a next gen card yet you've mentioned more than once you won't pay that price for a graphics card. Why the sudden change of mind?

[bill dennison]

Originally Posted by LordHawkwind

Bill that TV/Monitor costs $2,799 and you'd need a next gen card costing about $2,000 to get your 60fps so really cheap at $4,799. I bet it will be real fun sitting 2 feet from a 65" TV/Monitor

i'm setting 3 feet from a 4k LG OLED55CXPUA that was like around 1800+ with 5 year warranty
and it is fun

I can see 65" at the same 3 feet

and we don't know what the 7900 xt & 7800 xt will cost yet
but if it is as fast as they say they both would do 8k

[pax]

Originally Posted by LordHawkwind

Pax, as I've said to Bill, to run that monitor you'll need to spend $2,000 on a next gen card yet you've mentioned more than once you won't pay that price for a graphics card. Why the sudden change of mind?

I dont expect top end to be 2k+ next year. TSMC says many vendors are hoarding chips. Eventually that tactic to keep prices high will be counterproductive.

https://hothardware.com/news/tsmc-cu...cial-shortages

There are many factors that play into the chip shortage that continues to affect the market for all kinds of various electronics, from graphics cards and game consoles, to certain smart automobile features and everything in between. Some of the reasons include rabid demand for cutting edge hardware, manufacturing challenges spurred by the pandemic, and cryptocurrency mining (as it relates to GPUs). According to TSMC, you can also add chip hoarding to the pile.

TSMC is too inbred to the industry to make this kind of accusation lightly.

[acroig]

https://wccftech.com/amds-flagship-n...een-taped-out/

[bill dennison]

Originally Posted by acroig

https://wccftech.com/amds-flagship-n...een-taped-out/

so 6 to 9 months maybe

4th of july would be cool

[acroig]

Originally Posted by bill dennison

so 6 to 9 months maybe

4th of july would be cool

If they can get it out in volume a full quarter before nV's AL...... they'd clean up.

[bill dennison]

Originally Posted by acroig

If they can get it out in volume a full quarter before nV's AL...... they'd clean up.

if RDNA 3 is half as fast as they say they will do that anyway
as i exect Linda to be the same 30% to 45% max over Amperage anyway

and if it is really 3x faster than a 6900 xt NV is in trouble this time


but AMD needs to follow NVidia's lead in the LHR cards or gamers won't ever see them

[pax]

Cool thats its taped out means the rumors on specs will become a bit more solid soon now...

[Nagorak]

I guess it's all rumors anyway, but if it's already taped out shouldn't we be expecting it sooner than the end of next year? It seems a lot of sites are saying it's coming Q4 of next year, but does it actually take a year from tape out to production?

[bill dennison]

Originally Posted by Nagorak

I guess it's all rumors anyway, but if it's already taped out shouldn't we be expecting it sooner than the end of next year? It seems a lot of sites are saying it's coming Q4 of next year, but does it actually take a year from tape out to production?

used to be about 6 months if all goes well

so i think from May on if they want to and have supply

but it could take a few more months to stock up on cards to sell
being chiplets should help their less waste than a big monolithic GPU

depends on what Intel's cards do Q1 and the 3080 super and 3080 ti super soon after

AMD could well release early Q3 if they can beat the 3090 super by a good bit

[LordHawkwind]

This is from Lisa Su in the latest earnings call..

Yeah. So, again, what I would say is our market share is still, I would say underrepresented, whether you're talking about the client CPU or APU side or the GPU side. I think what we have seen here in the third quarter and then into the second half of the year is, our graphics business has performed quite well. It is channel driven in the sense that there's still strong demand amongst gamers for GPUs. As we go into 2022 though, I don't view the PC market as a headwind for the Company. I think as we look at all of these markets, of course, we do a bunch of scenario planning if the market is up or if it's down.

I think there are many who think that the market maybe up, there's some who think that market maybe down, and that's why we're choosing to model the base case is flattish. But even within that market, whether you're talking about client CPUs or client GPUs, we think we have opportunities to gain share and grow in that business. Just given the strength of our product portfolio and the fact that we are underrepresented to think the what we can expect given those products.

She also mentioned consoles and CPU's were a major driver for 2022 so as usual the GPU side is lagging behind. Don't blame them it's business but don't think they'll prioritise GPU's for a few generations yet. Don't get me wrong, like Bill says, RDNA 3 might knock it out of the park but it just won't be easily available. Just my 2c'c worth

[bill dennison]

AMD's next-gen RDNA GPUs could have 3D Infinity Cache technology

AMD's next-gen RDNA GPU architectures teased with 3D Infinity Cache technology, Navi 31 MCM GPU could have 512MB Infinity Cache

https://www.tweaktown.com/news/82628...ogy/index.html

[pax]

We kind of are expecting the amount of L3 to be increased up to 512 megs but they would use Infinity Fabric to run it? I suppose its not the case with the on die cache of the 6000 gen even tho they call it Infinity cache?

https://www.trustedreviews.com/news/...-cache-4128555

I think the interesting part is that if they do a 3d cache and remove the on die Inf Cache it would make more room available on the gpu die and make it possible to have that monstrous increase in shaders that have been rumored up to 15000~ vs the top end 5120 now on the 6900xt.

[bill dennison]

Originally Posted by pax

We kind of are expecting the amount of L3 to be increased up to 512 megs but they would use Infinity Fabric to run it? I suppose its not the case with the on die cache of the 6000 gen even tho they call it Infinity cache?

https://www.trustedreviews.com/news/...-cache-4128555

I think the interesting part is that if they do a 3d cache and remove the on die Inf Cache it would make more room available on the gpu die and make it possible to have that monstrous increase in shaders that have been rumored up to 15000~ vs the top end 5120 now on the 6900xt.

now add 24gb of HBM3

https://www.anandtech.com/show/17022...t-up-to-64gbps

and have a 8k card

[pax]

Godless monster and a price to go with it... and power useage... latest rumor is rtx 4800 is gonna double up on power use vs ampere so sounds like both will push the power envelop to the hilt trying to out perf the other.

Can they cool 600+ watts tho. Water would have to be the reference cooler.

[bill dennison]

Originally Posted by pax

Godless monster and a price to go with it... and power useage... latest rumor is rtx 4800 is gonna double up on power use vs ampere so sounds like both will push the power envelop to the hilt trying to out perf the other.

Can they cool 600+ watts tho. Water would have to be the reference cooler.

two or three fan AIO

[Nagorak]

Hopefully these power consumption rumors turn out to be wrong. Whether people care about power consumption or not, at the current rate power is increasing we will quickly approach the point where it's simply not viable to draw more power.

A 15 amp 120v circuit maxes out at 1800w, and usually more than one outlet on a circuit. At 1000w power draw you're already at the point where having two computers on the same circuit will trip the breaker. Much about 1000w and you're starting to run into problems if anything else high powered is on that circuit with your computer, or if you have crappy wiring, etc.

And then there's the cooling issues...

[Nunz]

Originally Posted by bill dennison

two or three fan AIO

360mm AIO can barely handle 350-400watt cards. Talking about 500+.. no chance.

[acroig]

Originally Posted by Nagorak

Hopefully these power consumption rumors turn out to be wrong. Whether people care about power consumption or not, at the current rate power is increasing we will quickly approach the point where it's simply not viable to draw more power.

A 15 amp 120v circuit maxes out at 1800w, and usually more than one outlet on a circuit. At 1000w power draw you're already at the point where having two computers on the same circuit will trip the breaker. Much about 1000w and you're starting to run into problems if anything else high powered is on that circuit with your computer, or if you have crappy wiring, etc.

And then there's the cooling issues...

This. Totally crazy.

[AllexxisF1]

Just got my electric permit from the county pre-approved for the home theater. Good thing I'm putting in three new 15amp circuits.

[bill dennison]

Originally Posted by Nagorak

Hopefully these power consumption rumors turn out to be wrong. Whether people care about power consumption or not, at the current rate power is increasing we will quickly approach the point where it's simply not viable to draw more power.

A 15 amp 120v circuit maxes out at 1800w, and usually more than one outlet on a circuit. At 1000w power draw you're already at the point where having two computers on the same circuit will trip the breaker. Much about 1000w and you're starting to run into problems if anything else high powered is on that circuit with your computer, or if you have crappy wiring, etc.

And then there's the cooling issues...

i ran two 20 amp circuits on 10 ga wire to my system a long time ago

[LordHawkwind]

Originally Posted by bill dennison

i ran two 20 amp circuits on 10 ga wire to my system a long time ago

TBH BIll I have no idea what you are talking about. Me and electrics don't mix From a non techie perspective if it takes 2x power to improve performance then to me that is a massive fail from both sides. Aren't things meant to get smaller and more power efficient or have I missed something?

If a graphics card needs to come with a water block to cool it (not AIO) it's a massive statement of failure. It makes enthusiast gaming a niche of a niche and marginalises its appeal greatly. TBH count me out I'll stick with what I've got thank you.

Also if the 4090/7090 are this power hungry what about the next gen and so on. This will kill PC gaming far quicker than consoles will.

[bill dennison]

Serving up big performance
Despite all of the trouble that comes with creating MCM designs, the payoffs are huge. Having multiple chiplets working together as a unified GPU means huge performance boosts and a much more efficient solution than bridged graphics cards.

Aside from massively improving performance with the power of two interlinked chiplets, AMD would also have the option to optimize each chiplet for performance or efficiency. This would help AMD either send performance to the moon at the cost of huge power draws, as rumors already suggest about 500W models, or moderately increase performance by setting more efficient clock frequencies on each chip.

https://www.pcinvasion.com/amd-rdna-...n-performance/


might not be so bad if they scale it to need somehow

lighter loads lower clock frequencies or even turn one chiplet off

then crank it up only when needed

[bill dennison]

Originally Posted by LordHawkwind

TBH BIll I have no idea what you are talking about. Me and electrics don't mix From a non techie perspective if it takes 2x power to improve performance then to me that is a massive fail from both sides. Aren't things meant to get smaller and more power efficient or have I missed something?

If a graphics card needs to come with a water block to cool it (not AIO) it's a massive statement of failure. It makes enthusiast gaming a niche of a niche and marginalises its appeal greatly. TBH count me out I'll stick with what I've got thank you.

Also if the 4090/7090 are this power hungry what about the next gen and so on. This will kill PC gaming far quicker than consoles will.

means i'm set up for up to 3 way SLI/CFX on both of two systems in one case running at the same time

i don't think even these new high watage GPU's will surpass 3 way sli 1080 ti that i bult it for when i had 980 sli and fury x cfx W/ a 1200 and 1050 watt PSU
i didn't know at the time they would kill SLI/CFX

the plug changes on PSU's will be a pain in the ass wonder if AMD will use NV's plug

[Nunz]

Originally Posted by bill dennison

https://www.pcinvasion.com/amd-rdna-...n-performance/


might not be so bad if they scale it to need somehow

lighter loads lower clock frequencies or even turn one chiplet off

then crank it up only when needed

sounds like a recipe for major framepacing issues and high latency... I don't think you want your GPU to do things like this.

[bill dennison]

Originally Posted by Nunz

sounds like a recipe for major framepacing issues and high latency... I don't think you want your GPU to do things like this.

i'm sure they could step it down for things like diablo 3 and up for 2077

[pax]

Gonna have to ask cavemanjim to get Ice Giant to make a gpu thermosiphon at this point...

[bill dennison]

Originally Posted by pax

Gonna have to ask cavemanjim to get Ice Giant to make a gpu thermosiphon at this point...

I don't think a 5 or 6 slot air cooler will do 500+ watts

the highest end cards are going to take a full waterblock or a 3 fan AIO cooler with a thick rad like a Alphacool NexXxoS Monsta


https://www.aquatuning.us/water-cool...360mm-radiator

or maybe a 420mm
https://www.aquatuning.us/water-cool...420mm-radiator


my 560mm
https://www.aquatuning.us/water-cool...560mm-radiator

cooled two gtx 980's and the cpu fine for two weeks without fans when i forgot to plug them back in once

[Nunz]

Originally Posted by bill dennison

i'm sure they could step it down for things like diablo 3 and up for 2077

I mean, cards already do this though. downclocking and downvolting for minimal loads..

Turning chips on and off is a much different thing though and it's impossible to do it without incurring latency or reducing performance. They still haven't figured out how to do that on CPUs yet, which is why disabling C-States and running an all-core OC is still the best way for lowest latency/input lag.

Originally Posted by pax

Gonna have to ask cavemanjim to get Ice Giant to make a gpu thermosiphon at this point...

Such an overrated cooler that flopped lol.. all that hype during the marketing was BS. Turned out to perform worse than a 360mm AIO, and lose in most cases to a Noctua D15. Terrible for the size and noise.

[pax]

Ice Giant is pretty much one of the best coolers for threadripper tho and has a high watts cooling capacity. I think the copper version will do a lot better on smaller packages.

I mean if the all aluminium model can handle 500w easy the copper one might be able to go a bit further.

https://www.icegiantcooling.com/blog