Discussion RDNA4 + CDNA3 Architectures Thread

[DisEnchantment]

With the GFX940 patches in full swing since first week of March, it is looking like MI300 is not far in the distant future!
Usually AMD takes around 3Qs to get the support in LLVM and amdgpu. Lately, since RDNA2 the window they push to add support for new devices is much reduced to prevent leaks.
But looking at the flurry of code in LLVM, it is a lot of commits. Maybe because US Govt is starting to prepare the SW environment for El Capitan (Maybe to avoid slow bring up situation like Frontier for example)

See here for the GFX940 specific commits

History for llvm/lib/Target/AMDGPU - llvm/llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies. - History for llvm/lib/Target/AMDGPU - llvm/llvm-project

github.com

Or Phoronix

More AMD "GFX940" Enablement Work Landing In LLVM - Phoronix

www.phoronix.com

There is a lot more if you know whom to follow in LLVM review chains (before getting merged to github), but I am not going to link AMD employees.

I am starting to think MI300 will launch around the same time like Hopper probably only a couple of months later!
Although I believe Hopper had problems not having a host CPU capable of doing PCIe 5 in the very near future therefore it might have gotten pushed back a bit until SPR and Genoa arrives later in 2022.
If PVC slips again I believe MI300 could launch before it

This is nuts, MI100/200/300 cadence is impressive.



Previous thread on CDNA2 and RDNA3 here

Question - Speculation: RDNA3 + CDNA2 Architectures Thread

Man I have been dying to make this one for a while now. First rumours for RDNA3 are here so new thread time! Just going to start off with this one for now: kopite7kimi on Twitter: "@VideoCardz Ah, I mean a simple mcm design with 10240 cores is not enough. Because the lift from RDNA2 to RDNA3...

forums.anandtech.com

 

[GTracing]

A 650w psu is all good if you have 7700X set to eco mode right?

I think they rate the PSU ratings against a 7950X

I wouldn't pair a 9070 XT with a 650W unless the PSU was a really nice one. They suggest higher wattage PSUs for a reason. Transient spikes can be way higher than the TDP of the GPU. The RTX 3000 series was where this issue first got some attention.

 

[gaav87]

Wonder how high thous 340W tbp boards will go if that small gigabyte went 3250mhz 56C in furmark xD
Easy 3.5?

 

[MrTeal]

There s no current balancing as such, in a 2 x 6 pin connector the 6 pins that handle the positive rail are connected in parralel in the PCB, and the same for the 6 of the negative rail, if the currents are unbalanced it just mean that there s some pins of the card that are not in tight contact with the cable connector pins, hence most of the current flow in the pins that are connected the tightest.

Do you have a picture of the Nitro+ PCB showing that the 6 pins are commoned after the connector?

 

[SolidQ]

I'm looking at this seems Steam stats are broken for now +4% and -5% for one month?

Was that in 2024, and next month is fixed

Wonder how high thous 340W tbp boards will go if that small gigabyte went 3250mhz 56C in furmark xD
Easy 3.5?

we will see in 2.5 days

 

[Abwx]

Do you have a picture of the Nitro+ PCB showing that the 6 pins are commoned after the connector?

I dont see how it could be otherwise for redundancy purposes and cost, controling each cable current would increase cost too much and wouldnt solve anything since
the problem for Nvidia FI is that the connectors have a loosy grip on the cards pins.

Only thing that could be done at very low cost is probing the currents by measuring the voltage drops between the PCB individual pins and the commoned track, this allow to know if your connector has the required contact torque for each pin.

 

[igor_kavinski]

World-first reveal of my idea to fix graphics card power delivery. Please contact me if you want to implement this solution and want to share the profits with me out of sheer honesty.

An actual dual slot card (needs two PCIe x16 slots right next to each other).

Card pulls 75W from each slot. Total 150W.

Both parts of the card have exactly one 8-pin PCIe connector. That's 300W.

So now we are up to 450W just like that.

For additional power, we have USB-C ports on the port cluster of each part of the card towards the outside. Let's assume just one USB-C port for the time being, each with ability to provide 140W. That's 280W. Not impossible to easily have two more ports.

So we have 75W+75W+150W+150W+140W+140W=730W for a relatively "normal card".

And for factory overclocked or halo cards, we can have above plus two 8-pin connectors and two more USB-C connectors for an additional 580W of power which equals total 1310W power.

See? Not an engineer but I have the future ALL FIGURED OUT.

Was that hard? NO.

Makes me wonder what these "engineers" smoke on their breaks.

 

[soresu]

World-first reveal of my idea to fix graphics card power delivery. Please contact me if you want to implement this solution and want to share the profits with me out of sheer honesty.

An actual dual slot card (needs two PCIe x16 slots right next to each other).

Card pulls 75W from each slot. Total 150W.

Both parts of the card have exactly one 8-pin PCIe connector. That's 300W.

So now we are up to 450W just like that.

For additional power, we have USB-C ports on the port cluster of each part of the card towards the outside. Let's assume just one USB-C port for the time being, each with ability to provide 140W. That's 280W. Not impossible to easily have two more ports.

So we have 75W+75W+150W+150W+140W+140W=730W for a relatively "normal card".

And for factory overclocked or halo cards, we can have above plus two 8-pin connectors and two more USB-C connectors for an additional 580W of power which equals total 1310W power.

See? Not an engineer but I have the future ALL FIGURED OUT.

Was that hard? NO.

Makes me wonder what these "engineers" smoke on their breaks.

Or get ATX/PCIe backing to standardise this:

Asus GPU power connector delivers 1,000W for cableless builds — GC-HPWR has a retractable design

BTF 2.0 branded GPUs will come with this new power connector

www.tomshardware.com

 

[igor_kavinski]

Asus GPU power connector delivers 1,000W for cableless builds — GC-HPWR has a retractable design

BTF 2.0 branded GPUs will come with this new power connector

www.tomshardware.com

That would be best if ASUS makes it royalty-free. Unfortunately, Intel/Nvidia suffer the most from "not invented here" syndrome so not much hope of that happening. These two idiot companies will prefer to double down on their stupid connector and try to make it more fireproof.

 

[soresu]

That would be best if ASUS makes it royalty-free. Unfortunately, Intel/Nvidia suffer the most from "not invented here" syndrome so not much hope of that happening. These two idiot companies will prefer to double down on their stupid connector and try to make it more fireproof.

At this point the well is pretty poisoned on the 16 pin connector for a lot of people.

It's better just to abandon it and try something else.

Plus GC_HPWR has the added bonus of spreading the weight of the gfx card further across the mobo - a thing that ATX really should have addressed more than a decade ago.

 

[soresu]

Another thing that could simplify maintenance problems is to engineer the x16 slot release mechanism so that it is activated from the under side of the mobo instead.

I've no idea how that would be achieved, but it would take up vastly less space than the solutions I've seen so far.

Hell, just making the release catch a bit sturdier and wider would go a very long way to making it remain usable once the gfx card is covering it.

 

[PJVol]

Curious about what hypothetical 96cu board connectors would look like, 4-5 8-pin? )

 

[Hail The Brain Slug]

Curious about what hypothetical 96cu board connectors would look like, 4-5 8-pin? )

3x8 pin would suffice. It offers 516W max according to spec, and wider parts are generally run lower and slower for better efficiency on the V/F curve.

 

[RnR_au]

A spreadsheet with slot width - https://docs.google.com/spreadsheets/d/18eQRucHX41A-O4OsoV96Qw2gFw1Qs2N7f6qQQs3kXx4/edit?usp=sharing

edit: not mine.

 

[PJVol]

Does anyone know what Fmax is set on XT gpu?

 

[adroc_thurston]

Curious about what hypothetical 96cu board connectors would look like, 4-5 8-pin? )

2*8pin since it would be a 350W GPU or therein.

 

[PJVol]

it would be a 350W GPU or therein

+15% power for 50% more compute units? Come on)
Only if severely reduced boost clocks for the large sku, like 2500mhz or even lower.

 

[adroc_thurston]

+15% power for 50% more compute units? Come on)

A wee bit of v/f dial back and there you go.
N31 already shipped 96 shader cores@355W.
N48 chop is 56@220W anyhow.

 

[PJVol]

A wee bit of v/f dial back and there you go.
N31 already shipped 96 shader cores@355W.
N48 chop is 56@220W anyhow.

Ok, it would probably fit 350W at sth like 2300-2500 targeting +30% performance, a hair above 4090 but still far behind 5090.
Btw, any guess on Fmax for the n48?

 

[adroc_thurston]

Btw, any guess on Fmax for the n48?

I don't even know.
Low 3's I guess?

 

[In2Photos]

I guess AMD plans on selling more than 3 units.

https://twitter.com/x/status/1896424499400307150

 

[marees]

I guess AMD plans on selling more than 3 units.

https://twitter.com/x/status/1896424499400307150

MLID said the stock levels will resemble 7800xt/7700xt (that was also a delayed launch like now)

 

[Win2012R2]

MLID said the stock levels will resemble 7800xt/7700xt (that was also a delayed launch like now)

Demand is likely to be much higher

 

[uzzi38]

Does anyone know what Fmax is set on XT gpu?

You mean the frequency limit with OC?

It's high enough that you'll run into the power limit long before you hit the clock limit.

 

[gaav87]

You mean the frequency limit with OC?

It's high enough that you'll run into the power limit long before you hit the clock limit.

Even on 340TBP boards ? that is 391W

 

[SolidQ]

Interesting what is AIB with those high memory temps
P.S I don't think it's with Honeywell PTM7950