Question Are there any 5-7nm GPUs in the 75 Watt range? Or, when will wee see them?

[amd6502]

Will we see a semi-affordable small die consumer GPU using an advanced node (5nm - 7nm )? And what sort of performance (relative to a RX-560) could we expect on such a card tuned to a factory 75 Watt TDP?

 

[Leeea]

yes:

NVIDIA Quadro RTX3000 MXM Module - M3T3000-QN | Embedded MXM | Industrial Graphics Card || Aetina Corporation

M3T3000-QN module features advanced NVIDIA Quadro graphics cards with Turing Architecture technology in the MXM 3.1 Type B form, bringing amazing visual computing performance, power-efficiency and real-time raytracing into every embedded system.

www.aetina.com

performance:

GeForce RTX 3000 laptop GPU benchmarks have started to appear

Phenomenal GPU power, itty bitty living space

www.techradar.com

They make mxm to PCI-E x16 adapters:

Eurocom MXM-to-PCIe x16 adapter lets you put a laptop GPU inside of a desktop PC

In what is essentially an eGPU docking station in reverse, the Eurocom Riser card allows users to install laptop MXM3 graphics cards directly onto their desktop PCIe slots. The product is aimed at professionals and workstation environments for specialized SFF PCs rather than a home office.

www.notebookcheck.net

 

[amd6502]

Thanks, though it looks like the RTX is not going to be a budget oriented GPU or small die. My best guess for the die size might be something around 170mm2.

I think something around 5 billion transistors and sub 100mm2 would hit the sweet spot for price/perf in this wattage category.

The riser card (notebook to desktop pcie) adapter is an intersting idea, but 'ouch' when i look at the prices.

 

[Insert_Nickname]

Will we see a semi-affordable small die consumer GPU using an advanced node (5nm - 7nm )? And what sort of performance (relative to a RX-560) could we expect on such a card tuned to a factory 75 Watt TDP?

This is as close as you can get:

https://www.amd.com/en/products/graphics/amd-radeon-rx-5300

You just can't buy one trough normal (r)etail channels.

 

[Leeea]

though it looks like the RTX is not going to be a budget oriented GPU or small die

AMD and Nvidia can make $$$ selling there top of the line units to seemingly infinite demand. Low margin budget parts? Not this year.

Budget is gone for the foreseeable future unless Intel steps in.

 

[Bouowmx]

Buying by marketing nanometers and die area. That's so strange.
No current product exists.

AMD Navi 24 "Beige Goby" entry-level GPU to offer 1024 cores - VideoCardz.com

AMD Navi 24 gets first confirmed specs AMD has come up with a fancy Linux name for its upcoming Navi 24 graphics processor. “Beige Goby” is the fourth codename for AMD RDNA2 GPU in Linux drivers. Appropriate patches have been issued recently by the AMD software team. The codename does to...

videocardz.com

 

[jpiniero]

Buying by marketing nanometers and die area. That's so strange.
No current product exists.

AMD Navi 24 "Beige Goby" entry-level GPU to offer 1024 cores - VideoCardz.com

AMD Navi 24 gets first confirmed specs AMD has come up with a fancy Linux name for its upcoming Navi 24 graphics processor. “Beige Goby” is the fourth codename for AMD RDNA2 GPU in Linux drivers. Appropriate patches have been issued recently by the AMD software team. The codename does to...

videocardz.com

That's gotta be OEM only. Maybe even mobile only.

 

[Magic Carpet]

This is as close as you can get:

https://www.amd.com/en/products/graphics/amd-radeon-rx-5300

You just can't buy one trough normal (r)etail channels.

3 gigs of vram @ ~100w tdp? My ~30w GT640 had 2Gb in ~2011. Now that was a great low power/htpc/cuda card. This amd card only makes sense in a pcie 4.0 slot, I suspect the vram deficit won’t be felt as much.

Best low power card today is likely the 1650 and even that comes with 4 gigs of vram and can be fed entirely off the pcie slot. But yeah, it’s out of the ops range being on a larger node design. But nano meters isn’t everything.

The last great low power (~22w TDP; 130nm (!!!)) AMD/ATI I had was the one-slot Sapphire 9600XT Ultimate with a passive heatsink. Now that was something, especially compared to the loud (such cooling design was the norm) and relatively power hungry competition at the time.

Spoiler

*Posts like these were very common on the forums:

FX5950 from XFX sounds like a vacuum cleaner...HELP

NB. It’s funny how the definition of “low power” varies throughout our computing history. To think, the top GeForce 5950 Ultra had only 75w TDP, now we need to look hard to find a speedy modern graphics card of the same wattage.

 

[amd6502]

Well considering that it looks roughly like a halved navi 23 (which is ~240mm2) 130mm2 sounds like a good guess. We are getting closer to the 100mm2 budget barrier.

Until that, I hope they can produce more RX-560. Someone said it might be difficult because of the older GDDR it needs, but I'm hoping for the best.

 

[Leeea]

Until that, I hope they can produce more RX-560. Someone said it might be difficult because of the older GDDR it needs, but I'm hoping for the best.

That sort of thing is just not worth the labor, time, or BoM.

Maybe they could get the chips made. But then they would be nerfing the supply of VRMs etc for their high end products. High end products which sell all day long over MSRP. Why would any manufacturer of graphics cards make a rx560 when they could be making a "oem version" rx6900 and then sell it way over AMDs MSRP?
https://www.newegg.com/p/pl?d=rx6900&Order=2&N=100007709&isdeptsrh=1
Why would ASUS, Sapphire, Power color, etc spend the time making a card that sells for <$200 when you can spend the same time making a card that sells 3x over MSRP and pocket $2000 profit?
Why would AMD reprint a rx560 they can sell for maybe $20 when they can turn out rx6900s they sell to the OEMs for several hundred $. Those old nodes still cost money to run, GF is having ones of its best years ever.


Both companies are going to put most of their energy into cranking out rtx3090s or rx6900s.

(personally, I was hoping AMD would continue production of the rx5000 series, but it was not to be ).

 

[aigomorla]

Why would AMD reprint a rx560 they can sell for maybe $20 when they can turn out rx6900s they sell to the OEMs for several hundred $. Those old nodes still cost money to run, GF is having ones of its best years ever.

because the majority of gamers aren't ballers who can afford a gpu even @ over 500 dollars.
This is why the under 500 dollar segment was good and healthy for gamers.

Now you can't even get a 1660ti/super for under 500 dollars, so most of the cards aren't even going to gamers, but miners and scalpers.

And no AMD won't only produce 1000 dollar videocards.
Thats a quick way to kill your self, as soon people will eventually dry up and those 1000 dollar videocards will just be sitting on the shelf collecting dust, unless AMD wants to take a hugh loss and price those 1000 dollar videocards at sub 500 dollars.

 

[Leeea]

Your only hope* is . . . Intel!:

Intel Xe-HPG DG2 GPU Spotted With Performance On Par With A NVIDIA GTX 1660 SUPER

A brand new Intel Xe-HPG DG2 GPU entry has made its way in GFXBench which shows performance on par with an NVIDIA GeForce GTX 1660 SUPER.

wccftech.com

However, with a 275 watt TDP, not exactly what you were looking for.

*hope, as in wccftech is a rumor mill of the first order, not even remotely reliable

 

[GodisanAtheist]

Buying by marketing nanometers and die area. That's so strange.
No current product exists.

AMD Navi 24 "Beige Goby" entry-level GPU to offer 1024 cores - VideoCardz.com

AMD Navi 24 gets first confirmed specs AMD has come up with a fancy Linux name for its upcoming Navi 24 graphics processor. “Beige Goby” is the fourth codename for AMD RDNA2 GPU in Linux drivers. Appropriate patches have been issued recently by the AMD software team. The codename does to...

videocardz.com

-Amd's card breakdown this gen has been so ****ing weird I cannot make sense of it.

80/40/32/16 CU versions?

Why not 80/60/40/20 with a natural 72/52/32/12 cut down part at each step? You end up with 4 dies each way anyhow...

 

[aigomorla]

i dont think well see a 5-7nm low wattage cpu for a while.

They still make 710's and it hasn't been replaced by a 1010 or a 2010, nor do i hear anything of a 3010.
We saw a 1030 but those were not very plentiful as people grab'd a 1050 with the very small price disparity, and vendors have even gotten desperate to ressurect the 730.

Meaning No....

I think were gonna be stuck on the 700 series for low wattage gpu's until you wait for Intels XPe.

But then APU's are getting more and more powerful so we probably wont even need them by then.

 

[Insert_Nickname]

But then APU's are getting more and more powerful so we probably wont even need them by then.

There is still a market for a low power card to upgrade older machines with newer video decoders. 3D performance isn't everything.

A small >100mm2 5-7nm with AV1 decode would be a really decent option there. The small process still allow you to pack a lot of transistors in such a package.

 

[Magic Carpet]

Yeah, they don’t want to bother with lower margin parts at like ALL. Maybe with the 4000 series we get to enjoy a <=40 watt SKU. Too bad Apple doesn’t want to enter the PC GPU market. We definitely could use more competition.

2) Also, I am really not a fan of higher performance SKUs with 3 slot coolers in order to quietly tame those 300w+ monsters. I like very everything in 3090, except the fact that it likes to heat my room on a hotter day, a bit too much.



This is the case.

 

[Insert_Nickname]

Maybe with the 4000 series we get to enjoy a <=40 watt SKU.

I suppose such a card sort of exist. Just not in consumer format.

https://www.pny.com/nvidia-t600?iscommercial=true

See also T400, T1000

 

[Magic Carpet]

@Insert_Nickname

Great find, everything's great except that tiny fan. That brings me the memories of my old Geforce 2 MX and how much I hated the noise of it, lmao. I ended up putting a giant heatsink on it just to save my hearing. I am pretty sure, it's not as bad today as it once was noise-wise, though. I'd still prefer single 90-100mm fans on 30-50w cards, personally.




It's not low profile, but otherwise still a great design:

 

[Insert_Nickname]

Great find, everything's great except that tiny fan.

I -would- prefer something like the cooler on my ASUS low-profile half-height GTX1650, but it is low-profile and single slot after all. Can't have everything it would seem.

That brings me the memories of my old Geforce 2 MX and how much I hated the noise of it, lmao. I ended up putting a giant heatsink on it just to save my hearing.

I was kind of lucky, my Geforce2 MX400 didn't have a fan. ASUS V7100PRO 64MB:

https://classiccomputershop.eu/Webw...-out-DX7-graphics-AGP-4x-PC-card-adapter.html

I didn't even have problems with it running hot.

 

[amd6502]

-Amd's card breakdown this gen has been so ****ing weird I cannot make sense of it.

80/40/32/16 CU versions?

Why not 80/60/40/20 with a natural 72/52/32/12 cut down part at each step? You end up with 4 dies each way anyhow...

I agree it is confusing; the naming and marketing is so confusing I've actually given up on even trying to keep track with AMD naming/marketing.

I really like the idea of a 16 CU die. It would be a major step up from the monolithic 8CU iGPUs in the flagship APUs. Secondly, you still can get 12 CU cut down units that will be a significant step up from the 7-8 CU iGPUs, especially if you pump about 70 watts through them.

 

[bigboxes]

Yeah, they don’t want to bother with lower margin parts at like ALL. Maybe with the 4000 series we get to enjoy a <=40 watt SKU. Too bad Apple doesn’t want to enter the PC GPU market. We definitely could use more competition.

2) Also, I am really not a fan of higher performance SKUs with 3 slot coolers in order to quietly tame those 300w+ monsters. I like very everything in 3090, except the fact that it likes to heat my room on a hotter day, a bit too much.

View attachment 48366

This is the case.

Very cool. What is exactly in that machine?

 

[aigomorla]

This is probably the fastest your going get for low wattage.
But its a Quadro... meaning its horrible when it comes to games, but will do everything else fine and then some, especially if you need a NVENC card.

NVIDIA Quadro P2200 Review

NVIDIA Quadro P2200 Graphics Card review with benchmark scores. See how it compares with other popular models.

benchmarks.ul.com

That card does not require any aux 6-8 pin.
If we could just find some hacked drivers to make it play its Gforce counter part, it could probably be "acceptable" for 1080p @60hz.

 

[Abwx]

If one can live with the 6/8 pin connector then it should be quite easy to be within 75W with a downclocked 6600XT, although that s a high cost solution, but 30-30% lower frequency should do it and would still provide a good amount of Tflops.

 

[fleshconsumed]

I'd like a card like that for my Plex server. I could buy a higher wattage card, but I don't really want to deal with additional PCIe power cables. I could also buy Quadro, but newer RTX cards have better NVENC engines. I guess I'll just have to wait till ETH goes PoS and video card situation normalizes, should only take another year or two.

 

[Insert_Nickname]

I could also buy Quadro, but newer RTX cards have better NVENC engines.

Have you considered this?

https://www.pny.com/nvidia-t600

You could go even lower to the T400, but I don't think the savings justify it.