Extensive test reveals AMD’s Radeon RX 7900 XTX draws 150 W more on average compared to the Nvidia RTX 4080

[Redaktion]

Despite both Nvidia's and AMD's new GPU generations being built on the TSMC 4 nm nodes, the power efficiency crown surprisingly goes to Team Green this time around. AMD has been working on driver fixes since December last year, but the abnormal power issues are still there.

https://www.notebookcheck.net/Extensive-test-reveals-AMD-s-Radeon-RX-7900-XTX-draws-150-W-more-on-average-compared-to-the-Nvidia-RTX-4080.733657.0.html

[Hey]

Is this with the latest Amd 23.7.1 driver released a few days ago that was said to tackle the power draw issue?

[Michael keggereis]

AMD does not use 4nm process.  It uses a combination of 5nm and 6nm.  They don't suck that bad at efficiency.  Along with those choices a manufacturing processes they use a chiplet design which in turn allows them to make a more cost efficient GPU

[nerodia]

The issue is weird, and I haven't experienced it with Mesa (vk_radv) 23.1 or 23.2 (git) on my 7900xtx. Power usage has been within expected ranges, idling at ~15w for 2 1440p monitors, ~22w for video decode acceleration (Youtube/VLC), and gaming varying widely based on usage as expected. It's a shame, I have no real complaints about the hardware but the drivers seem to be making it a poor choice currently for Windows users.
For the record, the Windows driver issues with 7900xtx VR gaming are also nonexistent with Mesa. Aside from the slowly improving but still not quite there AI support through ROCm it's been a very painless experience.

[S.Yu]

AMD does not use 4nm process.  It uses a combination of 5nm and 6nm.  They don't suck that bad at efficiency.  Along with those choices a manufacturing processes they use a chiplet design which in turn allows them to make a more cost efficient GPU

True...and it says so on this very site. They'd better be cheap because a mix of 5nm/6nm is vastly cheaper than 4nm.

[Halps]

Ofc the 4090 gets bashed due to a flaw in just a minute fraction of all units sold and most was due to human error. Doesn't discount the fact that the 4090 is the best card by far out of the current offerings. Manual undervolting allows the card to really stretch it's performance. I'm getting only 10mhz less clock speed on a flat .975 undervolt after 2790mhz.

Literally doubles my 3080 rasterization for nearly the same power draw and 20c+ cooler. Oh right, but 80 out of 800,000 units melted, so we need to mention that...

[Russel]

Ofc the 4090 gets bashed due to a flaw in just a minute fraction of all units sold and most was due to human error. Doesn't discount the fact that the 4090 is the best card by far out of the current offerings. Manual undervolting allows the card to really stretch it's performance. I'm getting only 10mhz less clock speed on a flat .975 undervolt after 2790mhz.

Literally doubles my 3080 rasterization for nearly the same power draw and 20c+ cooler. Oh right, but 80 out of 800,000 units melted, so we need to mention that...

Well....
Compare the prices of 4090 and 3080 as well.
A generational leap like that is something that we used to take for granted 'WITHOUT' tripling the price...

When you buy something for such a hefty premium, you'd at least expect them to get rid of qc issues completely.

[Bogdan Solca]

AMD does not use 4nm process.  It uses a combination of 5nm and 6nm.  They don't suck that bad at efficiency.  Along with those choices a manufacturing processes they use a chiplet design which in turn allows them to make a more cost efficient GPU

My bad, 4 nm is used on the mid-range models and iGPUs.

[Glegster]

Remember nvidia released comprehensive compression advantages to keep using the 28nm in the 900 series. Not like it rocket science,but  you have to remember that. And the compression techniques they use.
Where amd is more raw power.
If raw efficiency of amd could meet nvidia's compression techniques it be a power house

[Bogdan Solca]

Is this with the latest Amd 23.7.1 driver released a few days ago that was said to tackle the power draw issue?

From what I'm reading in the comments of that very video, it looks like the new driver does not fix the power draws in full loads.

[Matt is so cool]

I don't regret returning my defective 7900 xtx and buying a 4080 for $80 more at all.  And really dropped the ball this generation, which is a shame because 6000 was a great series.

[Kon]

Well that is certainly unfortunate but not unpredictable with the XTX, V8 will use more gas than a 4cyl engine even though they are both going 60kph. All that extra memory doesn't come free, the main issue for us Australians is 4080 is 300-400$ more than the XTX which makes the decision harder,

[g]

i have a xfx 7900xtx and it's more power efficient than my previous 3080 for the same output with a couple of optimizations.
firstly i run a 7600x with an igpu, be sure to plug gp to the motherboard instead of the gpu. during idle the 7900xtx consumes barely 8w.
during gaming i undervolt and use amd chill to limit fps to 60 which is my monitor's max anyway.
in apex legends my gpu consumes 150w which is slightly lower than my previous 3080.

happy with my 7900xtx which is $250 cheaper than a 4080.

[shanny]

Literally doubles my 3080 rasterization for nearly the same power draw and 20c+ cooler. Oh right, but 80 out of 800,000 units melted, so we need to mention that...

gpus of that power range are fire starters, melting cables shortly after release is very bad and needs to be fixed right away, the fact of the matter is there needed to be a media shitstorm to force Jensen and his Nvidiots to at LEAST have a good flagship, but anyway you look again since the 20, series we are seeing a product stack lower on our base models, with a melting top performer this would make every single gpu launch they had become controversial,

>4090, bad price, bad cable, HEAT
>4080, no generational leap in price/perf, only lower tdp
>4070ti, a mistake that was unlaunched, does not justify its existence
>4070, what the 4060 should have been, meh value
>4060ti, 100 bucks for RAM ONLY......
>4060, what the 4050 should have been, tdp lower than the 3050 just to support my point

unless the 4050 is a passively cooled GOD tier card to replace the 1630 I've no clue how Nvidia can have any single hope to fix the pr nightmare