• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

AMD Radeon RX 6950 XT Reference Design

W1zzard

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
28,753 (3.74/day)
System Specs
Processor Ryzen 7 5700X
Memory 48 GB
Video Card(s) RTX 4080
Storage 2x HDD RAID 1, 3x M.2 NVMe
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 10 64-bit
The AMD Radeon RX 6950 XT Reference Design card not only looks great, but also achieves excellent performance, matching the GeForce RTX 3090, but at much better pricing. Because of good optimization and surprisingly good energy efficiency, AMD's thermal solution works very well, too.

Show full review
 
I actually expected worse pwoer consumption, kinda impressive in some ways.
Still, these prices are too crazy to consider any of these cards (both AMD and Nvidia).
 
Nice review as always.

I see v-sync can an obviously make a big difference on power consumption.

Out of curiosity, if there a power draw difference running ray tracing vs. not running ray tracing?
 
Will testing the reference design change the current relative percentage value in the TPU GPU database? Currently the database shows only 3% over the 6900xt.
 
Wow I looked up AMD's website and us Canadians are not being served properly. My cousin lives in Boston and was telling me that he could everything from CPUs to GPUs at MSRP. In Canada the only listings that are available are the 6750XT and the 6950XT.
 
I don't know about 3090Ti difference, it seems correct though, the odd number is the 3090 efficiency, it seems wrong, it must be close to 3080Ti, @W1zzard should check it out:
IMG_20220519_012749.png
 
Last edited by a moderator:
We can draw the power/perf chart for Navi 21 between 6900 and the various custom 6950 designs.
 
3090 is just -8% on average at 4K and -5% at QHD vs 3090Ti and the TDP is 350W vs 450W (355W vs 445W measured or whatever), doesn't find it alarming that 3090Ti is 3% more power efficient than 3090?
 
ModEl4 said:
I don't know about 3090Ti difference, it seems correct though, the odd number is the 3090 efficiency, it seems wrong, it must be close to 3080Ti, @W1zzard should check it out:
Click to expand...
I mean, they also keep reposting the 600W+ power spike number for the 6900 XT, which was clearly an erroneous measurement.
 
TBH I hope AMD could untie the artificial power limit in the Driver.
There are clearly many headroom for 6800 series for overclocking
The only thing in between is just the +15% power limit set by AMD
 
Crackong said:
TBH I hope AMD could untie the artificial power limit in the Driver.
There are clearly many headroom for 6800 series for overclocking
The only thing in between is just the +15% power limit set by AMD
Click to expand...
This is easy enough bypass with MPT, but it doesn't really give you much without exotic cooling. Most benchmark world record holders with Navi 21 cards struggle to break 2800 MHz, which is achievable within the power limits with the right card.
 
Really impressed with this chart with what AMD did on RX 6950 XT vs the RX6900 XT and the thing of it is a AIB could improve that further if they saw fit to do so. It would be interesting if a AIB kept a good cooler on it like Sapphire utilizes and then also dials down the power limit to like 325w a bit below reference if they can do that on good chips reliably and stably at the same time they would have something quite nice on energy efficiency. These power consumption 20ms spikes can get really extreme and power hungry.

Something I wouldn't mind seeing on GPU's is a bypass fan header on the fan cooling power so that the MB or fan controller could power it instead. It's fine if the GPU still regulates the RPM's, but if they can offload the actual fan power being generated by the GPU itself in such way it would be better for the PCB and VRM temps. I haven't noticed a design like that yet, but seems very much within the realm of possibilities and offloading it externally would free up the power limit a bit. The fans themselves and at higher RPM's can consume a chunk of the power limit if I'm not mistaken.

1652925056624.png
 
kapone32 said:
Wow I looked up AMD's website and us Canadians are not being served properly. My cousin lives in Boston and was telling me that he could everything from CPUs to GPUs at MSRP. In Canada the only listings that are available are the 6750XT and the 6950XT.
Click to expand...
This has always been the case even back when it was Ati Radeons built in canada. The US always get it cheaper than we do and their dollar is worth abit more so its a win win for them.
 
Last edited:
Daven said:
Will testing the reference design change the current relative percentage value in the TPU GPU database? Currently the database shows only 3% over the 6900xt.
Click to expand...
Yup, that's part of the reason why I was so interested in getting my hands on the reference card. The GPU Database performance numbers for RX 6950 XT have been updated.

Oberon said:
I mean, they also keep reposting the 600W+ power spike number for the 6900 XT, which was clearly an erroneous measurement.
Click to expand...
Just woke up and retested this for you on the RX 6900 XT
https://imgur.com/oyWZKbi
(ignore the multi-monitor numbers, I was too lazy to plug in an out the 2nd cable)
 
if you want to talk about the "6950XT being more efficient than" then you must go beyond addition and subtraction and use multiplication and division too, for now you must assume that 3090ti is at 100% efficiency scale in the chart and then multiply that 100% by the ratio between the previous 6950XT 100% and the previous 3090ti 81%, so

100 * ( 100 / 81 ) = 123,46%

so

the 6950XT is 23.46% more energy efficient than the 3090ti

---
Ref Table from the article:

energy-efficiency.png
 
Last edited by a moderator:
It is very nice to see all the tested AIB cards together in the chart with the ref card. The only thing I am missing is the XTXH card.
 
ARF said:
The "multi-monitor" power consumption is terrible. AMD's engineers failed to find a way NOT to overclock the memory chips so heavily during this function:
Click to expand...
Yeah, this is pretty normal for AMD, has been like that for like a decade
 
Jism said:
@Wizzard, why is it you guys dont recieve directly from AMD anymore?
Click to expand...
For this launch AMD has only coordinated the launch with board partner cards. They haven't seeded their own reference designs, to anyone

I have Ryzen 5600 and 5700X here for review from AMD
 
W1zzard said:
They haven't seeded their own reference designs, to anyone
Click to expand...

Then, if so, you can safely assume that the fastest representation of the card (MSI Gaming X Trio) should be included in TPU performance table, rather than the slowest which virtually no one would be interested in to purchase, anyways.
 
ARF said:
which virtually no one would be interested in to purchase, anyways.
Click to expand...
Huh? They are not sending out samples so they can sell the cards
 
W1zzard said:
Huh? They are not sending out samples so they can sell the cards
Click to expand...

Really? How low can AMD fall? I mean they have to send no more than 5 cards for the whole world for the largest reviewers. And then, ask you and the other four reviewers to return the cards, if the supply is so pathetic.
 
You must log in or register to reply here.
Back
Top