Friday, November 4th 2022

AMD's Navi 31 Might Clock to 3 GHz, Partner Cards Will be Able to Overclock

Based on details from a PCWorld livestream following AMD's launch of the Radeon RX 7000-series, it was revealed that AMD has designed the Navi 31 GPU to be able to scale as high as 3 GHz. In other words, it appears that AMD has power limited its cards, at least for the SKUs that the company has announced so far. This could be for many reasons, but most likely to try to find a balance between power and performance. The details of the 3 GHz scaling did however not come from AMD directly, but rather from Jarred Walton over at Tom's Hardware. That said, the information was apparently shared with the media by AMD at the event.

In the livestream, it was also confirmed that partner cards will be able to overclock, so expect to see some factory overclocked cards, with higher power draw. This could be why, in part, that ASUS went with a much larger cooler on its TUF Gaming Radeon RX 7900-series cards. As ASUS didn't reveal any clock speeds or TDPs of its two cards, we don't really know what to expect, but we'd be surprised if these cards weren't factory overclocked to some degree when they launch in December.
Sources: PCWorld (on YouTube), via @akoago (on Twitter)
Add your own comment

51 Comments on AMD's Navi 31 Might Clock to 3 GHz, Partner Cards Will be Able to Overclock

#26
spnidel
AnotherReaderDidn't RDNA2 have better clock headroom than Ampere?
yep, as an example my 6800 xt pulse can oc from stock 2.31ghz to 2.58ghz with a slight undervolt so power draw remains lower than 300w
Posted on Reply
#27
Guwapo77
Vayra86WRONG
Wikipedia:Its a complete misconception that premium somehow indicates better, it never did. Premium sells an emotion.

Premium is literally overpaying for the same shit with perceived increased value 'because its premium'. Its empty. Its marketing. And it proves itself time and time again. Yet somehow people still want to feel like they get 'the premium treatment' and that's exactly why it exists.

People buying premium simply have more money than sense. 'Premium SMS'. Explain :) You get more characters? And what about an FE 4090 versus a 'premium AIB card'...? ;) You get more wattage... lol. Same GPU, same perf, same transistor count, but you burn moar watts doing so. The premium appears on your energy bill :D
I could buy a Ford Focus to get to A to B or I could buy a premium BMW M5 Competition (emotion version) to do exactly the same thing. I don't know man, that premium sure is nice though. And you ain't lyin' about premium appearing on your energy bill either. LoL
Posted on Reply
#28
revanoctis
spnidelyep, as an example my 6800 xt pulse can oc from stock 2.31ghz to 2.58ghz with a slight undervolt so power draw remains lower than 300w
How much better performance in games did that give you?
Posted on Reply
#29
spnidel
revanoctisHow much better performance in games did that give you?
core oc alone gets me 9% more perf, coupled with a vram overclock it's around 12% more performance compared to stock
mind you that's at 4k, at 1440p I wouldn't be surprised if the core oc gave a slightly bigger boost
Posted on Reply
#30
mb194dc
6800xt is a decent overclocker. Not to the level of unlocked bios 5600xt though. Have both.

Interested to see what custom Navi 31 designs will do. Probably huge diminishing efficency returns over 400w or so but we'll see.
Posted on Reply
#31
Minus Infinity
MLiD reported AIB cards will be able to go to 500W. Initial cards will be reference design though. Honestly the pricing is amazing in this day and age and AMD is still getting good margins. Ngreedia is looking p!ss poor value. I have a feeling the 7900XT will be my next card, but waiting for a MSI custom version. I just hope FSR3 is ready early(ish) 2023. Looking to update ancient 1700X system, and thinking 13700K but prepared to wait for 7900X3D reviews. 7900X3D and 7900XT has a nice symmetry just need DDR5 7900 to complete the trifecta. :)
Posted on Reply
#32
Jism
john_only come with two 8 pin connectors to limit overclockability.
You can easily extract far more power out of a single 8 pin cable then what PCI-E standard dictates.

www.techpowerup.com/review/amd-r9-295-x2/22.html



The peak power on only 2 x PCI-E 8 pin cables was 646 watts, is roughly 53 Amps.

Yes wires will get warm, but generally if it is a good build PSU with quality cables you can easily do this.

I have never seen such a thing that a card with 3 vs 2 8 pin cables suddenly overclocks worse. As long as input voltage supplied to your GPU's VRM is stable your good. And thats why (lots) of capacitors exists on graphics card anyway.
Posted on Reply
#33
Patriot
fancuckerThey're pioneers and leaders in temporal upscaling and ray tracing technologies, and it's safe to assume they have the rasterization crown. They deserve their market dominance. People cry for some miraculous AMD victory but 80 percent of sales across the performance segments will be Nvidia. People want a premium, reliable product and that's where team green delivers.
Yes, the burn your house down premium feel.
Posted on Reply
#34
wolf
Better Than Native
AnotherReaderDidn't RDNA2 have better clock headroom than Ampere?
It did indeed, but I was just talking RDNA2 VS RDNA3
Posted on Reply
#35
AnotherReader
wolfIt did indeed, but I was just talking RDNA2 VS RDNA3
We will see in December, but I suspect that RDNA3 will overclock and scale better than Ada. The 4090 doesn't scale well with clock speed. A 10% overclock only increased performance by 4.5% in theFounder Edition review at TPU. Compared to RDNA2, it might be slightly worse as the default TDP has increased so there's less margin for increased power. We also don't know if the MCD is clocked 1:1 to the GCD's shader clock. If it stays static while the shader clock increases, then overclocking won't be as impressive as RDNA2.
Posted on Reply
#36
Jism
AnotherReaderWe will see in December, but I suspect that RDNA3 will overclock and scale better than Ada. The 4090 doesn't scale well with clock speed. A 10% overclock only increased performance by 4.5% in theFounder Edition review at TPU. Compared to RDNA2, it might be slightly worse as the default TDP has increased so there's less margin for increased power. We also don't know if the MCD is clocked 1:1 to the GCD's shader clock. If it stays static while the shader clock increases, then overclocking won't be as impressive as RDNA2.
I suspect that some vital parts will be locked or limited in regards how and what you can overclock. Just as the initial RDNA and it's Infinity cache which was locked. You'd say a minor increase would yield huge bandwidth advantage. But we dont have the tools nor utilities to customize, test these ourself.

Oc'ing is dead anyway. GPU's pretty much clock to what it is capable of. The best you can get is a undervolt and lower consumption, with perhaps a little bit of performance. But thats about it. Many years ago i installed a vapochill on a Ati Radeon 2600 Pro card. The core clock ran at 600Mhz stock and i was running at somewhere around 1300Mhz with a 2.1V core voltage and -40 degrees temps. It did blew up during the 3Dmark nature run and to be honest you wont be getting these "fun" clocks anymore compared to these generation(s) of hardware.

You buy a card that matches to what you want and thats it.
Posted on Reply
#38
Zach_01
TheLostSwedeSomeone on Twitter did some further extrapolations, which may or many not be true.



If we accept that the 1.5x comes at 2.3GHz (GCD) then 2.7GHz is +17% clock. (I expect the same % clock increase in cache and a portion of this on GDDR6)
355W + 17% = 415W and that is only on the same voltages. If more volts then the +17% can easily be +25~30% in wattage (=440~460W)

1.5x + 17% = 1.75x ...and this is with linear increase for clock/wattage vs perf.

I think its too optimistic to say 2.0x

Maybe if we accept the 1.5~1.7x, then... + 17% = 1.75~2.0x (up to 2.0x) and again with linear increase of power/performance.

Could be or not... depends on the headroom of the efficiency curve, but I would assume that AMD already took advantage of it.

Personally I expect it to be trading blows with 4090 at the 430~450W region in rasterization and the same with 3090/Ti in RT.
Posted on Reply
#39
Zareek
TheLostSwedeWatch the video. Nothing confusing about it, he shared something he wasn't supposed to say, yet.
100% it was slip, hopefully he doesn't pay for it in someway.
Posted on Reply
#40
john_
Zareek100% it was slip, hopefully he doesn't pay for it in someway.
Probably he wouldn't, because there is already this out, so if this is not fake, it's not a secret anymore.
Posted on Reply
#41
crazyeyesreaper
Not a Moderator
Will be interesting to see how performance stacks up in all tests.

AMD is quoting a 60% RT performance increase with a maximum of 80%.

So in something like Control with no Ray Tracing we are seeing the 6950XT at 56.9 FPS at 4K Ultra and 23.6 with RT ON

Taking the above into account using the 6950XT as reference

3090Ti vs 4090 vs 7900XTX
Control: 40 vs 68 vs 40
Cyberpunk 2077: 24 vs 41 vs 21
Deathloop: 66 vs 121 vs 74
F1 2022: 49 vs 85 vs 54
Far Cry 6: 77 vs 96 vs 126
Metro Exodus: 71 vs 116 vs 78
Resident Evil Village: 102 vs 175 vs 142
Watch Dogs Legion: 48 vs 75 vs 64

Obviously not a perfect setup as RT and Rasterization performance will likely scale differently but if the 7900XTX is 1.6x faster in both rasterization and ray tracing a direct 1.6X to 1.7X best case would result in performance similar to the above.

Seems to me. This is how the product stacks will likely compare.
RTX 4090 = _________
__________ = 7900XTX
RTX 4080 = 7900XT
RTX 4070 = 7800
RTX 4060 = 7800
RTX 4060 = 7700
RTX 4050 = 7700
Posted on Reply
#42
64K
crazyeyesreaperWill be interesting to see how performance stacks up in all tests.

AMD is quoting a 60% RT performance increase with a maximum of 80%.

So in something like Control with no Ray Tracing we are seeing the 6950XT at 56.9 FPS at 4K Ultra and 23.6 with RT ON

Taking the above into account using the 6950XT as reference

3090Ti vs 4090 vs 7900XTX
Control: 40 vs 68 vs 40
Cyberpunk 2077: 24 vs 41 vs 21
Deathloop: 66 vs 121 vs 74
F1 2022: 49 vs 85 vs 54
Far Cry 6: 77 vs 96 vs 126
Metro Exodus: 71 vs 116 vs 78
Resident Evil Village: 102 vs 175 vs 142
Watch Dogs Legion: 48 vs 75 vs 64

Obviously not a perfect setup as RT and Rasterization performance will likely scale differently but if the 7900XTX is 1.6x faster in both rasterization and ray tracing a direct 1.6X to 1.7X best case would result in performance similar to the above.

Seems to me. This is how the product stacks will likely compare.
RTX 4090 = _________
__________ = 7900XTX
RTX 4080 = 7900XT
RTX 4070 = 7800
RTX 4060 = 7800
RTX 4060 = 7700
RTX 4050 = 7700
My guess is that Nvidia will come out with a RTX 4080 Ti at some point to compete directly with the RX 7900XTX but that's just speculation. I don't think they will count on the RTX 4090 to compete with the RX 7900XTX because it's so expensive.
Posted on Reply
#43
H3llB0und
spnidelyep, as an example my 6800 xt pulse can oc from stock 2.31ghz to 2.58ghz with a slight undervolt so power draw remains lower than 300w
Similar, but with a 6800XT Nitro+ SE.
Can easily get it to boost above 2.65GHz without ever going over 300W and keeping my temps under 80°
Posted on Reply
#44
ARF
RTX 4090 Ti ~~ RX 7900 XTX 24 GB
RTX 4090 ~~ RX 7900 XT 20 GB
RTX 4080 Ti ~~ RX 7900 / RX 7800 XT
RTX 4080 ~~ RX 7800 XT 16 GB
RTX 4070 Ti ~~ RX 7800
RTX 4070 ~~ RX 7700 XT
RTX 4060 Ti ~~ RX 7700
RTX 4060 ~~ RX 7600 XT
etc.
Posted on Reply
#45
Lucas_
fancuckerThey're pioneers and leaders in temporal upscaling and ray tracing technologies, and it's safe to assume they have the rasterization crown. They deserve their market dominance. People cry for some miraculous AMD victory but 80 percent of sales across the performance segments will be Nvidia. People want a premium, reliable product and that's where team green delivers.
that what you think, I m all about the reliability but I m not sure that they do care that much its just thier R&D has that much of load of $$ to be able to do that.
i dont wanna open a discussion about what what, but cause alot ppl have this understanding and mentality, we are now paying 2k for card ! keep doing so bro ! we need alot from you.
Posted on Reply
#46
Max(IT)
Now I’m just waiting for AIC prices before considering to buy a 7900XTX.
I really hope Nvidia cards will stay on the shelf forever.
64KMy guess is that Nvidia will come out with a RTX 4080 Ti at some point to compete directly with the RX 7900XTX but that's just speculation. I don't think they will count on the RTX 4090 to compete with the RX 7900XTX because it's so expensive.
Because RTX 4080 is cheap, right ? Putting that into perspective, the 4080 is even worse than 4090, price wise.
Posted on Reply
#47
spnidel
john_Probably he wouldn't, because there is already this out, so if this is not fake, it's not a secret anymore.
fuuuuuck, stop sharing good stuff like this! you're gonna force me to upgrade AGAIN!--oh wait, there are no games to play... nevermind
Posted on Reply
#48
Denver
Lucas_that what you think, I m all about the reliability but I m not sure that they do care that much its just thier R&D has that much of load of $$ to be able to do that.
i dont wanna open a discussion about what what, but cause alot ppl have this understanding and mentality, we are now paying 2k for card ! keep doing so bro ! we need alot from you.
I think we can all agree that Nvidia is already as good as Apple at zombifying consumers and overturning any judgment they have.

"Just buy this one, it has our logo, so it's the best product in the universe"

Nvidia launches $1600 GPU that fries the cables and there are people talking about reliability. Plus, If you're talking about software, I invite you to see the dozens of bugs that both AMD and Nvidia clearly announce with every driver release.
Posted on Reply
#49
Minus Infinity
RT increase is only disappointment to me. 60-80% is frankly useless, and RDNA3 will heavily rely on FSR 3.0. 100% increase would have been the bare minimum required. I guess until we see RDNA4 with it's largely new architecture AMD will still trail badly in RT. Still won't stop me buying 7900XT most likely.
Posted on Reply
#50
john_
64KMy guess is that Nvidia will come out with a RTX 4080 Ti at some point to compete directly with the RX 7900XTX but that's just speculation. I don't think they will count on the RTX 4090 to compete with the RX 7900XTX because it's so expensive.
Considering how much slower RTX 4080 is expected to be compared to RTX 4090, it's obviously a gap there for the RTX 4080 Ti. I believe who ever pays $1200 for the RTX 4080 will be a fool. I believe Nvidia postponing the RTX 4080 12GB, was a reaction to AMD's pricing for the RX 7900XT/X cards. So, I believe they will come out probably with an RTX 4080 16GB and an RTX 4080 Ti ??GB instead. Now I don't know when that RTX 4080 Ti will come out, maybe December 13th(?), but my guess is that it will be costing $1200-$1400 and the RTX 4800 16GB will drop to $1000-$1100 to directly compete with RX 7900 XTX. Even if RTX 4080 16GB loses in rasterization to RX 7900XTX, even if it doesn't win in RT(probably it will, but let's say it doesn't), we all know by now that people will still buy it, because of the Nvidia logo. Nvidia also knows it, so I guess an RTX 4800 at $1000-$1100 will happen soon.
Posted on Reply
Add your own comment
Nov 13th, 2026 23:00 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts