AMD RX 7000 series GPU Owners' Club

[AusWolf]

Oh I'd bet on finewine happening. I'm just not sure it's worth all this. 6700XT and 6750XT pretty much finewine'd their way past the 3070 and 3070 Ti in a bunch of games, but I never cared about that - I just wanted to get away from 8GB of VRAM.

On a single monitor, with a clean Driver Only install, card is now clocking as it should in SE5 (270-310W at 2.5GHz+). Previously with two monitors it was stuck a stuttering mess at 180W at 1900MHz. Performance probably still has room to improve, it's only outperforming 3070 Ti native by a little bit.

FSR simply doesn't work, perf equal to or worse than native. Also Vsync does not work in Vulkan, whereas on 3070 Ti both work fine. I know it's a FSR 1.0 implementation but the image quality at Ultra FSR on the 3070 Ti was stunning - I couldn't believe it wasn't DLSS. By comparison Ultra FSR on the 7900XT is a blurry trash heap

Anecdotally, reddit has blamed everything under the sun on cheap uncertified DP cables. The vapor chamber problem, burning cables, stuttering, crashing, bad perf.........we shall see tomorrow.

It's weird that Nvidia has been selling revisions of the same architecture under different names for 3 generations now, while AMD has been doing the exact opposite - selling 3 different architectures under the same name. I just wish they got them right. After the buggy RDNA 1, we had the awesome RDNA 2, and now it's back to bugs again with RDNA 3, it seems.

The weirdest thing is that we have people living on this planet who believe that progress is always linear, change is always positive and the ideal launch date of any new product is yesterday.

 

[kapone32]

It's weird that Nvidia has been selling revisions of the same architecture under different names for 3 generations now, while AMD has been doing the exact opposite - selling 3 different architectures under the same name. I just wish they got them right. After the buggy RDNA 1, we had the awesome RDNA 2, and now it's back to bugs again with RDNA 3, it seems.

The weirdest thing is that we have people living on this planet who believe that progress is always linear, change is always positive and the ideal launch date of any new product is yesterday.

This is the first run of Chiplet based Gpus. This is very different than anything that has come before it. As they get there head around optimizing it we could see tremendous improvements in performance.

 

[MrPerforations]

Alphacool™ water block is here, day and a half late, but its here.
it is very heavy, alarmingly so.

 

[tabascosauz]

Anyone in Canada looking at the XT, here's the first new sub-$1200 deal I've seen pop up after last week's MBA and MERC at $1150:



Might only last about a day, knowing Amazon. Powercolor seems to be leading the charge on pricing

 

[kapone32]

Alphacool™ water block is here, day and a half late, but its here.
it is very heavy, alarmingly so.

That means the block is quality. Tell me does the water area look like a pool or is it channelled?

 

[MrPerforations]

micro fin center but more like a pool than a channel. didn't really look and now its in the machine.
bad news is the memory junction temp and the hot spot temps didn't change much, save about 5c in that area.
i assume it just that hot.
the package temp is down in the 41c.

 

[tabascosauz]

AMD Software: Adrenalin Edition 23.1.1 for AMD Radeon™ RX 7900 Series Graphics Release Notes | AMD

New 23.1.1 driver, anyone on it yet? From the release notes:

Known issues

• High idle power has situationally been observed when using select high resolution and high refresh rate displays.
• Video stuttering or performance drop may be observed during gameplay plus video playback with some extended display configurations.

So, in the past 3 weeks, AMD fixed absolutely NOTHING

I just tried switching the second monitor over to HDMI 2.0 and disabling Freesync, card still completely incapable of handling a second monitor on 22.12.2. Gonna try 23.1.1 now

The funny thing is that while I was gathering Genshin metrics today with CapframeX, multi monitor and single monitor repeatedly come out the exact same on frametimes. Frametimes and FPS, almost identical data. Yet, it's blatantly obvious to anyone with a set of eyes that the multi monitor setup is CONSTANTLY stuttering. Maybe it's a Freesync problem that shit just hits the fan when there's a second monitor connected? As my vertical monitor, the S2721DGF already has its Freesync permanently disabled just to try and minimize any compatibility issues.

 

[Super Firm Tofu]

So, in the past 3 weeks, AMD fixed absolutely NOTHING

You must be new here. Still waiting on Adrenalin drivers to support AMD CPU monitoring of AM5 (since September FFS).

 

[tabascosauz]

You must be new here. Still waiting on Adrenalin drivers to support AMD CPU monitoring of AM5 (since September FFS).

You reckon I should just return it and get a reference 6950XT? $100 less, not a great deal, but at least there are no surprises.

 

[Super Firm Tofu]

You reckon I should just return it and get a reference 6950XT? $100 less, not a great deal, but at least there are no surprises.

Well, I had a Red Devil 7900 XT ready to be ordered until I read your feedback so far on the 7000 series. Holding off for now.

As far as the 6950XT, if you think your power supply is up to the transients of RDNA2, it's probably a less frustrating option.

*edit* - I forgot that the 6950XT resolved the huge 20ms spikes that the 6900XT has.

 

[AusWolf]

You reckon I should just return it and get a reference 6950XT? $100 less, not a great deal, but at least there are no surprises.

Last gen is always a safer option than the newest shiny and untested thing, imo. There are some sweet discounts on it too, unlike any new release.

 

[tabascosauz]

Last gen is always a safer option than the newest shiny and untested thing, imo. There are some sweet discounts on it too, unlike any new release.

I'm not sure about discounts. 6800, 6800XT, 6900XT have fully disappeared. 6950XT is all that's left and it's $1060, vs. the $1150 of the 7900XT.

I'm now testing the S2721DGF with Freesync on, at 60Hz, in landscape orientation. If Sniper Elite 5 doesn't shit the bed due to the S2721DGF's presence, I think I might be able to get around Genshin's problems by moving it to the Dell screen where it's 60Hz.

If.

 

[AusWolf]

I'm not sure about discounts. 6800, 6800XT, 6900XT have fully disappeared. 6950XT is all that's left and it's $1060, vs. the $1150 of the 7900XT.

That's a shame. Anyway, my point stands: last gen is always a safer option. New gen should only be bought by people who don't mind beta testing or putting up with teething issues.

 

[tabascosauz]

No change to the fan stop curve in 23.1.1. Not touching Radeon Software with a 10 foot pole anymore so I guess I'll live with the long wind down time. That shitty program crashed 3 times on me in the span of 1 hour yesterday. No bug report, no dialog box, no error logged, no nothing, just decided to leave.

Genshin works on the S2721DGF. Buttery smooth with just Windowed VRR ticked in Windows. Except the ridiculous 60Hz input lag, but I'll live. Still choppy af if I Win+Shift+arrowkeys the game onto the M32Q, I guess that's the beauty of purely software-based VRR

And Sniper Elite still works the same as it did yesterday on one monitor. Maybe Radeon just isn't prepared to accept vertical monitors?

Back to the testing suite.

That's a shame. Anyway, my point stands: last gen is always a safer option. New gen should only be bought by people who don't mind beta testing or putting up with teething issues.

Though, I wonder if that is actually reasonable to accept. Not qualified enough to comment on Radeon but at this point we should be paid for the investigative work that we do for the AGESA team. Outside of AMD launch day seems to still mean "the product is ready", not "congratulations, you ARE the engineering sample"

 

[NicklasAPJ]

Just got my 7900 XTX Red DEVIL limited, and is a great card, buuut the OC side is a bit werid? need more TDP for this card, the card can "run" at 3100-3200Mhz core but will lose alot of fps instead of 2900/3000 cause of hitting TDP limit.

Will make a video about it soon!?

 

[kapone32]

So I installed the latest driver and it did run the Stress test properly (for 10 seconds) which is better as last time it was 5 seconds. The idle power draw has also gone down and now the temp in Adrenline are about 3 degrees cooler but the Memory is still at 58 C in HWinfo.It is the only temp that is higher than 40C. I am really interested to see what adding a Waterblock will do to the memory temp.

 

[tabascosauz]

So I installed the latest driver and it did run the Stress test properly (for 10 seconds) which is better as last time it was 5 seconds. The idle power draw has also gone down and now the temp in Adrenline are about 3 degrees cooler but the Memory is still at 58 C in HWinfo.It is the only temp that is higher than 40C. I am really interested to see what adding a Waterblock will do to the memory temp.

I think there might still be room for power optimization on the new 20Gbps GDDR6 on these cards. At lowest possible idle power (15-20W), core is pretty much completely gated at 0.0W, yet memory still floats so many watts. Similar to chiplet Ryzen (cores parked) but also not (I doubt MCD are responsible for the power draw, have to double check the metrics).

At idle mine is also between 50-60C. No problems at load, around 82-88C memory, but I can imagine that the overclockers will need to get creative to cool this new generation of GDDR6 (just like GDDR6X). I saw someone mod the G6X on the TUF 3070 Ti/3080 with copper heatspreaders.

G6X on my last card ran a bit cooler (40ish idle, 80C load), but the cooler was also physically bigger, and density was half (1GB packages).

 

[kapone32]

I think there might still be room for power optimization on the new 20Gbps GDDR6 on these cards. At lowest possible idle power (15-20W), core is pretty much completely gated at 0.0W, yet memory still floats so many watts. Similar to chiplet Ryzen (cores parked) but also not (I doubt MCD are responsible for the power draw, have to double check the metrics).

At idle mine is also between 50-60C. No problems at load, around 82-88C memory, but I can imagine that the overclockers will need to get creative to cool this new generation of GDDR6 (just like GDDR6X). I saw someone mod the G6X on the TUF 3070 Ti/3080 with copper heatspreaders.

G6X on my last card ran a bit cooler (40ish idle, 80C load), but the cooler was also physically bigger, and density was half (1GB packages).

Agreed, I see the memory as being the culprit for the 110 C problem as well. I don't see the core but that top rank of 8 GB of Memory. could be the issue. As if the coolant is low having it horizontal showing those problems but not vertical speaks to some active part of the board Gaming. I wonder how much system RAM these GPUs even use. This new driver may help to mitigate it by a few degrees.

 

[tabascosauz]

Agreed, I see the memory as being the culprit for the 110 C problem as well. I don't see the core but that top rank of 8 GB of Memory. could be the issue. As if the coolant is low having it horizontal showing those problems but not vertical speaks to some active part of the board Gaming. I wonder how much system RAM these GPUs even use. This new driver may help to mitigate it by a few degrees.

23.1.1 doesn't change much in the way of anything. Dunno about Radeon Software since I don't have it.

The 110C issue is not quite the same issue. there are a variety of readings visible in a GPU Hotspots dropdown in HWInfo. If you upgraded from RX6000 you should reset HWInfo. GCD Hotspot, MCD Hotspot (1-6 for XTX), each VRM component, and GDDR6 all have separate hotspot readings.

I've been on the HWInfo beta, it has some Navi 31 improvements. Looks like the release has been merged with the beta today

 

[kapone32]

23.1.1 doesn't change much in the way of anything. Dunno about Radeon Software since I don't have it.

The 110C issue is not quite the same issue. there are a variety of readings visible in a GPU Hotspots dropdown in HWInfo. If you upgraded from RX6000 you should reset HWInfo. GCD Hotspot, MCD Hotspot (1-6 for XTX), each VRM component, and GDDR6 all have separate hotspot readings.

I've been on the HWInfo beta, it has some Navi 31 improvements. Looks like the release has been merged with the beta today

I have the updated version and have commented before that i find that the Memory temp is higher than the Hot Spot Temp. I was using the PC with zero fan enabled and had 96 C on the Memory with a 2300 RPM Fan(s) speed.

 

[AusWolf]

Though, I wonder if that is actually reasonable to accept. Not qualified enough to comment on Radeon but at this point we should be paid for the investigative work that we do for the AGESA team. Outside of AMD launch day seems to still mean "the product is ready", not "congratulations, you ARE the engineering sample"

I completely agree, though it has been happening all across the board. You can look at Hollywood and all the half-written rubbish they're dumping on us. Or the gaming industry and the scandals of half-finished games that get patched to an acceptable state at a later time. Every new generation of product or service you're buying nowadays would have been classed as "in beta state" 10 years ago. This is what happens when the whole Western world's business model has been set up to have X amount of time to develop a product that has to be X times better than the last and the scale never changes. It's all about meeting business targets and quotas - quality doesn't matter. It's disgusting, but this is the world we're consuming, unfortunately.

 

[Space Lynx]

man this thread is really making me want to keep my 6800 XT now. I fixed the hotspot temp on my 6800 XT, it was a gpu sag issue since this is a giant ass card. measured and cut a wooden pencil to prop up the end of it, so there is no more sag, and my hotspot went from 99 celsius to 84 celsius in same tests in same config, etc.

i don't think I want next gen anymore, I mean I would gain what 30 fps in several games, 40 in some, but my 6800 xt is oc'd like a mofo and still runs good temps, getting 2520 core in games now, and gained 10 fps, so really i am looking at what 7900 xt maybe be 20 fps faster at 1440p... based on the reviews of games I play anyway. God of War I would still benefit from like a 40 fps gain, but that is an outlier in the benchmarks.

 

[AusWolf]

man this thread is really making me want to keep my 6800 XT now. I fixed the hotspot temp on my 6800 XT, it was a gpu sag issue since this is a giant ass card. measured and cut a wooden pencil to prop up the end of it, so there is no more sag, and my hotspot went from 99 celsius to 84 celsius in same tests in same config, etc.

i don't think I want next gen anymore, I mean I would gain what 30 fps in several games, 40 in some, but my 6800 xt is oc'd like a mofo and still runs good temps, getting 2520 core in games now, and gained 10 fps, so really i am looking at what 7900 xt maybe be 20 fps faster at 1440p... based on the reviews of games I play anyway. God of War I would still benefit from like a 40 fps gain, but that is an outlier in the benchmarks.

I agree. Upgrading every gen is a waste anyway (unless you just like to tinker with stuff).

There is this thread, though...

 

[Space Lynx]

I agree. Upgrading every gen is a waste anyway (unless you just like to tinker with stuff).

There is this thread, though...

Yeah, I already saw that thread, it doesn't worry me. My particular 6800 XT is a monster. 3 power plug design, so my guess is, w.e AMD may have fucked up in the drivers won't effect a card of this design which is designed for higher voltages out the door. I may be wrong, but I am not worried as I have been running latest drivers for over a month now with stress tests and gaming a lot, 0 issues.

 

[AusWolf]

Yeah, I already saw that thread, it doesn't worry me. My particular 6800 XT is a monster. 3 power plug design, so my guess is, w.e AMD may have fucked up in the drivers won't effect a card of this design which is designed for higher voltages out the door. I may be wrong, but I am not worried as I have been running latest drivers for over a month now with stress tests and gaming a lot, 0 issues.

That's cool. There's a few things suspicious about it anyway which you may pick up in the thread. Here's a couple:
1. Why is the GPU chip cracked? How does a driver do that?
2. Why only 6800 and 6900 cards?
3. What does the video's maker mean by "December 8 driver update"? Is it 22.11.2, or did the 7900 series' 22.12.1 driver install with these cards somehow? Both were issued on 8th Dec.
4. We don't know if the cards were overclocked, or if any OC tool was installed, etc.
And so on...