Blackmagic Forum

Hi all,

I’m looking to get a graphics card for use with Resolve and I was initially set on an Nvidia card because they were ranked higher for productivity tasks, but it seems AMD has stepped up and become a lot more comparable recently. I’m currently looking at the RTX 4080 or the RX 7900 XTX. Unfortunately, both are likely going to end up being around the same price ($1,200) as I'll probably need to upgrade my current 750W PSU for the 7900 XTX (according to AMD, the minimum requirement is 800W). Given that I won’t likely be saving much (if any) money by going with AMD, is it still worth it over Nvidia?

I tend to edit in 1080, but I’m leaving 4k open as a possibility. I will not likely be doing any 8k or RAW footage, so these are not a concern for me.
I make frequent use of Magic Mask, Paint node, optical flow, Text+, tracking, various types of blur and color, etc.
I would like to be able to use the noise reduction, speed warp and depth mask but without any graphics card, my system will not process these effects at all (trying to get them to cache is far too slow to deal with).

I’ve looked at the Puget Systems benchmarks for Resolve which include both of these cards:
https://www.pugetsystems.com/labs/artic ... on-review/
The two benchmarks that I care about are the effects and encoding/decoding. They have the 4080 slightly ahead for effects and the 7900 slightly ahead for encoding/decoding. Based on those two, it’s essentially tie for me, but certainly there are far more factors to consider that were not included in those two charts.

I’ve also watched several videos comparing them, but of course, content creators can have biases. And it also appears AMD has had a driver update since some of the videos were made which may have fixed many of the issues brought up in those videos.

My system specs are in my signature. Any recommendations and anything else I should be considering before making the purchase?

Thank you.

Perilous though this subject is, both sound excessive for your purposes, unless the benchmarks for either one indicate you'll get real time with these cards but not lesser ones for your frequently used fx.

Look at heat, noise, manufacturer/driver support, the size and scope of the user base for each. Assess the importance, if any, of CUDA to you. Consider frequency of reports of anomalies for either manfacturer in places like this, though difficult to compare usefully since the nvidia/Resolve base is much larger. Confirm, don't assume, that any known issues for either card have been fixed with driver updates. If anyone tells you only idiots and dumb youtubbers have problems with any given card, assume for safety's sake that you're also an idiot unless you want to spend the next few weeks or months trying to prove otherwise.

John Paines wrote:Perilous though this subject is, both sound excessive for your purposes, unless the benchmarks for either one indicate you'll get real time with these cards but not lesser ones for your frequently used fx.

Hi John, thanks for the reply.

For performance, Casey Faris did a side by side comparison of the 3070 vs the 4080 and the difference is staggering. Jump to about 2:26 for a speed comparison using the Magic Mask in the Color page, it's a night and day difference.


That alone is almost worth the price due to the amount of headache I deal with re-running the Magic Masks because they frequently break. If that process was nearly instantaneous, then this would be so much less of a hassle. Casey also stated that just using Fusion was a lot more responsive, less waiting time for things to show on screen, even over the 3070.

I would be willing to take a performance hit to spend less, however, even previous generation cards are excessive in price so there seems to be no reason to spend the same (or more) for an older card. For example, Gigabyte 3080 12GB is $1,300 on both Newegg and Amazon. At that price, I might as well get a 4080 16GB founders edition for $100 cheaper and have a better card.

Similar on the AMD side, the 7900 XT is pretty significant downgrade in performance but it's only $100 cheaper, so not a good price/performance ratio in comparison. I don't want to go with older AMD cards as even the 3060TI beat out far more expensive AMD cards in Resolve according to Puget's benchmarks.

As much as I'd love to go cheaper, there just doesn't seem to be good options. I would be spending a lot anyways, even to get an older card. The GPU market is terrible right now, so I've picked a really bad time to try to pick one up.

As for specific brands, for Nvidia I'd go with the founders edition made by Nvidia themself. For AMD, the AMD manufactured 7900 XTX is out of stock and from what I've seen it had problems with the card itself, and might not be coming back (not sure)? The one I was looking at was by ASRock, but I'm open to other brands.

Thanks again.

Puget bench marked both

https://www.pugetsystems.com/labs/artic ... lve_Studio

ZRGARDNE wrote:Puget bench marked both
https://www.pugetsystems.com/labs/artic ... lve_Studio

Hi Zeb, thanks for the reply.

Funny, I linked to another Puget bench mark comparing both cards in my initial comment... but the results in your link are different. The one you posted has the 4080 ahead in both effects and encoding/decoding, where the one I posted had the 4080 ahead for effects and the 7900 XTX ahead for encoding/decoding. Yours was a more recent benchmark test and it put the 4080 ahead for both of the areas of interest to me.

Thanks again.

I bought what I could afford for my hobbyist 4K machine in my signature and I could not be happier with my decision.

I went with Nvidia simply for the solid driver reputation with DR.

Tekkerue wrote:

ZRGARDNE wrote:Puget bench marked both
https://www.pugetsystems.com/labs/artic ... lve_Studio

Hi Zeb, thanks for the reply.

Funny, I linked to another Puget bench mark comparing both cards in my initial comment... but the results in your link are different. The one you posted has the 4080 ahead in both effects and encoding/decoding, where the one I posted had the 4080 ahead for effects and the 7900 XTX ahead for encoding/decoding. Yours was a more recent benchmark test and it put the 4080 ahead for both of the areas of interest to me.

Thanks again.

True, Apparently the version of the Benchmark changed from V0.93.1 to V0.93.2. No clue what changed in the bench between them. Looks like the hardware stayed the same, Thread Ripper 32 core, 128gb ram. Some differences may be result of drivers.

The 93.2 version isn't even available on their site, so there is no change log for it either. Strange they are using an unpublished benchmark....
https://www.pugetsystems.com/labs/artic ... olve-1523/



I would argue the GPU effects score is the most telling #. This is how long your effects take to cache to preview your grade and render for your final export. The 'dnxhr to h.264' tests have no effects applied to them, so not really representative of real work.


7900XTX V0.93.1 Vs V0.93.2
173, 182

4080 V0.93.1 Vs V0.93.2
175, 189

4080 Vs 7900xtx for the two versions
101.1%, 103.8%


The 24gb of the AMD might be more valuable than the small performance increase from NV

ZRGARDNE wrote:The 'dnxhr to h.264' tests have no effects applied to them, so not really representative of real work.

My understanding is that the encoding/decoding will help with timeline speed, caching, and final export. So it's still important. In my case, since I use a lot of effects, you are probably correct that the effects would be the dominating factor here. But in your link, the 4080 won in both of these.

The 24gb of the AMD might be more valuable than the small performance increase from NV

Yeah, that's the one thing really making me consider the AMD card, because everything else seems to be leaning towards Nvidia. It's a tough call.

I also got a response from a YouTuber who uses Resolve and had tested the 7900 XTX and the 4080. He ended up going with the 4080 and said that when he tested the 7900 XTX in Resolve the AV1 codec did not work, so I should verify that it's working now if I was interested in the 7900 XTX.

I'm not in an immediate hurry to purchase, I'd rather take my time to get as much input as possible to make the best decision.

Tekkerue wrote:

ZRGARDNE wrote:The 'dnxhr to h.264' tests have no effects applied to them, so not really representative of real work.

My understanding is that the encoding/decoding will help with timeline speed, caching, and final export. So it's still important. In my case, since I use a lot of effects, you are probably correct that the effects would be the dominating factor here. But in your link, the 4080 won in both of these.

The 24gb of the AMD might be more valuable than the small performance increase from NV

Yeah, that's the one thing really making me consider the AMD card, because everything else seems to be leaning towards Nvidia. It's a tough call.

I also got a response from a YouTuber who uses Resolve and had tested the 7900 XTX and the 4080. He ended up going with the 4080 and said that when he tested the 7900 XTX in Resolve the AV1 codec did not work, so I should verify that it's working now if I was interested in the 7900 XTX.

I'm not in an immediate hurry to purchase, I'd rather take my time to get as much input as possible to make the best decision.

Any update? I'm comparing both cards now. Well, trying...

Paulos wrote:

Any update? I'm comparing both cards now. Well, trying... [/quote]

Definitely 4080, look at this post about Magic Mask
viewtopic.php?f=21&t=185869

Paulos wrote:Any update? I'm comparing both cards now. Well, trying...

I've decided on the 4080, but haven't made the purchase yet. I've been holding out hoping the price would come down at some point, but the price still hasn't budged on them.

Reviving this thread. I noticed that the latest "Puget System Verified" results for Resolve are showing hugely improved performance for the 7900 XTX compared to the RTX 4080. They've been uploading many throughout August. Scroll through for yourself:

https://benchmarks.pugetsystems.com/ben ... ults-table

An Intel Core i9-13900K with the RX 7900 XTX scores 3355 and 3970 in the extended and standard tests, respectively:
https://benchmarks.pugetsystems.com/ben ... ?id=155090

The same processor with an RTX 4080 scores 2980 and 3367, respectively:
https://benchmarks.pugetsystems.com/ben ... ?id=154008

That's quite a huge delta in favor of the AMD part. Unfortunately there is no accompanying article to explain what has changed over the past month or two. Again, these are Puget System Verified results.

stretchsje wrote:That's quite a huge delta in favor of the AMD part. Unfortunately there is no accompanying article to explain what has changed over the past month or two.

Thanks for the info. I've seen some stuff on AMD driver updates improving performance, but they were usually talking about games. Maybe these updates have also improved performance in Resolve? If so, then that's good news. I'll definitely keep an eye on this. Hopefully Puget will do a full article on it.

I think AMD card may not able to produce YUV 420 10 bit output video which is very common in video production industry, at least in 6000 serials cards it cannot.
Not sure in 7000 serial card improved or not.
Better to check detailed specifications of 7000 serial card.

I wouldn’t recommend that output for reference anyway.
That feature would be insignificant for grading.

stretchsje wrote:That's quite a huge delta in favor of the AMD part. Unfortunately there is no accompanying article to explain what has changed over the past month or two. Again, these are Puget System Verified results.

From just glancing at the breakdown the big difference is Fusion. Almost a third better. It's worse on 8k. About the same on 4K. Slightly better on effects.

Like always with the Puget tests you need to think about your own use case. Are you doing lots of Fusion? Concentrate on that. Are you doing 8K? Think about that. The GPU effects need even more thinking because you need to check which effects they're testing. Then think if those are the ones you need.

IMHO the best use of those numbers is to test against your current machine. See if it's improving on what you have.

Nick2021 wrote:From just glancing at the breakdown the big difference is Fusion. Almost a third better. It's worse on 8k. About the same on 4K. Slightly better on effects.

I'm not using 8k (and don't expect I will), but I use a lot of Fusion. Improved performance for Fusion and other GPU intensive effects is definitely what I need.

I'll try and throw my hat into the mix, as like most people here I've found no good source or recent info on the 7900 xtx performance on DR studio.

I ordered one today and I'll be updating this forum once it's installed and I've had a chance to compare it to my 3070.

If anybody would like to benchmark it with pretty heavy timelines, both in UHD and 8K, please PM me.

alphakrusher wrote:I'll try and throw my hat into the mix, as like most people here I've found no good source or recent info on the 7900 xtx performance on DR studio. I ordered one today and I'll be updating this forum once it's installed and I've had a chance to compare it to my 3070.

Awesome, I look forward to hearing your thoughts on it. I'm in no big hurry, and I might even check around Black Friday to see if there are any substantive deals.

Any updates, anyone? I think the 4080 will be out of my budget - the Ava cards have gone up in price in my country - I would look at used but it's still expensive. I'm now looking at 4070 Ti - but, the 12gb vram rubs me the wrong way. A used 3090 or 3090 Ti can be had pretty easily but I am wondering if it's a better choice over a 7900 XTX - which I sometimes find used - and it's close to doable.
I'm surprised that there's quite a few 7900 XTX users/owners here. I suspect some are having issues - I have read a few posts of ppl having problems?
But, there's a few 'AMD fans' too, though.

I plan to use Blender, also, though - and I would like to use these programs in Windows and Linux.
HIP-RT is still experimental and doesn't work in Linux yet. Nvidia CUDA/Optix - is generally solid.
The 7900 XTX, theoretically, has good performance in Davinci Resolve though - it's been tested by Puget Systems, techgage and a few other reviewers - the performance has been good - acceptable to me - I just dunno about real world experience - as reviews/reports seem to be mixed.

It would also cost me a bit more - but, at least, I get a current gen gpu and it is AMD's flagship - there's some advantages to using it in Linux - although, I'm aware that setup and configuration of it in Linux for DR can be trickly - or so it appears to be.

Black Friday is coming up in a few weeks or so - so, maybe some deals or at least used gpu sellers may need to lower their asking price? I was thinking of buying this month (Nov).

There are more people around here that will ''recommend'' a 3060 over anything AMD because of random reasons like CUDA or ''bad drivers''. And it gets repeated ad nauseaum without any evidence. Both brands have good and bad drivers and quirks like VRAM memory full on nvidias part, thumbnails delay on amd side and so on.

And it all boils down to : whats your workflow?

Do you plan on using Resolve 95% of the time with RAW codecs, Prores and h264/265 non 4:2:2? Get the 7900XTX, its as close to the 4090 as possible. For the other 5% spent in Blender it works good enough.

If it's the other way around get the 4080.

With that said none of these cards are good for serious 3D work as most complex scenes don't fit in the actual VRAM so unless you do low res previz... meh...And viewfinder performance is the same afaik.

If you have a specific case you want to ask about, as a 7900XTX and former 6800XT owner let me know and I'll try to answer. For me , on Windows 10, with a 5800X3D these have been stable cards.

Here is a comparison with neural engine:

The biggest issue with the 40 series, is that the power cable design is crap. If it goes south, it can take out the gpu, your pc, or worse......

Dan Sherman wrote:The biggest issue with the 40 series, is that the power cable design is crap. If it goes south, it can take out the gpu, your pc, or worse......

Making sure the cables are fully and properly seated will mitigate that -- but *NOTHING* will mitigate the price of an RTX4090

RCModelReviews wrote:

Dan Sherman wrote:The biggest issue with the 40 series, is that the power cable design is crap. If it goes south, it can take out the gpu, your pc, or worse......

Making sure the cables are fully and properly seated will mitigate that -- but *NOTHING* will mitigate the price of an RTX4090

I personally know a person it happened to. His was seated properly, but over the course of 9 or 10 months of maintenance (he's anal about dust) and upgrades, it looks to have opened up the pins.

He and I both assumed it was because the cable is so stiff. Bumping it while doing maintenance probably just torqued the connector.

Honestly, for as expensive as the cards our and the amount of power they can pull through the cable, they should probably be secured with screws like a DVI, VGA, and serial cables.

panos_mts wrote:Here is a comparison with neural engine:

Might need to be retested.

23.11.1

DirectML improvements and optimizations for Stable Diffusion, Adobe Lightroom, DaVinci Resolve, UL Procyon AI workloads on AMD Radeon RX 600M, 700M, 6000, and 7000 series graphics.

Driver was released Nov 2nd.

Thanks for the info everyone. And if anyone finds any good holiday/Black Friday sales for either card, please post them.

VMFXBV wrote:There are more people around here that will ''recommend'' a 3060 over anything AMD because of random reasons like CUDA or ''bad drivers''. And it gets repeated ad nauseaum without any evidence. Both brands have good and bad drivers and quirks like VRAM memory full on nvidias part, thumbnails delay on amd side and so on.

And it all boils down to : whats your workflow?

Do you plan on using Resolve 95% of the time with RAW codecs, Prores and h264/265 non 4:2:2? Get the 7900XTX, its as close to the 4090 as possible. For the other 5% spent in Blender it works good enough.

If it's the other way around get the 4080.

With that said none of these cards are good for serious 3D work as most complex scenes don't fit in the actual VRAM so unless you do low res previz... meh...And viewfinder performance is the same afaik.

If you have a specific case you want to ask about, as a 7900XTX and former 6800XT owner let me know and I'll try to answer. For me , on Windows 10, with a 5800X3D these have been stable cards.

I'm going by a number of ppl who comment in the comment section of YT videos - of DR-based and Blender-based videos. There's a few other websites, too - AMD cards users often don't have positive reviews and some even switch to a Nvidia card, ultimately. All these cards are expensive - I've now 'downgraded' to looking for a 3090 or 7900 XT (around the same price; the 7900 xt might be a bit more expensive - both used). One is 24gb and the other 20gb - so, vram amt is covered.

In a perfect world, I am trying DR in Linux, too. An AMD card would probably be 'preferable' in that case but who knows - Blender support there is really slow and I'm not sure of the user experience for DR users in Linux. I suspect it's not as smooth as in Windows? I suspect the 'safe' choice is a Nvidia card but the 7900 XT would be a current gen card, would offer more features - AV1 - probably wouldn't need that but I like the bonus features.

It would probably be the 95% use case mentioned - I hope to create/edit videos (movies/home movies) - in 1080p/4k....just basic stuff to start with but hopefully, advance in time. I know I could pick any gpu to start with but, I'll use it for other things as well - but, DR would be the majority of the time, for sure. Probably starting in Windows - to keep things simple.

Dan Sherman wrote:

RCModelReviews wrote:

Dan Sherman wrote:The biggest issue with the 40 series, is that the power cable design is crap. If it goes south, it can take out the gpu, your pc, or worse......

Making sure the cables are fully and properly seated will mitigate that -- but *NOTHING* will mitigate the price of an RTX4090

I personally know a person it happened to. His was seated properly, but over the course of 9 or 10 months of maintenance (he's anal about dust) and upgrades, it looks to have opened up the pins.

He and I both assumed it was because the cable is so stiff. Bumping it while doing maintenance probably just torqued the connector.

Honestly, for as expensive as the cards our and the amount of power they can pull through the cable, they should probably be secured with screws like a DVI, VGA, and serial cables.

I see ppl still trying to get 4090s - mostly for gaming - they'll get the cable connector figured out - there's already aftermarket cables one can buy - and they'll have a new cable out and maybe just wait to see if they finally fix that issue?
Too much $$ for my blood.
How's your 7900 XTX in DR?

I noticed one of the most recent benchmarks by Puget - the 7900 XTX performance looks good - although, a used 3090 seems to keep up on a lot of tasks. It's way cheaper - but at a much higher power envelope. The 4070 Ti performs well but is behind - and the price is around the same as a 7900 XT.

Since we are talking about comparing performance here, which of you monitors hotspot temperature indicators when rendering?
I have a card more modest than yours - AMD 6700 xt, but no tests and benchmarks can compare with real noise reduction in 4K, when I have to cut the final film into intermediate 15-minute parts, so as not to burn the GPU memory.
And I know a whole list of NVIDIA 3 and 4 series cards that have big problems with video memory cooling, even for serious brands.

All the tests in the world are far from the real DaVinci load in noise reduction.

Z-Good wrote:Since we are talking about comparing performance here, which of you monitors hotspot temperature indicators when rendering?
I have a card more modest than yours - AMD 6700 xt, but no tests and benchmarks can compare with real noise reduction in 4K, when I have to cut the final film into intermediate 15-minute parts, so as not to burn the GPU memory.
And I know a whole list of NVIDIA 3 and 4 series cards that have big problems with video memory cooling, even for serious brands.

All the tests in the world are far from the real DaVinci load in noise reduction.

- 90C junction when rendering and NR on 7900XTX
- 105-110C junction on my former 6800XT reference
- 110C junction on my Radeon VII from my other workstation

All within spec. Why would you burn your memory?

Remember if you get around these numbers, they're well within spec and design to work at this temp.

While AMD chose for some weird reason to reveal junction temperature (well, the fans are based on it but still weird), Nvidia doesn't show junction temp in any software which is why some people are confused with temperatures thinking AMD cards are hotter. Hotspot temp is not GPU temp.

As long as its below 115C I wouldn't worry about it. Just make a custom curve in Adrenalin and ramp up the fans when temp goes over 100. Most default curves are set for quiet operation, not cooling.

Paulos wrote:I see ppl still trying to get 4090s - mostly for gaming - they'll get the cable connector figured out - there's already aftermarket cables one can buy - and they'll have a new cable out and maybe just wait

I expect they will have a completely new connector design for the 50 series.

Paulos wrote:How's your 7900 XTX in DR?

I noticed one of the most recent benchmarks by Puget - the 7900 XTX performance looks good - although, a used 3090 seems to keep up on a lot of tasks. It's way cheaper - but at a much higher power envelope. The 4070 Ti performs well but is behind - and the price is around the same as a 7900 XT.

So far its been perfectly fine for me. It has good performance, 24 GB of vram, and it is power efficient.

I was originally planning on getting a 4090, but Puget's benchmarks from the end of last year proved to me I would just be wasting money. Not to mention they have released new drivers that should make it even better in DR.

The scores noted above are averages ("extended"); they may not reflect actual use of the OP, who might want to consult the full benchmark series if he wants to go the benchmark route.

It also bears noting that there are no bargains in this field, none of these manufacturers has a secret. And it's hard to see how buying any well-established, well supported and widely used product, in professional facilities, is is a "waste" of money. Driver support and size of the user base are not insignificant factors for anyone who wants to install and forget.

Threads which deal with individual product lines, and trouble with them, rather than "what should I buy", may be more helpful in making purchases, as you're less likely to encounter brand enthusiasms and personal or commercial biases for either product.

VMFXBV wrote:

Z-Good wrote:Since we are talking about comparing performance here, which of you monitors hotspot temperature indicators when rendering?
I have a card more modest than yours - AMD 6700 xt, but no tests and benchmarks can compare with real noise reduction in 4K, when I have to cut the final film into intermediate 15-minute parts, so as not to burn the GPU memory.
And I know a whole list of NVIDIA 3 and 4 series cards that have big problems with video memory cooling, even for serious brands.

All the tests in the world are far from the real DaVinci load in noise reduction.

- 90C junction when rendering and NR on 7900XTX
- 105-110C junction on my former 6800XT reference
- 110C junction on my Radeon VII from my other workstation

All within spec. Why would you burn your memory?

Remember if you get around these numbers, they're well within spec and design to work at this temp.

While AMD chose for some weird reason to reveal junction temperature (well, the fans are based on it but still weird), Nvidia doesn't show junction temp in any software which is why some people are confused with temperatures thinking AMD cards are hotter. Hotspot temp is not GPU temp.

As long as its below 115C I wouldn't worry about it. Just make a custom curve in Adrenalin and ramp up the fans when temp goes over 100. Most default curves are set for quiet operation, not cooling.

Learn to undervolt, those tempereatures will drop to around 70°C

My 6800XT is set to 1050mV and a target of 2400 MHz, default was 1150 mV. The 7000 series is able to drop below 1V for everyday use.

John Paines wrote: And it's hard to see how buying any well-established, well supported and widely used product, in professional facilities, is is a "waste" of money.

It's a pretty simple concept John. With regards to DR, a gpu is just a tool. And just like with any other tool, not everyone needs the same thing.

For example, Festool sells a $1500 miter saw, but not everyone needs a $1500 miter saw. I have a friend who is a professional contractor, and even he doesn't have a Festool, he has a Dewalt, that costs less than half what the Festool does.

John Paines wrote:Driver support and size of the user base are not insignificant factors for anyone who wants to install and forget.

Both companies have a large user base, and most people just install graphics cards and forget about them,

John Paines wrote:Threads which deal with individual product lines, and trouble with them, rather than "what should I buy", may be more helpful in making purchases, as you're less likely to encounter brand enthusiasms and personal or commercial biases for either product.

Do you want me to send you a photo of the 1070 i have in the machine a few feet away?

Dan, an even simpler concept is, unhelpful analogies to power drills, assertions that proven and widely used products in professional settings are a waste of money and anecdotal claims that you'll destroy your machine if you buy the disfavored item ("I personally know a person it happened to") may not be as illuminating as their contributors wish.

A common feature of these threads, known to promote one particular manufacturer (guess which one?) is that the loudest advocates, by all appearances, are neither editors nor colorists. That wasn't always the case, but the editors and colorists, most of whom aren't interested in gear, quickly discovered that the argument wasn't worth the grief.

John Paines wrote:Dan, an even simpler concept is, unhelpful analogies to power drills,

On the contrary john, I've found that a simple analogy can help people understand even complex subjects.

John Paines wrote:A common feature of these threads, known to promote one particular manufacturer (guess which one?) is that the loudest advocates, by all appearances, are neither editors nor colorists.

So does that mean you don't want me to post photos of the Nvidia and Intel products I own?

Dan Sherman wrote:
I was originally planning on getting a 4090, but Puget's benchmarks from the end of last year proved to me I would just be wasting money. Not to mention they have released new drivers that should make it even better in DR.

The 4080 and 4090 are many times faster than 7900 XTX at certain things in Resolve. It also has Cuda and that means just about everything AI related is compatible with it, AMD users would be very sad looking at all the great free software out there that isn't compatible with it.

Also worth noting is the upcoming 16GB 4070ti super and 4080 super, with 4070 ti super only slightly slower than current 4080, and rumours of 4080 super pricing for as low as US$1000. There would be no reason to buy AMD after this release unless AMD adjust pricing

ohimbz wrote:
Learn to undervolt, those tempereatures will drop to around 70°C

My 6800XT is set to 1050mV and a target of 2400 MHz, default was 1150 mV. The 7000 series is able to drop below 1V for everyday use.

I know how to undervolt. I'm undervolting my recently updated 5800X3D with a -30curve and its going strong. My old 5800X first batch didn't accept -30. Barely -5 so I didn't even bother.

Regarding the GPUs, I have / had one of the first batches of reference cards with the Radeon VII and 6800XT and didn't win the silicon lottery so they don't really accept a good undervolt. -25mV at max so I didn't bother with them.

Just ramped the fans for 100C when rendering and kept them on quiet mode for when I gamed on the 6800XT for example.

The temps you saw up there were on quiet mode. They do drop quite a bit on 100% fan.

The 7900XTX is different and not reference but I didn't have time to play with it. Will try to give it an undervolt when I'll have more time.

CougerJoe wrote:The 4080 and 4090 are many times faster than 7900 XTX at certain things in Resolve.

We have work stations at work that have RTX 6000s. However I wouldn't recommend them to most DR users. DR supports up to 8 gpus if memory serves, but very few users are going to see a benefit from more than 1.


As with everything tech related, it all comes down to what you need.

CougerJoe wrote:
The 4080 and 4090 are many times faster than 7900 XTX at certain things in Resolve.

While being slower and many times more expensive in other things...

CougerJoe wrote:
It also has Cuda and that means just about everything AI related is compatible with it, AMD users would be very sad looking at all the great free software out there that isn't compatible with it.

Name a few known pieces of software that don't work on AMD and are Nvidia only? Resolve "AI" features? They work fine. Stable diffusion like stuff? Works fine. Adobe? Works fine. What else?

Its also weird that these things keep popping up since they don't really have anything to do with Resolve.

Why would other pieces of software that aren't even video editing related (some of them) even matter for someone who asks specifically about Resolve? Do all the Nvidia users suddenly do programming in CUDA or start using Tensor Cores scientifically or something?

Its just odd but whatever floats your boat.

VMFXBV wrote:

CougerJoe wrote:
The 4080 and 4090 are many times faster than 7900 XTX at certain things in Resolve.

While being slower and many times more expensive in other things...

CougerJoe wrote:
It also has Cuda and that means just about everything AI related is compatible with it, AMD users would be very sad looking at all the great free software out there that isn't compatible with it.

Name a few known pieces of software that don't work on AMD and are Nvidia only? Resolve "AI" features? They work fine. Stable diffusion like stuff? Works fine. Adobe? Works fine. What else?

Its also weird that these things keep popping up since they don't really have anything to do with Resolve.

Because people do more than edit on Resolve with the same computer. There are many releases that are cuda only because evidently it's so easy to program with, and this locks out AMD, which was always the intentions of Nvidia .

The list is too many to mention. But the amount of times I've read 'must have Nvidia GPU' it would be downright depressing for an AMD owner. For this reason if you're interested in opensource github AI releases I'd stick to Nvidia.

CougerJoe wrote:
Because people do more than edit on Resolve with the same computer.

Last time I looked at this thread, the title was 4080 or 7900XTX for Resolve.

The vast majority of computer users don't care about CUDA as an API or even know what CUDA or Machine Learning is (the so called "Ai") which makes CUDA irrelevant to the vast majority.

There are many releases that are cuda only because evidently it's so easy to program with, and this locks out AMD, which was always the intentions of Nvidia .

The list is too many to mention. But the amount of times I've read 'must have Nvidia GPU' it would be downright depressing for an AMD owner. For this reason if you're interested in opensource github AI releases I'd stick to Nvidia.

So, as expected, you have no examples... Maybe because 99% of software out there is GPU and CPU agnostic...Just maybe...

Also, as an AMD user I've been called a bunch of stuff on this forum, but first time I hear I should be depressed because I can't run some random software that the vast majority don't care about. All those Mac users must also be popping up depression pills like crazy not being able to use CUDA goodness.

Just weird. Good luck.

You would have heard terms Cuda, Pytorch, Tensorflow. That is Nvidia GPU only or CPU, that's much of the AI software on github The sort of software that's Resolve adjacent could be faster subtitle generation, better optical flow, voice cloning, removing silences in a video. For the longest time the only '1 click' Stabile diffusion installs for windows was Nvidia only.

You need to include those provisos otherwise anyone buying AMD on your recommendation may learn to hate you. GIve them the full picture.

As expected, no examples that are relevant to the majority of PC users. The vast majority doesnt care about github. And more so editors, colorists and most of the people that ask about gpus on this forum.

But as with all your examples so far, I'll byte.

https://pytorch.org/blog/amd-extends-support-for-pt-ml/

Tensorflow also works on AMD cards.

DirectML (most important ML framework imo) is hw agnostic and works.

https://learn.microsoft.com/en-us/windo ... ow-plugins

Can you come up with more stuff?

2 out of your 3 terms work on AMD cards and the third one is an imposibility since its a proprietary black boxed API. But hold on, even that works with HIP...which translates cuda code.

Also really, people buying AMD on my or other user's recommendations for Resolve will hate me and love you? Kumbaya?

CougerJoe wrote:You would have heard terms Cuda, Pytorch, Tensorflow. That is Nvidia GPU only or CPU, that's much of the AI software on github The sort of software that's Resolve adjacent could be faster subtitle generation, better optical flow, voice cloning, removing silences in a video. For the longest time the only '1 click' Stabile diffusion installs for windows was Nvidia only.

You need to include those provisos otherwise anyone buying AMD on your recommendation may learn to hate you. GIve them the full picture.

AMD you would want Vulkan or OpenCL for efficient use of an AMD GPU. It really is unfortunate that Nvidia cornered and now dominates the market for AI. Even the AI in Resolve is so much faster than AMD

CougerJoe wrote:AMD you would want Vulkan or OpenCL for efficient use of an AMD GPU. It really is unfortunate that Nvidia cornered and now dominates the market for AI. Even the AI in Resolve is so much faster than AMD

You keep moving goalposts. We moved from AMD doesn't work with Ai (and surprise it actually works fine) to Nvidia dominates the market and its faster in Resolve with AI.

Who cares about market dominance? Who cares about Ai on this forum outside of Resolve's own AI tools and for which benchmarks clearly show which is faster and which isn't. Will nvidia's market dominance make a 3060 debayer 8K RED faster using CUDA? The more AI market share it has the faster it gets?

To quote a famous movie, ''Activate Ludicrous speed!''.

Still waiting on that huge list of examples but all I'll probably get is another wildy inaccurate take on what makes AMD cards inneficient.

Simples. Blender. AMD is no where near close Nvidia there.
Actually in pretty much any 3D GPU render except their own.
If you di 3D work you have a strong reason to go Nvidia.
Only in Mechanical CAD Viewport AMD have some good cards.

CougerJoe wrote:AMD you would want Vulkan or OpenCL for efficient use of an AMD GPU. It really is unfortunate that Nvidia cornered and now dominates the market for AI. Even the AI in Resolve is so much faster than AMD

I agree with this - Although, DR will be my main program I'll use - I want a gpu that is pretty good in other areas, too - AI/ML, Blender, video editing in general. I might try these programs in Linux, too - which is another potential roadblock.

With all that in mind, perhaps, a Nvidia gpu is the best choice? The problem with Nvidia, however - is twofold (for me): A) many gpus are lacking vram - many are only 8, 10 or 12gb - and the more expensive cards are 16 and up - or the higher vram cards are older/previous gen. The previous gen Nvidia cards - Ampere for e.g. - are higher power consumption and can have transient spikes so you need a high quality psu - and at least 850w - preferably more wattage.

I know you can mess with voltage, undervolt, fan curves, power limits - but, when you try outside the Windows ecosystem, it gets a lot more complicated? At least, that's my impression.

On the AMD side, I'm only interested if it's a 7900 series/rdna 3 card - they have the improvements in architecture - that's better for productivity - ROCm/AV1/OpenCL/HIP-RT* (*no Linux support yet) and as the various benchmarks posted here, show - theoretically, good performance in Davinci Resolve.

I would like at least 16gb of vram - JUST IN CASE - I don't want to be re-visiting 'upgrading' or replacing a gpu with another gpu because the vram isn't enough. I'm fine if I pick Nvidia - and there's issues - but, it still functions well enough - and vice versa, I choose AMD gpu - but, spending over $800 and the vram is insufficient is a silly mistake I don't want to risk. If 12gb is enough, well, having more than another is a nice luxury.

Nvidia is going to release these new cards to try to take sales away from the 7900 XT - a 4070 Super, 4070 Ti Super and a 4080 Super - but, I am afraid these cards will also be expensive in my country - so, although, they look good on paper - I might be looking at whether the regular 4080 is posted to more used classifieds - I doubt it though - it'll probably be older 30 series?

The way I see it, I perceive a conundrum - Nvidia or AMD cards are good for Davinci Resolve - but, once you use the cards for other use or in other operating systems - then one or the other is a better choice on that. The other question is how are these AMD gpus for Linux - if you want to try DR - I think there's more complications there. I don't know if you can use openCL & then need proprietary elements (i.e. requiring amdgpu-pro) or whether ROCm & OpenCL Mesa is enough (some ppl claim it is).
The thread for Linux is currently 42 pages long - so, to me, that means, there's been some troubles getting it to work - but, obviously, Linux has a much more varied configuration and complicated software/code system - so much, that DR went with official support on Red Hat/CentOS - which I feel isn't up-to-date anymore - as CentOS is pretty much obsolete, now.

The advantage of Nvidia-proprietary-one driver - and the longevity of it being straight forward - install the Nvidia repo for that distro - is that all the components should be there for DR - right?

Sorry to ramble.... in sum, my contender list is: 3090 (24gb), 4070 Ti Super (16gb)*, 4080 (16gb), 7900 XT (16)
All those gpus - I'm gonna pursue a used card except for the 4070 Ti Super - which isn't released yet.