Review AMD Radeon RX 9070 XT and RX 9070 review: An excellent value, if supply is good

[-Fran-]

People rarely talk about the real-world experience of various drivers, just nebulous statements. I will say I encounter oddities with AMD more than with Nvidia, but for the RDNA 4 launch things were mostly smooth. Why only mostly? I'll give you one concrete example.

I installed and tested the RX 9070 XT card. Everything went fine. When finished, I shut down the PC and installed the RX 9070 card. And... nothing. Oh, I saw the POST screen, but the Windows login screen never came up. But the PC was alive! I could ping it! So I tried remotely connecting. Nothing. I futzed around for possibly 30 minutes, wondering if I had a DOA 9070 sample.

So then I tried it in my old testbed, and had the exact same problem. Nothing. Sigh... more time wasted. But then I though, "You know, I had the 9070 XT in both systems. I wonder if the drivers are failing to load because they're looking for the 9070 XT and crapping out when they find the 9070?" So I put the 9070 into my even older 12900K testbed, and it booted right up to Windows. (No drivers installed yet.)

I then had to put the 9070 XT back into my testbed, boot into Windows, run Display Driver Uninstaller to wipe out the drivers, shutdown and put the 9070 in... and it worked.

I wish I could say this was a rare occurrence, but AMD GPUs don't really seem to like switching between graphics cards. Usually I'll get the PC to boot and the first launch it will BSOD and restart and then work. Other times, it will boot but act like no AMD drivers are installed, so I need to reinstall the drivers. On Nvidia, though, swapping between cards while using the same drivers almost never causes issues.

So why not just wipe the drivers every time I swap? I could, I probably should, but sometimes you don't want to wait an extra 10 minutes to wipe, shutdown, swap cards, boot, reinstall. Remember: MASSIVE time crunch for the past couple of weeks.

------------

Now, that's the AMD side. Nvidia's drivers with Blackwell are still problematic in a couple of games in my test suite. Minecraft at "ultra" (24 RT render chunks) doesn't work right on the 9800X3D system at 1080p, sort of 1440p and 4K. The slower the card (i.e. 5070) the less problematic it is. But the 5090? It runs terribly right now in Minecraft at 1080p — but not with 8 RT render chunks setting. This has been a known issue since early January now.

Warhammer 40K: Space Marine 2 also has some performance oddities. On both AMD and Nvidia Blackwell, performance is capped to around 150~170 FPS. Put in an RTX 40-series GPU and suddenly it will hit 200+ FPS. Again, Nvidia is aware of this but hasn't provided a fix in nearly two months. It did fix weaker than expected 1080p medium/ultra results for quite a few games compared to the launch 5090 drivers at least.

The black screens on Nvidia are another example of issues. I sent an MSI Suprim Liquid SOC back because I couldn't finish all the tests and MSI offered no explanation. Two days after I shipped it back, it released a VBIOS update to fix black screen issues and Nvidia put out new drivers. LOL. Why couldn't anyone just tell me, "Yes, we're aware of the problem and a fix is impending?" and save me almost a week of chasing my tail?

And how on earth did a bunch of GPUs with one missing ROPS cluster get shipped out? The "bad batch" explanation doesn't hold water, since it happened on 5090, 5080, and 5070 Ti cards IIRC. None of the cards I have exhibited this problem, but that's two completely different chips and three configurations, so it wasn't a single "batch."

The bottom line is that new architectures are usually far more problematic from a drivers standpoint. It's surprising how bad things can get, and AMD and Nvidia have been less kind to the AIBs lately in order to try to lessen the number of leaks. I'm pretty sure the entire 40-series was leaked weeks before it launched. The 5090 and 5080 had plenty of rumors but a LOT of wrong information that obviously wasn't leaked, just bad guesses.

/notes from testing new GPUs...

Sorry, but you don't run a clean windows installation per card per test? Did I understand that correctly?

I know we all trust DDU and all, but wouldn't that be a variable you can't control in the testing? Specially with pre-release hardware?

Not trying to criticize, but understand your rationale behind it.

Regards.

 

[palladin9479]

Sorry, but you don't run a clean windows installation per card per test? Did I understand that correctly?

It would take them a month or more to do a round of testing if they had to do that.

You see how they get almost no sleep on the leadup to a launch, and that's with them keeping one set of drivers installed.

 

[-Fran-]

It would take them a month or more to do a round of testing if they had to do that.

You see how they get almost no sleep on the leadup to a launch, and that's with them keeping one set of drivers installed.

For Windows installations and a separate disk for games, you don't need much space and it may even save you time, as you'd have a "golden" NVMe drive copy which you copy and transfer the system to. This gives you as many NVMes you can swap out per system as you need.

It's more expensive, for sure, but with open beds and/or easy access case and leaving the NVMe cover aside, it may save time in the long run, but you need to have at least 5 of those on rotation. This makes it 5 OS drives and 1 game drive per system you want to have running in parallel.

Again, not trying to criticize, but understand.

Regards.

 

[palladin9479]

For Windows installations and a separate disk for games, you don't need much space and it may even save you time, as you'd have a "golden" NVMe drive copy which you copy and transfer the system to. This gives you as many NVMes you can swap out per system as you need.

They already use a default master install image, that's standard. It's all the work that comes afterwards.

Plus I'm not sure you've caught up but it's not the 2000's anymore, you can't just move a secondary disk with games on it around and have them work. You need to reinstall each game to have it recreate it's registry keys and reregister all it's DLL's, not to mention Steam or the games installer is going to redownload the entire thing anyway.

The way everyone generally works is they have a standard platform already built and ready with all the tools / games / etc they need. They install the appropriate drivers, then run the benchmark on each game under strict conditions, record the results, change settings, rerun the benchmark, record the result, so on and so forth. Once one game is done, restart system (usually), run the next set of benchmarks, and you repeat this process until you get done one product. Now shutdown, remove the product, install next product, boot up, ensure drivers working, now start your benchmarking all over again with the new product. When that's done, power down, install next product, start up, check drivers, restart benchmarking.

You run a complete cycle for each product you need data for. If you need ten cards worth of data, that is ten full benchmark cycles they have to go through. If you complain they should of had two other cards, that's two more cycles they need to run.

Then once you've exhausted your team collecting the data, you start analyzing it and writing the articles, generating the graphics, and that whole editorial process.

 

[-Fran-]

They already use a default master install image, that's standard. It's all the work that comes afterwards.

Uh... I think you didn't get my original question, then.

Plus I'm not sure you've caught up but it's not the 2000's anymore, you can't just move a secondary disk with games on it around and have them work. You need to reinstall each game to have it recreate it's registry keys and reregister all it's DLL's, not to mention Steam or the games installer is going to redownload the entire thing anyway.

The way everyone generally works is they have a standard platform already built and ready with all the tools / games / etc they need. They install the appropriate drivers, then run the benchmark on each game under strict conditions, record the results, change settings, rerun the benchmark, record the result, so on and so forth. Once one game is done, restart system (usually), run the next set of benchmarks, and you repeat this process until you get done one product. Now shutdown, remove the product, install next product, boot up, ensure drivers working, now start your benchmarking all over again with the new product. When that's done, power down, install next product, start up, check drivers, restart benchmarking.

You run a complete cycle for each product you need data for. If you need ten cards worth of data, that is ten full benchmark cycles they have to go through. If you complain they should of had two other cards, that's two more cycles they need to run.

Then once you've exhausted your team collecting the data, you start analyzing it and writing the articles, generating the graphics, and that whole editorial process.

No, you can do that. I do it all the time when migrating PCs and disks around the house. Steam just checks the files and only re-downloads whatever it finds not matching. Steam, if you install it in the same exact location, it even keeps your libraries across disks, if they match the config of the Windows installation. EDIT: I will re-run the installers if the registry for the games is in Windows, if that's what you were talking about, but that's a very small time spent "installing" things.

I'm talking 100% from experience with this, that's why it caught my attention.

Regards.

 

[palladin9479]

Do not ask questions if you don't want the answers.

 

[-Fran-]

Do not ask questions if you don't want the answers.

Ok, hold up. Are you gaslighting me here?

You're not even the person I asked the question to.

Ugh, I'll just stop here.

Regards...

 

[JayGau]

Best Buy has some in-store inventory, no? What about Target and Walmart?

At the moment I checked they were all out of stock, Best Buy included. Walmart and Target don't sell graphics cards in-store, only online (and the Target selection looks very limited).

 

[Avro Arrow]

As glad as I am that the Radeon 9000-series worked out well (except for their mentally-deficient naming scheme), I'm still shaking my head at the RX 9070. They pulled that same stupidity between the RX 7900 XT/XTX and again with the RX 7700 XT/7800 XT. It didn't work then and it won't work now.

With AMD giving $50 rebates to sellers, I also hope that the cards won't suddenly jump $50 in price if AMD stops doing that. It's not for me, I won't be upgrading my XTX until many years in the future but I want a market that is balanced to drive innovation up and prices down.

My coworker was looking at getting an RX 7800 XT and I asked him today about it. He said that he was waiting for them to come down to $700CAD. Well, I looked it up, kinda shocked that they hadn't yet, and found TONNES of availability for the Gigabyte RX 7800 XT Gaming OC for $700. He was clearly excited and was going to buy one until I showed him a white XFX RX 9070 for $800CAD. He didn't want a white card but he didn't care for that price. He's gone to get one and I hope that he succeeds.

With everything that's going on, Canadians are snapping them up because Radeon is Canadian.

 

[JayGau]

Lol what????? The 9070XT>XTX. You just saw a Nvidia shill YT channel review. Here is an avg from TP 14 games:

14 Avg
5080: 124,17
9070XT: 115,97
7900XTX: 113,9 (+2%)
5070Ti: 111.35


~Alan Wake 2
5080: 104.5
9070XT 106.1
5070Ti 93
7900XTX 102.4

~Assassin's Creed Mirage
5080: 145.6
9070XT 157.4
5070Ti 136.4
7900XTX 138.3

Baldies Gate 3
5080: 186.8
9070XT: 175.4
5070Ti: 168.8
7900XTX: 180.2

~Cyberpunk 2077
5080: 142.9
9070XT: 134.3
5070Ti: 125.2
7900XTX: 129.2

~Ghost of Tsushima
5080: 134.8
9070XT: 137.7
5070Ti: 119.4
7900XTX: 121.1

~God of War Ragnarok
5080: 148.4
9070XT: 132.7
5070Ti: 133.1
7900XTX: 120.2

Hogwarts Legacy
5080: 107
9070XT: 94
5070Ti: 94.8
7900XTX: 94.5

~Horizon Forbidden West
5080: 130.5
9070XT: 122.8
5070Ti: 116.2
7900XTX: 118.3

Resident Evil 4
5080: 156.5
9070XT: 152.4
5070Ti: 140.3
7900XTX: 165.5

~Silent Hill 2
5080: 79.7
9070XT: 58
5070Ti: 71.6
7900XTX: 56.3

Stalker 2
5080: 91
9070XT: 77
5070Ti: 84
7900XTX: 76

Star field
5080: 110
9070XT: 105.4
5070Ti: 101.2
7900XTX: 111.1

Star Wars Outlaws
5080: 87.1
9070XT: 74.3
5070Ti: 74.5
7900XTX: 68.5

The Last of Us
5080: 113.6
9070XT: 96.1
5070Ti: 100.5
7900XTX: 105.9

Looks like you did the exact opposite. You picked some AMD fanboy review to make your point. Because when I check other random reviews (like the one right here, on Tom's Hardware), the 9070 XT is always behind the 7900 XTX (not by much, but still behind), even in the games you listed here.

 

[Crazyy8]

I have been disappointed. No 9070 XTs at MSRP, all scalped to $750+. Out of stock online, scalped in store. Only the 9070 remains, and that's still a bad deal. AMD has failed, billions must buy the RTX 5010 1Gb 64 bit bus DDR3 Edition

 

[drivinfast247]

Looks like you did the exact opposite. You picked some AMD fanboy review to make your point. Because when I check other random reviews (like the one right here, on Tom's Hardware), the 9070 XT is always behind the 7900 XTX (not by much, but still behind), even in the games you listed here.

That is until ray-tracing comes into play.

 

[palladin9479]

As glad as I am that the Radeon 9000-series worked out well (except for their mentally-deficient naming scheme), I'm still shaking my head at the RX 9070. They pulled that same stupidity between the RX 7900 XT/XTX and again with the RX 7700 XT/7800 XT. It didn't work then and it won't work now.

With AMD giving $50 rebates to sellers, I also hope that the cards won't suddenly jump $50 in price if AMD stops doing that. It's not for me, I won't be upgrading my XTX until many years in the future but I want a market that is balanced to drive innovation up and prices down.

My coworker was looking at getting an RX 7800 XT and I asked him today about it. He said that he was waiting for them to come down to $700CAD. Well, I looked it up, kinda shocked that they hadn't yet, and found TONNES of availability for the Gigabyte RX 7800 XT Gaming OC for $700. He was clearly excited and was going to buy one until I showed him a white XFX RX 9070 for $800CAD. He didn't want a white card but he didn't care for that price. He's gone to get one and I hope that he succeeds.

With everything that's going on, Canadians are snapping them up because Radeon is Canadian.

GN did a thing on the 9070 and it makes sense in specific scenarios.

View: https://www.youtube.com/watch?v=LhsvrhedA9E


The 9070 is pretty close to the 9070 XT but only use's 220W worth of power compared to the 304W+ for the XT and it's OC models. It also can be a physically smaller card making it easier to fit into smaller cases.

Two slot / two fan model that was benchmarked
https://www.sapphiretech.com/en/consumer/pulse-radeon-rx-9070-16g-gddr6

Three slot / three fan model from same benchmark.
https://www.sapphiretech.com/en/consumer/pulse-radeon-rx-9070-xt-16g-gddr6

 

[palladin9479]

That is until ray-tracing comes into play.

Yes RT is it's own thing, the RX 7000 series was pretty bad at RT while the RX 9000 series has made significant leaps in that regard. Depending on the RT title, the 9070 XT will either match or beat a 7900 XT.

Previously if someone wanted to do RT they had to buy Nvidia, now they can buy AMD.

 

[alceryes]

I have been disappointed. No 9070 XTs at MSRP, all scalped to $750+. Out of stock online, scalped in store. Only the 9070 remains, and that's still a bad deal. AMD has failed, billions must buy the RTX 5010 1Gb 64 bit bus DDR3 Edition

Unfortunately, the damn bots got all the MSRP online stock (if there ever was some).

Those close to a Best Buy or Micro Center could've probably got in line when the store opened and still got one. The Micro Center I went to (New Jersey, USA) had like 300+ in stock. I think at least half of those were being sold at MSRP.

It looks like, going forward, you'll need to be at a brick-n-mortar store for any chance of getting an MSRP card, at MSRP. I can't blame AMD for this though. I mean, 300+ cards on launch day is pretty damn good!

 

[JarredWaltonGPU]

Uh... I think you didn't get my original question, then.


No, you can do that. I do it all the time when migrating PCs and disks around the house. Steam just checks the files and only re-downloads whatever it finds not matching. Steam, if you install it in the same exact location, it even keeps your libraries across disks, if they match the config of the Windows installation. EDIT: I will re-run the installers if the registry for the games is in Windows, if that's what you were talking about, but that's a very small time spent "installing" things.

I'm talking 100% from experience with this, that's why it caught my attention.

Regards.

It works with Steam, yes, mostly. It doesn't work with Epic, GoG, EA, Microsoft Store, or some others I don't use. No, I do not do a complete reimage of the testbed for every GPU. That would take way too much time. I do look for any anomalies in testing. Outside of some 50-series cards underperforming in a few games, I haven't seen anything unusual.

 

[palladin9479]

Unfortunately, the damn bots got all the MSRP online stock (if there ever was some).

Those close to a Best Buy or Micro Center could've probably got in line when the store opened and still got one. The Micro Center I went to (New Jersey, USA) had like 300+ in stock. I think at least half of those were being sold at MSRP.

It looks like, going forward, you'll need to be at a brick-n-mortar store for any chance of getting an MSRP card, at MSRP. I can't blame AMD for this though. I mean, 300+ cards on launch day is pretty damn good!

If they can keep up the production volume them those scalpers will end up sitting on huge inventories that aren't moving and eventually have to dump them onto amazon for at or below cost just to write them off. Commodity speculation only works if the commodities value goes up and more importanly stays up.

 

[stuff and nonesense]

If they can keep up the production volume them those scalpers will end up sitting on huge inventories that aren't moving and eventually have to dump them onto amazon for at or below cost just to write them off. Commodity speculation only works if the commodities value goes up and more importanly stays up.

We can only hope!

 

[valthuer]

It's not for me, I won't be upgrading my XTX until many years in the future but I want a market that is balanced to drive innovation up and prices down.

Hey man, good to see you back after all those months of absence! Glad you 're OK.

XTX is still a solid performer in 4K.

 

[qwertymac93]

They already use a default master install image, that's standard. It's all the work that comes afterwards.

Plus I'm not sure you've caught up but it's not the 2000's anymore, you can't just move a secondary disk with games on it around and have them work. You need to reinstall each game to have it recreate it's registry keys and reregister all it's DLL's, not to mention Steam or the games installer is going to redownload the entire thing anyway.

The way everyone generally works is they have a standard platform already built and ready with all the tools / games / etc they need. They install the appropriate drivers, then run the benchmark on each game under strict conditions, record the results, change settings, rerun the benchmark, record the result, so on and so forth. Once one game is done, restart system (usually), run the next set of benchmarks, and you repeat this process until you get done one product. Now shutdown, remove the product, install next product, boot up, ensure drivers working, now start your benchmarking all over again with the new product. When that's done, power down, install next product, start up, check drivers, restart benchmarking.

You run a complete cycle for each product you need data for. If you need ten cards worth of data, that is ten full benchmark cycles they have to go through. If you complain they should of had two other cards, that's two more cycles they need to run.

Then once you've exhausted your team collecting the data, you start analyzing it and writing the articles, generating the graphics, and that whole editorial process.

It works with Steam, yes, mostly. It doesn't work with Epic, GoG, EA, Microsoft Store, or some others I don't use. No, I do not do a complete reimage of the testbed for every GPU. That would take way too much time. I do look for any anomalies in testing. Outside of some 50-series cards underperforming in a few games, I haven't seen anything unusual.

I may be off base here as my world is mostly virtualized servers and VDI but... Is a WDS/WDT server holding preconfigured images with all the games and benchmarking software already installed and activated not possible? Have sysprepped images been attempted in the past and they just don't work?

 

[Avro Arrow]

So, 3 monitors using display port for FS2024 is a non-starter? Have to mix technologies (DP and HDMI?) Is this going to be true for most RX 9070 XT cards?

Stupid question (I've never used multi-monitors so I'm unfamiliar with that kind of setup):
Does mixing DP and HDMI cause problems?

Nvidia, without price adjustments, is no longer competitive in this market segment, IMO.

Unfortunately, that has been true for more than a decade and look where we are. I only hope that it changes because I want a balanced market to increase innovation and decrease prices.

Ok, actually, you have to break perf between raster and RT as the 9070 XT is faster than the 7900 XTX in raster.

I think that you meant to say that the RX 9070 XT is faster than the RX 7900 XTX in RT, not raster. The RX 7900 XTX is still the king of the Radeon hill when it comes to raster performance, being 10% faster than the RX 9070 XT on average. When it comes to RT, the RX 9070 XT leaves the RX 7900 XTX in its dust. I own an RX 7900 XTX and I'm perfectly ok with its (lack of) RT performance because don't care about RT, like, at all. I'm also perfectly happy with FSR3.x. I haven't had to use it yet but I still tried it out just to see what it's like and it's more than good enough for me if I ever have to use it. I tried it out in Starfield and had actually played it for a week with FSR enabled without realising it. That tells me it's definitely good enough for my purposes.

That's pretty huge, and that's what AMD was promoting in their marketing (that RT perf was significantly better relative to raster and last gen).

Yep, that is pretty huge. If you care about RT that is. I don't care about it at all but I know that a lot of people do and there are a good number of people who bought GeForce specifically because Radeon wasn't good enough at RT. That's who the RX 9000-series is aimed at.

Still, I felt like AMD was more honest this time around than with their RDNA3 launch.

Yes they were, but they weren't as honest this time around as they were with their RDNA(1) and RDNA2 launches.

 

[systemBuilder_49]

Paper Launch. 90% of people paying less than $750 seems to have had their order cancelled. Gee THANKS AMD !!! THANKS FOR 3 MONTHS OF NONSTOP GASLIGHTING !!!

 

[Crazyy8]

Unfortunately, the damn bots got all the MSRP online stock (if there ever was some).

Those close to a Best Buy or Micro Center could've probably got in line when the store opened and still got one. The Micro Center I went to (New Jersey, USA) had like 300+ in stock. I think at least half of those were being sold at MSRP.

It looks like, going forward, you'll need to be at a brick-n-mortar store for any chance of getting an MSRP card, at MSRP. I can't blame AMD for this though. I mean, 300+ cards on launch day is pretty damn good!

That ain't my problem, my micro center has around 40 in stock, it's that all of them are $250 or more above MSRP. At that point, it ain't super competitive anymore(I know it beats the 5070 ti, which is at the same price, sometimes but that ain't the point).

 

[Gururu]

Not sure if this has been posted yet.

https://videocardz.com/newz/retaile...o-first-shipments-price-set-to-increase-later

That ain't my problem, my micro center has around 40 in stock, it's that all of them are $250 or more above MSRP. At that point, it ain't super competitive anymore(I know it beats the 5070 ti, which is at the same price, sometimes but that ain't the point).

Yes in MD Microcenter there are tons (50-100 cards) of 9070s starting at $620. All the XTs are gone.

 

[SyCoREAPER]

If I wasn't doing something special with my 4090 and buying said parts today, I would have probably caved and replaced my 7900XT with one. Then again my 7900XT is doing exactly what it needs to