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

[JarredWaltonGPU]

There are lies, damn lies and benchmarks.
All that can be done is to look for how your individual use case is covered across the benchmarks, not just TH, not just GN or HUB but across the spectrum of reviewers.

With that in mind, I lament the loss of written word sites. I prefer to not take notes but to flick between tabs on Firefox.

I hear that, and not just because I write for a living. It's so much easier to find the data I'm looking for in a written review than in an video — especially when the video pauses to play an ad if you skip around too much.

 

[King_V]

With that in mind, I lament the loss of written word sites. I prefer to not take notes but to flick between tabs on Firefox.

I hear that, and not just because I write for a living. It's so much easier to find the data I'm looking for in a written review than in an video — especially when the video pauses to play an ad if you skip around too much.

Absolutely 100% this. Sure, sometimes I don't mind watching videos, but typically, I like to just get a hold of the numbers really quickly, and that's way easier to do for written articles. Can't exactly do a CTRL-F to search in a video.

 

[thestryker]

With that in mind, I lament the loss of written word sites. I prefer to not take notes but to flick between tabs on Firefox.

This is why I'm appreciative of the fact that HUB reviews end up on Techspot in written form some time after the video. GN also posts some of their reviews in their website after the videos have been up for a while.

 

[KyaraM]

Technically, I think it's important to note that I show the 5070 Ti as faster than the 9070 XT in all four of the aggregate rasterization charts. The 9070 meanwhile is faster than the 5070 in all four aggregate rasterization charts. But I guess KyaraM was looking at the 4070 Ti and 4070... which... why?

And by that, I mean why is that the main comparison? You can't really buy 4070 class cards at a reasonable price now, so while the comparison point is important on some level, it's not my primary comparison. New AMD vs old Nvidia? Just like I wouldn't compare new Nvidia vs old AMD as the main item. New AMD vs old AMD makes more sense, or new Nvidia vs old Nvidia.

Anyway, I know with brand-new, up-to-date test results (I'm not relying on numbers run sometime last year or even last month), the 9070 XT ends up quite a bit faster than a 4070 Ti Super. The 9070 ends up just barely faster than a 4070 Ti — 2% faster at 1080p medium and 1440p ultra, and 1% slower at 1080p ultra, but then 6% faster at 4K ultra. That makes more sense considering 12GB vs 16GB.

I don't have really any old and undemanding games in my current test suite. Flight Simulator 2020 definitely hits CPU limits on these cards, and Baldur's Gate 3 is pretty tame in GPU requirements as well. Just about everything else ends up fully GPU limited, and that will factor into the overall rankings.

Is it really so unreasonable for a 4070Ti owner to look at how new cards stack up to yours? I looked at the 5070Ti the same way, even if you didn't get to see it because the info has been out a while and this topic is about the AMD cards.

I think quite a few owners of previous cards are interested in such benchmarks for that very reason, comparing them to their own cards, since they also give a prediction of where the journey goes so to speak, and an estimate for how long your own card might stay relevant. Besides, those cards were previously compared to the 4070s, and based on that prediction people expected quite a bit. So please don't blame anyone for measuring up previous news and comparisons to reality. Or for making purchase decisions based on wether or not the performance gain would be substantial enough. After all, I was under the impression that was the point for including these older cards in the first place.

 

[JarredWaltonGPU]

Is it really so unreasonable for a 4070Ti owner to look at how new cards stack up to yours? I looked at the 5070Ti the same way, even if you didn't get to see it because the info has been out a while and this topic is about the AMD cards.

I think quite a few owners of previous cards are interested in such benchmarks for that very reason, comparing them to their own cards, since they also give a prediction of where the journey goes so to speak, and an estimate for how long your own card might stay relevant. Besides, those cards were previously compared to the 4070s, and based on that prediction people expected quite a bit. So please don't blame anyone for measuring up previous news and comparisons to reality. Or for making purchase decisions based on wether or not the performance gain would be substantial enough. After all, I was under the impression that was the point for including these older cards in the first place.

If you have a 4070 Ti, then who cares about 4070 Ti Super? That's why the comparison struck me as weird. You're not talking about one prior gen comparison point but instead referenced two. And I get that those comparisons are useful on some level, but really the 9070-series needs to go up against Nvidia's 50-series counterparts.

But you're missing the main point of my post: I tested newer games, all new testing results, nothing from old testing. That can and will make a difference. CS2 is a great example of this. It's a very lightweight game running at hundreds of FPS. No one would really notice the difference between 300 and 400 FPS, but it can skew benchmark data. I don't test that sort of game for a reason.

 

[KyaraM]

If you have a 4070 Ti, then who cares about 4070 Ti Super? That's why the comparison struck me as weird. You're not talking about one prior gen comparison point but instead referenced two. And I get that those comparisons are useful on some level, but really the 9070-series needs to go up against Nvidia's 50-series counterparts.

But you're missing the main point of my post: I tested newer games, all new testing results, nothing from old testing. That can and will make a difference. CS2 is a great example of this. It's a very lightweight game running at hundreds of FPS. No one would really notice the difference between 300 and 400 FPS, but it can skew benchmark data. I don't test that sort of game for a reason.

People expected 4080 Super performance from the 9070XT, yet it doesn't really get there. They expected that due to news posted on THIS website. I made an observation to the contrary, plain and simple. I see nothing wrong with that.

And I honestly don't know where I missed anything; or rather, I think we speak past each other this whole time. All I did was point out that it makes a difference for benchmark results. I didn't intend to offend or tell you to use other benchmarks, just that it makes a difference. That wasn'tdirected at you, but in general at people. If you picked a different suite of newer games, like some of those GN (Dragon's Dogma and Veilguard, for example) used, the result would also be different, which I also mentioned. I made observations. That's the long and the short of it.

Also, I think it best to drop this here since it's a fruitless discussion leading nowhere. I also don't have the mental capacity with everything going on at my end.

 

[JarredWaltonGPU]

People expected 4080 Super performance from the 9070XT, yet it doesn't really get there. They expected that due to news posted on THIS website. I made an observation to the contrary, plain and simple. I see nothing wrong with that.

And I honestly don't know where I missed anything; or rather, I think we speak past each other this whole time. All I did was point out that it makes a difference for benchmark results. I didn't intend to offend or tell you to use other benchmarks, just that it makes a difference. That wasn'tdirected at you, but in general at people. If you picked a different suite of newer games, like some of those GN (Dragon's Dogma and Veilguard, for example) used, the result would also be different, which I also mentioned. I made observations. That's the long and the short of it.

Also, I think it best to drop this here since it's a fruitless discussion leading nowhere. I also don't have the mental capacity with everything going on at my end.

We posted rumors of how fast RX 9070 XT was supposed to be -- no one had real data. Or else someone used a GeekBench result as potentially useful data. I really despise that and try not to let those things slip past without heavy caveats, but it does happen from the news team. Because, unfortunately, people click through to read that sort of news. It's a no win situation.

A year ago, I think a lot of people might have assumed the 9070 XT (then unnamed) could match the 4080 / 4080 Super. But a year ago, most of us probably also hoped the 5080 would be 40% faster than the 4080. This is why the "grain of salt" thing is laughably overused. "Take this early results with a pinch/grain/spoonful of salt" is the horrible refrain.

There are certain workloads, I'm sure, where a 9070 XT can match or maybe even beat a 4080 Super. But in general? No, raw specs alone tell us it shouldn't be faster. And I know paper specs aren't the best way to compare, but with correct insight they can tell a lot about what to expect. Fundamentally:

9070 XT: 48.66 TFLOPS FP32, double that for FP16, and 16GB with 640 GB/s of bandwidth
4080 Super: 52.22 TFLOPS FP32, 209 for FP16 (double with sparsity), and 736 GB/s of bandwidth

FP32 TFLOPS is close, but traditionally Nvidia gets more benefit per TFLOPS. So AMD would need to radically change its architecture to close the gap and beat a 4080 Super, and while it did change, I don't think it was enough to expect a win there. Basically, it's roughly on par (slightly faster) than a 4070 Ti Super overall, but 10% slower than a 4080 Super, which would be 7% slower than a 4080. That's really good overall I think.

Shame about all the retail prices, though!

 

[AnandHulk]

This is what it looks like when business is so good you can have fun with it. Think Blockbuster in the '90's, or IBM in the early '80's.
You can sell your goods at MSRP out of a food truck and people will consider themselves "lucky" to pay full price and get one.
Be careful nVidia...times can change.

 

[King_V]

This is what it looks like when business is so good you can have fun with it. Think Blockbuster in the '90's, or IBM in the early '80's.
You can sell your goods at MSRP out of a food truck and people will consider themselves "lucky" to pay full price and get one.
Be careful nVidia...times can change.

I'm guessing you meant to post that in this thread?

 

[AnandHulk]

I'm guessing you meant to post that in this thread?

Oops! Sorry about that!