Question 4070ti vs 7900xt

[Cableman]

It's time for me to upgrade my aging GTX1070 and I am trying to decide between the 4070ti and the 7900xt. The card will be bought through my employer ($900 is my hard limit) so those are my two options. I have a 4k 144hz monitor that I haven't been able to drive properly with the 1070.

I read and watched all the reviews I could find. The cards are basically the same price. The 7900xt beats the 4070ti in raster pretty much in every game. The weakness of the 7900xt seems to be RT, but it doesn't seem to be as bad as I expected. There are even games where it beats the 4070ti even with RT on. The 2 worst cases I saw were Control and Cyberpunk 2077. The one variable I'm not sure about is DLSS vs FSR - do we expect FSR to gain similar traction to DLSS? Is the 7900xt missing any features?

So overall, the two are priced the same (in practice), the 7900xt is faster without RT and just a bit slower with RT on. Is there a reason to consider the 4070ti over the 7900xt?

 

[Stuka87]

you should look at 1080p data if you want CPU vs CPU comparisons (that I recited from memory), this isn't good for that, because 4K puts the pressure elsewhere.

Yeah, but the point was that the CPU is less important at 4K. And if somebody is buying a GPU that cost over $1000, chances are they are gaming at 4K. The CPU is rarely the bottleneck at 4K. And thus, DLSS 3 is kind of worthless for these pricy GPUs as it really only works for CPU bound use cases where the GPU has extra cycles that arent doing anything. If the GPU is pegged out, DLSS 3 won't add much.

 

[Cableman]

Yeah, but the point was that the CPU is less important at 4K. And if somebody is buying a GPU that cost over $1000, chances are they are gaming at 4K. The CPU is rarely the bottleneck at 4K. And thus, DLSS 3 is kind of worthless for these pricy GPUs as it really only works for CPU bound use cases where the GPU has extra cycles that arent doing anything. If the GPU is pegged out, DLSS 3 won't add much.

I agree 100% with you. I find the whole discussion of bottlenecks to be very academic and very disconnected from the real world, besides some weird edge cases.

No one buys a current get GPU to game at 1080p, it's completely nonsensical. Even my 1070 is ok for 1080p. What is the point showing CPU "bottlenecks" with something like a 4090 at 1080p? Firstly, no one is using a 4090 to play at that resolution and the "bottleneck" is what exactly - you are "only" getting 200 frames (or whatever) instead of 230 (or whatever). For someone to both be using a 4090 to play at 1080p AND to care about losing a few FPS when most monitors are beyond maxed with the frames they are getting is very niche.

I challenge anyone to give an example of a CPU bottleneck that is realistic. And no, playing Cyberpunk 2077 with an Athlon II and a 4090 at 240p is not realistic.

 

[Cableman]

Yeah, but the point was that the CPU is less important at 4K. And if somebody is buying a GPU that cost over $1000, chances are they are gaming at 4K. The CPU is rarely the bottleneck at 4K. And thus, DLSS 3 is kind of worthless for these pricy GPUs as it really only works for CPU bound use cases where the GPU has extra cycles that arent doing anything. If the GPU is pegged out, DLSS 3 won't add much.

I agree 100% with you. I find the whole discussion of bottlenecks to be very academic and very disconnected from the real world, besides some weird edge cases.

No one buys a current get GPU to game at 1080p, it's completely nonsensical. Even my 1070 is ok for 1080p. What is the point showing CPU "bottlenecks" with something like a 4090 at 1080p? Firstly, no one is using a 4090 to play at that resolution and the "bottleneck" is what exactly - you are "only" getting 200 frames (or whatever) instead of 230 (or whatever). For someone to both be using a 4090 to play at 1080p AND to care about losing a few FPS when most monitors are beyond maxed with the frames they are getting is very niche.

I challenge anyone to give an example of a CPU bottleneck that is realistic. And no, playing Cyberpunk 2077 with an Athlon II and a 4090 at 240p is not realistic.

 

[Heartbreaker]

I challenge anyone to give an example of a CPU bottleneck that is realistic. And no, playing Cyberpunk 2077 with an Athlon II and a 4090 at 240p is not realistic.

I'm in total agreement about CPU gaming bottlenecks not really being a thing, in the real world.

I find it kind of hilarious, how the enthusiast CPU market seems to hinge entirely on gaming performance. When AMD Zen 3 finally beat Intel at gaming, they launched 6 core CPU at $300 about 50% higher than previous releases, just because they eeked out a tiny edge on 1080p gaming, with a top end monster GPU.

Then Intel pushed back with it's next gen to eek back a tiny 1080p gaming win, then AMD added 3D cache... It's all about gaming, when in reality none of these "wins" would be detectable except by turning on an FPS counter.

There can always be edge cases.

It seems like the main edge case mentioned lately is MS Flight Sim. IIRC, it didn't really respond to more cores though. Don't know if it was patched with better multi-core support.

 

[Khanan]

I agree 100% with you. I find the whole discussion of bottlenecks to be very academic and very disconnected from the real world, besides some weird edge cases.

No one buys a current get GPU to game at 1080p, it's completely nonsensical. Even my 1070 is ok for 1080p. What is the point showing CPU "bottlenecks" with something like a 4090 at 1080p? Firstly, no one is using a 4090 to play at that resolution and the "bottleneck" is what exactly - you are "only" getting 200 frames (or whatever) instead of 230 (or whatever). For someone to both be using a 4090 to play at 1080p AND to care about losing a few FPS when most monitors are beyond maxed with the frames they are getting is very niche.

I challenge anyone to give an example of a CPU bottleneck that is realistic. And no, playing Cyberpunk 2077 with an Athlon II and a 4090 at 240p is not realistic.

But it’s not. With a proper cpu test at low res you can find out which CPU is faster in the long term. It’s pretty important. Also the faster CPUs always have better 1% lows. And who pairs a high end GPU with something like a 10100? Yea right, nobody. That’s more academic than CPU tests at 1080p low.

 

[Zepp]

No one buys a current get GPU to game at 1080p, it's completely nonsensical.

I agree with your overall point I just wanted to add there is a surprising amount of GPU buyers who are clueless about much beyond the hype. They buy prebuilt machines and are only techie enough to swap the GPU. If you ask what resolution they play at they'll say "I dunno, it's 27" so they may get hyped into buying the latest mid-range or high end GPU just to play at 1080p/60Hz now with RTX!

 

[Aapje]

It seems like the main edge case mentioned lately is MS Flight Sim. IIRC, it didn't really respond to more cores though. Don't know if it was patched with better multi-core support.

Simulator games tend to have a heavy load on one core. This is not necessarily something that you can just change easily or at all.

 

[MrTeal]

It's not even just low res "720p CPU Tests" that can show differences. The linked Techpowerup review of the 7600 shows even perfectly competent CPUs like the 5600X significantly down in some games at 1440p using what is now a middle of the pack GPU like a 3080. In Far Cry 6 the 5600X got 95.8FPS vs the 123.8FPS stock on a 7600. It's certainly not unplayable like that, but a 30% difference between CPUs is no different than moving up a whole class in GPUs. FC6 FPS doesn't really change between running at 720p up to 1440p FWIW, the game is pretty much entirely CPU bound with a 3080 up until you get to 4k. Spending more on a better GPU will certainly bring much more consistent performance increases, but it's being willfully ignorant to say there's not games out there that would show notable differences between even good modern CPUs.

 

[Khanan]

It's not even just low res "720p CPU Tests" that can show differences. The linked Techpowerup review of the 7600 shows even perfectly competent CPUs like the 5600X significantly down in some games at 1440p using what is now a middle of the pack GPU like a 3080. In Far Cry 6 the 5600X got 95.8FPS vs the 123.8FPS stock on a 7600. It's certainly not unplayable like that, but a 30% difference between CPUs is no different than moving up a whole class in GPUs. FC6 FPS doesn't really change between running at 720p up to 1440p FWIW, the game is pretty much entirely CPU bound with a 3080 up until you get to 4k. Spending more on a better GPU will certainly bring much more consistent performance increases, but it's being willfully ignorant to say there's not games out there that would show notable differences between even good modern CPUs.

That’s a worst case scenario for the 5600X but shows well that average graphs are unimportant if you happen to play that one game that is super cpu heavy and will tank your fps. I wanted to say it earlier, averages of all games combined do not tell the full picture of the story.

 

[IEC]

It's not even just low res "720p CPU Tests" that can show differences. The linked Techpowerup review of the 7600 shows even perfectly competent CPUs like the 5600X significantly down in some games at 1440p using what is now a middle of the pack GPU like a 3080. In Far Cry 6 the 5600X got 95.8FPS vs the 123.8FPS stock on a 7600. It's certainly not unplayable like that, but a 30% difference between CPUs is no different than moving up a whole class in GPUs. FC6 FPS doesn't really change between running at 720p up to 1440p FWIW, the game is pretty much entirely CPU bound with a 3080 up until you get to 4k. Spending more on a better GPU will certainly bring much more consistent performance increases, but it's being willfully ignorant to say there's not games out there that would show notable differences between even good modern CPUs.

Exactly my experience upgrading from a Zen 3 chip to my Ryzen 7700X targeting 240Hz 1440p.

My *average* FPS became my new 0.1% low FPS. Averages are much higher. And that's with a last gen card.

Newest gen cards show CPU bottlenecks even more easily.

 

[Khanan]

Exactly my experience upgrading from a Zen 3 chip to my Ryzen 7700X targeting 240Hz 1440p.

My *average* FPS became my new 0.1% low FPS. Averages are much higher. And that's with a last gen card.

Newest gen cards show CPU bottlenecks even more easily.

Was even more crass for me when I upgraded from 3700X to 5800X3D. So much smoother. I think Zen 3 is still fine, but Zen 2 is clearly a step behind

 

[Heartbreaker]

It's not even just low res "720p CPU Tests" that can show differences. The linked Techpowerup review of the 7600 shows even perfectly competent CPUs like the 5600X significantly down in some games at 1440p using what is now a middle of the pack GPU like a 3080. In Far Cry 6 the 5600X got 95.8FPS vs the 123.8FPS stock on a 7600. It's certainly not unplayable like that, but a 30% difference between CPUs is no different than moving up a whole class in GPUs. FC6 FPS doesn't really change between running at 720p up to 1440p FWIW, the game is pretty much entirely CPU bound with a 3080 up until you get to 4k. Spending more on a better GPU will certainly bring much more consistent performance increases, but it's being willfully ignorant to say there's not games out there that would show notable differences between even good modern CPUs.

As I just finished say above. There are edge cases. This is one. I don't think anyone said no such games exist.

It appears again, that the game is not well multi-threaded, so is dependent on how much work a single core can do, so clock speed and IPC become dominant. Newer architectures have more of both.

So yeah, in 2023, you make a game CPU bottleneck if your game engine is still heavily reliant on one or two single threaded loops.

 

[Stuka87]

Was even more crass for me when I upgraded from 3700X to 5800X3D. So much smoother. I think Zen 3 is still fine, but Zen 2 is clearly a step behind

I did the exact same upgrade, and it was a noticeable change. Even with my older GPU.

 

[MrTeal]

As I just finished say above. There are edge cases. This is one. I don't think anyone said no such games exist.

You said that, after opening your comment with

I'm in total agreement about CPU gaming bottlenecks not really being a thing, in the real world.

You can call MSFS and Far Cry 6 edge cases, or that big differences in CS:GO don't matter because FPS is ridiculous already, or that you won't notice FPS differences in a game like Civ6, but it's a bit disingenuous to say that CPU bottlenecks don't exist in the real world, except for all those cases where they do that I'm just going to ignore.

MSFS and FC6 may not be true scotsmen but they certainly are well received recent releases in major gaming franchises.

 

[Aapje]

There are also people that play nothing but MSFS.