Current Generation Games Analysis Technical Discussion [2022] [XBSX|S, PS5, PC]

[snc]

A 3090 is 3x the power of the consoles yet it fails to yield 3x the performance.

Only if you looking on teraflops which defenitly inflate in terms of performance/teraflop on rtx3xxx (metrics that isn't very important to be honest)
https://www.techpowerup.com/review/zotac-geforce-rtx-3090-trinity/32.html
rtx 3090 is "only" around 2x faster in 4k than rtx2070super that is on similar level vs consoles (tough advantage increase when rt enabled)

 

[BitByte]

A 3090 is 3x the power of the consoles yet it fails to yield 3x the performance. I’ve tested the demo on my pc with a 3080 and 5900x and it was awful. When I say that pc hardware is performing worse that consoles, I’m talking about expected vs actual performance. The consoles have mobile zen 2 cpus and modified rx6600s. If we use that as a baseline, then it is normal to expect a significant increase in performance when using hardware that is significantly more powerful. The only time I get acceptable performance is when all the settings are at 2. Then it yields close to 2x the performance with significantly worse visuals. Finally, keep in mind that Nvidia’s RT hardware is more performant than Amds yet we’re getting middling results.

Only if you looking on teraflops which defenitly inflate in terms of performance/teraflop on rtx3xxx (metrics that isn't very important to be honest)
https://www.techpowerup.com/review/zotac-geforce-rtx-3090-trinity/32.html
rtx 3090 is "only" around 2x faster in 4k than rtx2070super that is on similar level vs consoles (tough advantage increase when rt enabled)

Isn't that chart referring to raster performance? In ray-tracing, I'm quite sure that the 3090 is 3x faster than the 2070super or I may just be misremembering.

 

[Seanspeed]

Isn't that chart referring to raster performance? In ray-tracing, I'm quite sure that the 3090 is 3x faster than the 2070super or I may just be misremembering.

Might be 3x faster than like a 6700XT in ray tracing or something, but definitely not a 2070S.

Ampere's improvements in ray tracing over Turing were not that dramatic in general.

 

[snc]

Isn't that chart referring to raster performance? In ray-tracing, I'm quite sure that the 3090 is 3x faster than the 2070super or I may just be misremembering.

Yeah raster, in rt could be closer to 3x over consoles indeed

 

[PSman1700]

A 3090 is 3x the power of the consoles yet it fails to yield 3x the performance. I’ve tested the demo on my pc with a 3080 and 5900x and it was awful. When I say that pc hardware is performing worse that consoles, I’m talking about expected vs actual performance. The consoles have mobile zen 2 cpus and modified rx6600s. If we use that as a baseline, then it is normal to expect a significant increase in performance when using hardware that is significantly more powerful. The only time I get acceptable performance is when all the settings are at 2. Then it yields close to 2x the performance with significantly worse visuals. Finally, keep in mind that Nvidia’s RT hardware is more performant than Amds yet we’re getting middling results.

Edit: When this demo initially came out on console, I was very vocal about it being nothing special due to its awful performance and my opinion hasn’t changed. There is something really wrong with this demo and ue5 imo. It badly fails to utilize hardware. GPU utilization is often low and cpu utilization is also not the best. Combine in the awful shader compilation stutters and frankly, I can’t say I’m impressed. Visually, it looks good and it has high quality assets but that’s irrelevant to me if it fails to utilize hardware properly.

Not at the same settings, as digital foundry already pointed out. Also, unpolished relatively compared to the console versions. Early days, engines running bad on all platforms, we cant conclude hardware performance based on a tech demo and a developer sample.

 

[chris1515]

Isn't NX specifically talking about asset decompression being handled by the additional logic blocks/coprocessors found in the consoles SSD controller or I/O complex, and not necessarily speaking about raw SSD speeds/bandwidth, but more so about freeing up CPU resources for other things other than asset decompression needs? In theory by doing so, the consoles could have an advantage CPU wise in like-for-like scenarios where both console and PC CPUs have similar matching clock speeds and core counts (maybe performing better in single threaded cases such as The Matrix Awakens demo).

What he is saying is not clear and another things I don't like is comparing PC with unlocked framerate and consoles capped at 30 fps and without the same settings. I think this is again not a good comparison. I prefer again the Digitalfoundry video.

For sur PC CPU has more work to do from an I/O point of view but this is maybe not the reason of what he is seeing. I did not see the full video because there is too much approximation.

EDIT: He talk about things only developer with devkit of consoles and developing the demo can know. This is what he always does.

 

[cwjs]

EDIT: He talk about things only developer with devkit of consoles and developing the demo can know. This is what he always does.

dude just randomly guesses

 

[Globalisateur]

A 3090 is 3x the power of the consoles yet it fails to yield 3x the performance. I’ve tested the demo on my pc with a 3080 and 5900x and it was awful. When I say that pc hardware is performing worse that consoles, I’m talking about expected vs actual performance. The consoles have mobile zen 2 cpus and modified rx6600s. If we use that as a baseline, then it is normal to expect a significant increase in performance when using hardware that is significantly more powerful. The only time I get acceptable performance is when all the settings are at 2. Then it yields close to 2x the performance with significantly worse visuals. Finally, keep in mind that Nvidia’s RT hardware is more performant than Amds yet we’re getting middling results.

Edit: When this demo initially came out on console, I was very vocal about it being nothing special due to its awful performance and my opinion hasn’t changed. There is something really wrong with this demo and ue5 imo. It badly fails to utilize hardware. GPU utilization is often low and cpu utilization is also not the best. Combine in the awful shader compilation stutters and frankly, I can’t say I’m impressed. Visually, it looks good and it has high quality assets but that’s irrelevant to me if it fails to utilize hardware properly.

Or the hardware is simply inadequat for the task. Outdated. Why do you think Nvidia is putting effort into GPU decompression (hw and sw) and such? Because they understand (like they understood the need for hardware RT) that the next step is not simply more CPU or more GPU power. It's how can you stream data in and set it up quickly enough to feed the GPU memory all ready to render the scene. Unfortunatly the PC is currently an antique hardware with tons of bottlenecks whether hardware or APIs. The CPU and GPU in PCs are working wonders when the scene is all compiled and there is no need to stream in new high quality assets.

But otherwise you get UE5 demo running on dinosaurs PCs costing a leg that get handily outperformed by mobile CPU and GPU running on a closed box with modern and efficient APIs.

 

[cwjs]

Why do you think Nvidia is putting effort into GPU decompression (hw and sw) and such? Because they understand (like they understood the need for hardware RT) that the next step is not simply more CPU or more GPU power.

Sure, that's true enough as a general statement. What does that have to do with how ue5 is bottlenecked? What does that do with how ue5 works?

 

[zed]

There is something really wrong with this demo and ue5 imo. It badly fails to utilize hardware. GPU utilization is often low and cpu utilization is also not the best. Combine in the awful shader compilation stutters and frankly, I can’t say I’m impressed. Visually, it looks good and it has high quality assets but that’s irrelevant to me if it fails to utilize hardware properly.

UE5 Visually is the best looking realtime thing ever made by leaps and bounds.
OK performance does suck and prolly not acceptable for games ATM, but like you say if its not utilizing GPU/CPU fully then there is quite a bit of room for optimization, in 2 ways. Better utilize the CPU/GPU fully and use the power they are using already using better (better algorithms) So there is quite good hope they will be able to hit 60fps in time.
Though of course they cant use too much of the CPU as you want to keep enuf for AI/gamecode etc

 

[Dictator]

Regarding 3090 vis-a-vis the consoles - I found that it tends to be 2x performance in games without ray tracing, and around 3x performance in games with.

 

[PSman1700]

Regarding 3090 vis-a-vis the consoles - I found that it tends to be 2x performance in games without ray tracing, and around 3x performance in games with.

Which is probably what the other user ment. Games targetting next gen without RT are going to be close to zero.

 

[PSman1700]

Or the hardware is simply inadequat for the task. Outdated. Why do you think Nvidia is putting effort into GPU decompression (hw and sw) and such? Because they understand (like they understood the need for hardware RT) that the next step is not simply more CPU or more GPU power. It's how can you stream data in and set it up quickly enough to feed the GPU memory all ready to render the scene. Unfortunatly the PC is currently an antique hardware with tons of bottlenecks whether hardware or APIs. The CPU and GPU in PCs are working wonders when the scene is all compiled and there is no need to stream in new high quality assets.

But otherwise you get UE5 demo running on dinosaurs PCs costing a leg that get handily outperformed by mobile CPU and GPU running on a closed box with modern and efficient APIs.

So far from reality lol.

 

[see colon]

Not at the same settings, as digital foundry already pointed out. Also, unpolished relatively compared to the console versions. Early days, engines running bad on all platforms, we cant conclude hardware performance based on a tech demo and a developer sample.

Every video I watch where some "tech guy" on the internet downloads and compiles the city demo and then complains that it runs poorly on their hardware compared to the consoles has somehow completely blocked from their mind the fact that the Matrix demo that was shown on consoles was tuned at a fairly granular level to reach that performance. I remember the sentiment (not here, but elsewhere) that the Coalition had helped optimize the demo having some sort of negative connotation toward the performance of Xbox hardware, but I think it should be clear now that everything is going to struggle a bit with this content, unless you spend some time getting it right. Epic didn't just click the compile button and send the demo off, and it shows when comparing to the sample city on PC.

 

[davis.anthony]

Every video I watch where some "tech guy" on the internet downloads and compiles the city demo and then complains that it runs poorly on their hardware compared to the consoles has somehow completely blocked from their mind the fact that the Matrix demo that was shown on consoles was tuned at a fairly granular level to reach that performance.

They also seem to forget that PS5/XSX are (from a performance point of view) in a much better place vs PC than PS4/OneS was at release.

If the consoles are 1080p30 then even an RTX3090 isn't enough for 1440p/60 without dropping settings.

And I think that's where all the complaining is coming from, PC gamers have been so used to running current games built with the 1.2Tflop One S in mind at silly resolutions and frame rates that they expect that to carry on now even with the base line quadrupling.

Only for it not to be happening and now their complaining about it, I saw one guy on Twitter ranting that his RTX2060 wasn't performing as well as PS5/XSX was and was demanding an answer from Nvidia.

 

[snc]

Every video I watch where some "tech guy" on the internet downloads and compiles the city demo and then complains that it runs poorly on their hardware compared to the consoles has somehow completely blocked from their mind the fact that the Matrix demo that was shown on consoles was tuned at a fairly granular level to reach that performance. I remember the sentiment (not here, but elsewhere) that the Coalition had helped optimize the demo having some sort of negative connotation toward the performance of Xbox hardware, but I think it should be clear now that everything is going to struggle a bit with this content, unless you spend some time getting it right. Epic didn't just click the compile button and send the demo off, and it shows when comparing to the sample city on PC.

Interesting that when demo arrived I read here we shouldn't read too much into benchmarks results as demo is not very optimized on consoles, now we cant read benchamarks results too much because apparently demo is only optimized on consoles ;d

 

[iroboto]

Interesting that when demo arrived I read here we shouldn't read too much into benchmarks results as demo is not very optimized on consoles, now we cant read benchamarks results too much because apparently demo is only optimized on consoles ;d

neither is representative of actual game performance and even the console versions were done under a tight deadline, to be fair they'd probably perform even better 3-4 years from now as the engine matures.
It's still sound advice overall to not judge things by this demo whether it represents good or bad performance.

my general take away here, is that perhaps Epic has indirectly implied that UE5 has catered for movies/cinematics/TV shows (hence we designed it around 30fps). And that the power of this engine is able to see close to near-baked final renders on something as powerful as a PS5.

But the changes made to UE5 may not necessarily be gaming focused.

I'm not sure if you participated in a previous discussion around Epic vs Unity, but I remember distinctly discussing Unity's push for the Entity Component / Job system (i think it's called DOTS now) vs Epic's push for Lumen and Nanite. And back then we discussed that UE5 would be completely hampered by single threaded performance and Unity would fly. And it came true. The development time was invested to create these amazing visuals, that would be very handy for cinematic artists, but not necessarily great for dealing with massively parallel processing of lots of calls/objects.

 

[chris1515]

We need to wait what the coalition do with the engine, they said they will continue to target 60 fps and they have some idea how to reach it on UE5. The demo was designed to run at 30 fps, it doesn't mean we won't see 60 fps game on UE 5.

And I suppose performance will improve too with every new version of UE 5 like with UE4.

 

[Silent_Buddha]

Every video I watch where some "tech guy" on the internet downloads and compiles the city demo and then complains that it runs poorly on their hardware compared to the consoles has somehow completely blocked from their mind the fact that the Matrix demo that was shown on consoles was tuned at a fairly granular level to reach that performance. I remember the sentiment (not here, but elsewhere) that the Coalition had helped optimize the demo having some sort of negative connotation toward the performance of Xbox hardware, but I think it should be clear now that everything is going to struggle a bit with this content, unless you spend some time getting it right. Epic didn't just click the compile button and send the demo off, and it shows when comparing to the sample city on PC.

Hell, that isn't even anything new to UE5. People continually fail to realize that a LOT of optimization must happen at the engine level regardless of the engine used (UE, UE2, UE3, UE4, UE5, Unity, Frostbite, whatever) before a title ships.

Demo's or vertical slices of "gameplay" from an title in development may have some level of optimization but rarely will it be as fully optimized as a shipping game title. Of course, there also exists shipping titles that aren't well optimized due to time, budget or technical expertise of the engineers working on the project's engine. Even well optimized shipping titles are often not "fully" optimized as you'll see further optimizations in future patches if budget allows.

It makes comparisons difficult and fraught with numerous pitfalls if you are trying to use them to somehow discern the capabilities of various pieces of hardware, each of which may or may not be as optimized for in code as another piece of hardware. And when it comes to comparing a demo that was prepared with a set target in mind by software engineers compared to just random people compiling a project with almost zero knowledge of what they are actually doing? Yeah...

Regards,
SB

 

[pjbliverpool]

Isn't NX specifically talking about asset decompression being handled by the additional logic blocks/coprocessors found in the consoles SSD controller or I/O complex, and not necessarily speaking about raw SSD speeds/bandwidth, but more so about freeing up CPU resources for other things other than asset decompression needs? In theory by doing so, the consoles could have an advantage CPU wise in like-for-like scenarios where both console and PC CPUs have similar matching clock speeds and core counts (maybe performing better in single threaded cases such as The Matrix Awakens demo).

I think this would certainly be true if you're CPU bound with full CPU utilisation, but I'd be surprised given the demo's seeming reliance on single threaded (or a small number of threads) performance if they were putting the decompression onto one of those already bottlenecked threads. From what we've been told the streaming requirement in the demo is pretty modest, so the decompression requirements shouldn't be significant. It's likely they could easily farm that off to one of the less utilised threads so that it doesn't form part of the bottleneck.

My son's PC which has good specs (slightly above PS5/X in terms of specs) runs the demo slightly worst than my PS5.

But at the same settings? It does seem the PC is heavier hit on the CPU front at least compared to console but I think it's quite likely this is simply down to the reality of optimised demo vs user compiled project straight from the editor. That's got to count for a fair bit of performance.

A 3090 is 3x the power of the consoles yet it fails to yield 3x the performance. I’ve tested the demo on my pc with a 3080 and 5900x and it was awful. When I say that pc hardware is performing worse that consoles, I’m talking about expected vs actual performance. The consoles have mobile zen 2 cpus and modified rx6600s. If we use that as a baseline, then it is normal to expect a significant increase in performance when using hardware that is significantly more powerful. The only time I get acceptable performance is when all the settings are at 2. Then it yields close to 2x the performance with significantly worse visuals. Finally, keep in mind that Nvidia’s RT hardware is more performant than Amds yet we’re getting middling results.

Edit: When this demo initially came out on console, I was very vocal about it being nothing special due to its awful performance and my opinion hasn’t changed. There is something really wrong with this demo and ue5 imo. It badly fails to utilize hardware. GPU utilization is often low and cpu utilization is also not the best. Combine in the awful shader compilation stutters and frankly, I can’t say I’m impressed. Visually, it looks good and it has high quality assets but that’s irrelevant to me if it fails to utilize hardware properly.

I'm not sure if the 3090/RT comparison is really relevant here since GPU performance seems to be massively held back by CPU performance making any GPU performance comparison impossible. As you note though, that hardware always nets you much better performance in every other game. So surely the simplest explanation is the most likely one - because on PC it's not an optimised demo, it's a user compiled project straight from the editor.

Or the hardware is simply inadequat for the task. Outdated. Why do you think Nvidia is putting effort into GPU decompression (hw and sw) and such? Because they understand (like they understood the need for hardware RT) that the next step is not simply more CPU or more GPU power. It's how can you stream data in and set it up quickly enough to feed the GPU memory all ready to render the scene. Unfortunatly the PC is currently an antique hardware with tons of bottlenecks whether hardware or APIs. The CPU and GPU in PCs are working wonders when the scene is all compiled and there is no need to stream in new high quality assets.

But otherwise you get UE5 demo running on dinosaurs PCs costing a leg that get handily outperformed by mobile CPU and GPU running on a closed box with modern and efficient APIs.

Wow were you drunk when you wrote that?

They also seem to forget that PS5/XSX are (from a performance point of view) in a much better place vs PC than PS4/OneS was at release.

If the consoles are 1080p30 then even an RTX3090 isn't enough for 1440p/60 without dropping settings.

True on the CPU front but as noted above the 3090 can achieve 3x the real world performance of the PS5 in games that make decent use of RT. I'm not sure we had any GPU that came close to that at the PS4's launch.