Future Hardware & Software Purchases

Quoted for Truth.

I think this is a rather big issue for those who (for whatever reason) want to push frame rates to very high levels. Whereas reducing GPU rendering load is typically trivial via settings, reducing CPU load by a factor of five is generally less so.

Not to mention if the console CPU code actually take advantage of the 500 GB/s RAM bandwidth. That could prove embarrassing for PCs.

 

I think the 360hz is aimed at the esports side of things.

I think that needs abit more explanation.

 

He’s saying typically PCs don’t use much more memory bandwidth than 40Gb/s. You can only get so much bandwidth from DDR4.

but these console CPUs have access to much more bandwidth. So In theory they could compute more. For specific AVX workloads in the DS space it is sometimes faster to use AVX than it is the GPU. It just really depends on cache and fitting etc. In those cases your bandwidth may reach and be fed at higher levels than typically the norm for a CPU.

 

Yes i understood that, but i see not much use cases for the cpu to access the ram at 448gb/s all that often to matter so much. Or what a 3.5ghz zen 2 could achieve in terms of bw.
Wasnt avx workloads supposed to be a problem anyway?

 

Exactly. Say, for instance that you have a game engine that does clever things leveraging the CPU, generating 1 GB/frame traffic over the memory bus. Well, on the consoles at 30fps that’s less than a tenth of the nominal bandwidth, so no problem. Even at 60fps you are perfectly fine albeit with a slightly compromised. bandwidth available to the GPU. No real problem though.
On a PC however, a good scenario today would be that the CPU was connected to dual channels of DDR4 3200, for a total nominal bandwidth of 51GB/s. You won’t ever reach that with actual game code, but the 30GB/s at 30 fps of our example should be OK. But as desired frame rates increase you run into a brick wall. No matter what graphics card you use, or what high frequency multi core CPU you own, you won’t get around your CPU bandwidth limitation. Throwing money on the PC-standard top of the line components won’t help you much in this case.

So if console frame rates are CPU processing limited, you will be hard pressed as a PC gamer to improve matters much beyond another 50% or so, but if the console CPU code uses even a relatively modest part of its available bandwidth, then you’re really up shit creek if you want high frame rates.

And even five years from now, with 32core CPUs, DDR5 and RTX 5080Ti GPUs, that will still be true.
It’s a system wide architectural weakness of the PC platform.

Now, multiplatform titles will probably not utilize the console CPUs in such a way as to totally cripple PC port performance. But assuming that they will all use a tiny fraction of the bandwidth resources of the consoles, in order to facilitate running at 144+ frame rates on PCs with a tenth of the available bandwidth is probably not a good idea either.

(All in the Jensen context of 360fps gaming.)

 

This is the reason with exclusives games I will take a PS5.

https://devblogs.microsoft.com/directx/directstorage-is-coming-to-pc/

RTX I/O and Direct Storage will only be in preview next year

And not all NVMe SSD will be compatible with Direct Storage, maybe some SSD will not be able to directly access GPU VRAM

 

Well they aren’t really a problem per se. it’s just when a developer wants to use them for whatever reason they want to it’s available to use. It will be heavy on power consumption but it doesn’t mean it’s a bad thing.

as for why we would? I guess it depends on what you’re trying to do. Some workloads may fit a CPU better than a GPU. Or if you have cycles to spare, that work may benefit more to be done on the CPU than on the GPU which may be already over taxed.

anyway, Entropy is right when you look at bandwidth as a function of framerate. If you want to hit those higher frame rates the bandwidth on the CPU side keep going up. You will hit a frame rate wall when you run out of bandwidth.

Which is more or less what happens on the GPU side of things but we don’t often think about that on the CPU side

 

It seems Direct Storage will work with new SSD and new motherboard

 

@Entropy mem BW on 2 channel for daily taps out around 65-67k read writes if you know how to tune mem. From there you can pick up latency gain if you tune sub timings. This helps a lot for maxing out games and esp in the min FPS. Unfortunately, it needs a fair bit of knowledge to pull it off and it's very much a YMMV.

Interestingly enough, qual channel like HEDT platforms have way more BW and can at times give better gaming performance even if their clock speeds are lower. Especially true in SLI scenarios where the mem BW becomes the bottleneck.

 

Yes, the GDDR memory in the consoles are shared between GPU, CPU, the audio block etc. Since in special in PS5's case, it seems bw limited already i think regarding the GPU. Is the CPU realistically going to consume more then 50gb/s for example in real world game applications?
Then we have latency too, although no idea how important that is to the CPU these days.

On pc you got between 760 and 1tb/s of gddr6x dedicated to the vram (?), and then the main DDR for CPU tasks. I doubt there will be problems as to 'embarrasing' though.
360hz gaming indeed seems unrealistic but that applies to all platforms. My original 360hz statement wasnt even related to console nor pc, but the phone market. Yes, there are phones out there claiming over 300hz.

I think this ultra high FPS gaming is aimed towards the esports scene, games like CSGO live by those 300+ fps standards for some.

 

Yea so In this instance, if your CPU requires 1GB of memory per frame, you would require 300Gb/s bandwidth to feed the cpu for 300FPS. It’s something that consoles aren’t as memory bound at, since they can access a larger bandwidth pool as a result of their setup. I don’t think it really matters but it’s an interesting discussion point for the sake of discussion. You’re going to need a hell of a GPU to get 300FPS.

 

True, and if the CPU would be consuming over 300GB/s worth of bandwith, there wont be much left for the most important part, the GPU. Forget about audio.

 

completely agreed! The new nVidia hardware feels like the second golden age of 3D graphics cards.

 

if DLSS allowed PCs to take advantage of the 360Hz on all games, it is cool, otherwise, 360Hz is almost useless or so niche like you say.

240Hz.... I find that more palatable and realistic. Still I've had a 240Hz monitor which I changed for the current one -'cos of dead pixels- and the sacrifices and stress you put on the CPU specially, but also on the GPU makes it impossible for many games to achieve such framerate and many games get little to no benefit from it.

Currently, it's quite enjoyable to play Battlefield 1, Mass Effect 2 (both on EA Access in Origin, best service ever!) Assetto Corsa and Diablo 3 at 165fps though.

 

Sounds great!

I tried out a couple of high refresh monitors last year - one IPS with GSync and the other VA with Freesync (which I couldn't use due to the age of my graphics card). High frame rates on both felt great, but the things that struck me most were 1) just how good GSync felt even as it wobbled between 50 and 70 fps, and 2) just how much VA can't cut it (at least for me) at high frame rates when the wrong colours are on screen.

At times the VA felt great, and certainly the contrast was wonderful, but during fast movement in dark scenes I had this strange mix of smooth motion and great, dark streaks across the screen - even with BFI turned on. It felt .... well ... like a big delicious pizza, but with streaks of turd across it. Sure, some parts of the pizza are incredible, but as a whole I just had to reject the experience.

Oh, and unfortunately the IPS was a monster curved ultra wide thing, which initially blew me away, but in dark areas of games IPS glow towards the edges wiped out detail and made the games look like I was experiencing a hallucinatory vision from god. So it had to go even though it performed pretty damn well.

High frame rate with good variable refresh is really immersive, I just wish panel technology was up to the rest of the pipeline. My best compromise is probably OLED at 120hz, but they don't make 'em small enough for a desk and with a PC use they'd burn in pretty fast. And I couldn't afford to replace it every 18 months ...

 

we have to wait for more information coming down the pipe.

however I will still a pc with 32 or 64gigs of ram and a 10-20gig video card over a ps5.

I doubt the ps5 will deliver the frame rates or resolution that the 3080/3090 or even navi 2 on pc will deliever. We already see the ps5 struggling with unreal engine 5.

The great thing about the pc is its not a static device. It morphs as time goes on. Next year we will most likely see the first ddr 5 capable motherboards and chipsets and another generation of intel and amd cpus. If both companies support Direct IO in the new boards and new Direct IO nvme drives also hit next year then it will all be right on time for unreal engine 5.

Its why I'm taking the $500 bucks of a xsx or ps5 and putting it into my pc.

 

A 3060/zen3 will be better already. RTX io seems faster, too.

 

I play games and the game I want will be available only on PS5 with no guarantee to be available on PC for a long time maybe Sony will release only PS4 games on PC until a PS6 is launched. And it does not guarantee all the games will release on PC.

There is some good reconstruction algoritms 1440p is ok and we need to wait if they succeed to reach the target of 60 fps on consoles for the UE5 demo.... I suppose reconstruction algorithm will improve too.

 

RTX io seems faster? How? Because Nvidia quote gen 4 max speed? Or because they showed their compressed speed? I don't see RTX io being faster than PS5 storage solution. It is fast for something that can go directly to the vram but that's it. PS5 showed 5.5 for normal speed and 9 for typical compressed (peak is 22). And remember that 5.5 is minimum speed that devs can expect when developing for PS5. You can fit a faster SSD in PS5, it just that devs probably can't make a game that rely heavily on 7 to make it run smooth.

 

That's great but doesn't really speak to anything I said.

My point was simply that the pc is not a static device. When I buy a ps5 I'm stuck with it and as time goes on its only going to become slower vs the pc.

Take a look at Unreal engine 5. It was running at 1440p 30fps on ps5 correct ? What would my 1700x 32gigs of ram pci-e 3 nvme drive with a vega 56 run it at ? What happens if i buy a 3080 or navi 2 ? How does the equation change? What if i decide to put in a 3x00 ryzen or 4x00 ryzen ? What if I wait for next year and get a new generation ryzen with ddr 5 and perhaps have 64 gigs of ram and a pci-e 4 with direct IO ?

Thats my point. I can upgrade where needed to keep gaming at the resolution and frame rates I want. PS5 will only ever get slower. We might get a ps5 pro or something but that isn't really a guarantee. We have just as much of a chance of getting the ps5 exclusives on pc .