ResilientBanana
Member
You get what you pay for.
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Digital Foundry: PC Gaming on the Xbox Series X CPU
crazyriddler
Member
Fast?? The tech is evolving waaay more slower than decades ago, same with the graphics evolution. And people where fine with it. The "need" of pro consoles it's the whinnig of nonconformist people with a need for consumption and who would equally complain about a ps5 pro, or ps5 pro ultra after a few months of launch.
There is no need of it, the only improvements they can bring are literally more pixels on screen, more frames or maybe some extra reflection, things that do not represent any considerable improvement or in many cases not even noticeable and that only serve to show off and perhaps cover some insecurity.
SportsFan581
Member
The 4090 is not mainstream. It's impossible to find a $250 or even $400 GPU today that can best the base consoles by the same percentage the 1060 was for $250 last gen.
I think that is what he is getting at.
Zuzu
Member
That explains some of the fps numbers we see in games. No wonder Starfield can’t run at constant 60fps. Still I guess for the cost of the console and that consoles have routinely targeted 30fps it’s capable of doing the job decently. And with VRR it’s not as bad that it may not be able to always hit 60fps in those games that support a performance mode.
hinch7
Member
Yeah ignoring the API differences, hardware configurations etc optimization on a large portion of newly released PC ports leaves a lot to be desired. With that said.. UE5 can perform well if we look at games like Fortnite.
Hopefully we'll see better optimizations on the engine and games using UE5 (and other updated engines), additionally more titles that make use of DirectStorage.
Last edited:
Ronin_7
Member
If it's done much better than Ps5 version why are multi platform games still shit on either Xbox?
Lmao
Bojji
Member
Exactly.
2014:
980ti $649 - 100% power
960 $199 - 48% power
2016:
1080ti $699 - 100% power
1060 $299 - 49% power
2018:
2080ti $999 - 100% power
2060 $349 - 60% power
2020:
3090 $1,499 - 100% power
3060 $329 - 45% power
2023:
4090 $1,599 - 100% power
4060 $299 - 32% power
And 4060 is better than 3060 in some games and worse in others! Only 1/3 of (almost) full chip performance. Prices of top GPUs went absurd and mainstream ones are weaker and weaker compared to them every gen, for 3xxx series 3080 was the best deal but almost no one could buy it at launch.
If it's done much better than Ps5 version why are multi platform games still shit on either Xbox?
Lmao
I meant how AMD has done the whole board, 4700S had shit PCIE slot so you couldn't even use any decent GPU on it and only 2 SATA ports. They made it better with this Xbox APU.
Last edited:
Ronin_7
Member
So Sony had to customize the whole shit? Crazy
The 4090 is mainstream? Those type of high-end products aren't considered average for most PC gamers.
Bojji
Member
Both of those apus are failed chips that didn't make it into consoles (probably something wrong with GPU part), some time ago AMD decided to sell them (instead of throwing away). First they made 4700S from PS5 chip bit was not a great product, with Xbox APU they improved some aspects in design.
They sell this only in china to make some money.
Gaiff
SBI’s Resident Gaslighter
And that's only if you compare the top available chip to the 60-tier card. In actuality, the 4090 is only 90% of the full AD-102 die whereas the 1080 Ti was 93.3% and the 2080 Ti around 94%. The 4090 despite its name has more in common with traditional 80 Ti cards but NVIDIA knew they couldn't sell a 4080 Ti for $1600. 4090 sounds much nicer because the 90 suffix only made a return with Ampere. Before that, it was a dual-GPU configuration that stopped being made with Kepler (GTX 690). Historically speaking, the 4090 is an 80 Ti-class card, not a Titan one like its name would have you believe. Lovelace actually has great performance improvement but everything below the 4080 has been gutted so badly so as to not make the top guys look bad that they ended up screwing up the mid-range. I don't think a 60 card getting beaten by its predecessor in so many instances has ever happened before. Usually, the new 60 outpaces the previous 70 and sometimes even comes within striking distance of the previous 80.
NVIDIA has been charging us more while giving us less over the years. People were outraged at the $1000 2080 Ti. Well, the 4090 is a $1600 4080 Ti.
Last edited:
Unknown Soldier
Member
It's long overdue for Sony to quit x86 and move PlayStation to ARM architecture based SoC for the PS5.
Dr.Morris79
Gold Member
Playing Demons Souls
ReBurn
Gold Member
I mean if you have a PS3 copy RPCS3 is a thing...Playing Demons Souls
BennyBlanco
aka IMurRIVAL69
Yea, it’s so good that you’re in every PC thread complaining about all the issues you have with your PC
Stooky
Banned
that confiq with the amount of ram needed would cost too much. devs like the single large pool of ram. just need more cache on cpu
rofif
Can’t Git Gud
Maybe you have. I am fine.
Nothing with my pc is responsible for shitty devs that can’t make the game not crash in 10gb vram cards!
Just to comment on this since I watched it recently. I believe Cerny had said that the decompression unit was equal to about 9 Zen Cores of processing which is ALOT to take off of the CPU
Last edited:
Stooky
Banned
they don’t want the cost of designing a chip. much cheaper to partner with Amd. anything outside of that does not equal $400 console.
A tough ask. Given how most of gamers buy into an ecosystem, rather than hardware itself. You'd have to have some amazing tools/software to ensure porting and backward compatibility (with an acceptable performance).
HeisenbergFX4
Gold Member
Or give me upgrades every 3 years or so
Dr.Morris79
Gold Member
That sounds untoward, i'm phoning F.A.C.T.
Dream-Knife
Banned
But the Zen 4 in the pro consoles would be cut down too. It might be better for a year, but then we're back at where we started.
2080ti was $1200
Why though? ARMs only real benifit is power savings, and that's because they build real wide. Do you want a chip with the cost of a 3090 and the performance worse than what you already have?
Clear
CliffyB's Cock Holster
Or give me upgrades every 3 years or so
That's basically an insane ask when dev cycles are now mostly 3 years plus!
Constantly moving the goalposts isn't going to result in more optimal code, it'll likely do the reverse and incentivize functionality over efficiency and waiting for hardware to brute force it.
True optimization takes time, because it requires experience and understanding of where new efficiencies can be made and where bottlenecks are likely to occur. The longer the turnaround time on projects -including engines- then the slower the advancement is likely to be because there are limits on what can be re-engineered on a "live" project. The traditional way being; work, ship, do a post-mortem, and apply what was learned on the succeeding project.
The only way you can break this cycle is by adding technologies that give "easy" gains, and that's generally the province of hardware designers and middleware suppliers, not game developers.
twilo99
Member
The "next gen" consoles should employ some 3D cash based APU with a potent CPU to cover UE5 engine based games properly
Only if they can get Apple silicon...
Also, great for back compatibility, but I guess PS gamers are used to that kind of thing by now.
Only if they can get Apple silicon...
Also, great for back compatibility, but I guess PS gamers are used to that kind of thing by now.
Unknown Soldier
Member
If Apple acquires Sony, I can see future PS consoles switching to Apple Silicon. That would certainly be something. Apple Silicon humiliates anything on the x86 in perf/watt comparisons.
Typing this post while watching Leagues Cup on the Apple TV+ app on my PS5
Last edited:
AnotherOne
Member
I ditched my xsx and ps5 to pc for 99% of my gaming so for me it's another one 99% of the time
thicc_girls_are_teh_best
Member
Oh okay I see you did a stealth edit. Sony acquires Apple? Maybe in 1996 they could've
Yep. Now, there's probably a caveat there; Cerny probably meant equal to 9 Zen cores in terms of doing decompression of I/O tasks, which is a bit different than being equal to 9 Zen cores in whole.
Still though, that's a hell of a lot of offloading from the CPU, and something the Xbox Series consoles don't quite really manage. Microsoft went for a more hardware-agnostic approach because they want a solution that's also scalable on PC. That works for them, but the problem seems to be with Series X, there just simply isn't enough raw hardware power overhead to implement that type of solution without sacrificing some type of performance.
In many cases with 3P games, coming up short of PS5 in some area, usually framerates.
Last edited:
Team Andromeda
Member
It's not about being optimized to get every last cycle. I remember the days when games like Doom 3 or Half-Life 2 were made for the PC 1st and foremost and with no consideration about any console ports
Team Andromeda
Member
You're doing a shit job of employing it *rollseyes*
Team Andromeda
Member
For consoles, they're a big jump over what has gone before, which seems to be last on many. Never mind 3 years in there's still only a couple of games where a PS5 or Series X owner could say 'this' game couldn't be done on a One or PS4.
I also like to go back to the days of a separate GPU and CPU. So we get monster GPU's like the old Xbox or 360 days
sendit
Member
Based on which benchmarks? I would expect the Zen 4 mobile CPUs to be roughly competitive with the M2 Pro.
Last edited:
rofif
Can’t Git Gud
Low Moral Fibre
Member
There was a point when CPU performance didn't seem to matter and you could have the same CPU for a decade but things have started to escalate quickly once AMD actually started to fight back and it's interesting to see how big an impact a better CPU can have in certain situations.
I tried my 4080 with my 3900x before putting it in my 7800x3d PC and the minimum framerates in some games was nearly 5x higher (as in they dropped to ~30fps on the 3900x but stayed at 140fps on the 7800x3d). Not every game of course but even those least impacted by CPU performance are much smoother on a better CPU.
The 3D cache seems to make short work of many problem games and I never seem to have problems with performance now even in the so called broken ports.
Obviously the higher you go towards being GPU limited the smaller the gap between the highs and the lows but I aim for my monitor refresh rate in every game now unless it is super heavy to run.
I can see why there are so few 120fps games on console as it would likely lead to very erratic frametimes but for the kind of GPU performance they have I don't think the CPUs are bad and certainly much better than last gen.
It's all about getting a decent balance between price, performance, cost and power draw and these consoles are well balanced imo.
It's pointless having 7800x3d level CPUs and 2070s level GPUs unless you are playing simulation heavy stuff which tends to only be on PC anyway but at least if the GPU is stronger than the CPU you can aim for higher resolution and more effects etc which is why I doubt the Pro consoles will have some crazy powerful CPU but I guess we'll see.
I tried my 4080 with my 3900x before putting it in my 7800x3d PC and the minimum framerates in some games was nearly 5x higher (as in they dropped to ~30fps on the 3900x but stayed at 140fps on the 7800x3d). Not every game of course but even those least impacted by CPU performance are much smoother on a better CPU.
The 3D cache seems to make short work of many problem games and I never seem to have problems with performance now even in the so called broken ports.
Obviously the higher you go towards being GPU limited the smaller the gap between the highs and the lows but I aim for my monitor refresh rate in every game now unless it is super heavy to run.
I can see why there are so few 120fps games on console as it would likely lead to very erratic frametimes but for the kind of GPU performance they have I don't think the CPUs are bad and certainly much better than last gen.
It's all about getting a decent balance between price, performance, cost and power draw and these consoles are well balanced imo.
It's pointless having 7800x3d level CPUs and 2070s level GPUs unless you are playing simulation heavy stuff which tends to only be on PC anyway but at least if the GPU is stronger than the CPU you can aim for higher resolution and more effects etc which is why I doubt the Pro consoles will have some crazy powerful CPU but I guess we'll see.
winjer
Member
No it doesn't. Apple's cores are very good, but they are on par with AMD and Intel's offering.
In fact, Jim Keller, stated that ISA is pretty much irrelevant for performance or power usage.
And this guy developed CPUs for AMD, Intel and Apple. So he knows exactly what he is talking about.
The real advantage that Apple has is that, most of the time, they get first access to newer process nodes from TSMC.
It also helps that Intel had Brian Krzanich, who screwed up Intel's lead, by disinvesting on R&D, so he could make a quick buck with stock manipulations.
YeulEmeralda
Linux User
So buy a PC then. The PS5 is selling so clearly neogaf is in the minority. Just like the people who have a 4090.
PaintTinJr
Member
I thought this video was pretty good - surprisingly - and Richard did a good job with context about what he was actually looking at, and the comparison test - without the games - were very informative IMO.
It is interesting to see that the CPU is a measly 60nanoseconds more in latency (on average) than the desktop counterparts, giving very good evidence for why console hand-rolled deterministic code on lesser mobile grade processor could easily hide and amortise that typical random workload latency at the beginning of processing.
The L3 cache sizes difference is logically different between desktop and console, mainly because the last level cache (LLC) at L3 is really a cache for unexpected workloads needing sizeable data close for efficiency, which is really a PC requirement, in highly optimised console code situations the size of L3 give a performance lift will get progressively smaller because of the predictable nature of the processing and the latency of the IO systems being so much smaller in this gen's consoles reducing the need for hiding IO latency with a large L3 cache or with less transfers
It is interesting to see that the CPU is a measly 60nanoseconds more in latency (on average) than the desktop counterparts, giving very good evidence for why console hand-rolled deterministic code on lesser mobile grade processor could easily hide and amortise that typical random workload latency at the beginning of processing.
The L3 cache sizes difference is logically different between desktop and console, mainly because the last level cache (LLC) at L3 is really a cache for unexpected workloads needing sizeable data close for efficiency, which is really a PC requirement, in highly optimised console code situations the size of L3 give a performance lift will get progressively smaller because of the predictable nature of the processing and the latency of the IO systems being so much smaller in this gen's consoles reducing the need for hiding IO latency with a large L3 cache or with less transfers
Last edited:
winjer
Member
It's not. For some very strange reason, their 3600 is only doing 90ns of memory latency.
A normal value of latency for it should be around 75ns. With tuned ram, should be around 65ns.
My 3700X was doing 63.7ns with tuned memory. This is close to 30ns lower than what DF is getting.
They must have something very wrong with their memory config to get a latency that high.
For a long time I was finding their benchmarks with this CPU to be somewhat off, to what I was seeing in other sites gaming benchmarks, and even on my own benchmarks.
This huge latency that their CPU has can affect performance quite a bit, since Zen2 really likes low latency.
Last edited:
PaintTinJr
Member
True, but it doesn't impact the first party console games, when their OS is a real-time OS with minimal resource usage optimised for games and the workloads are predictably batched so the latency hit happens only when kicking off an new unpredicted process and then hidden within the workload being batched.
The scale of latency difference being tiny was my main point, rather than the specific value. Even at an extra 40ns wouldn't break the bank for quality hand-rolled code by first party devs, and is only an issue with devs treating the consoles like a budget PC which has been rife through the cross-gen faceoffs.
Last edited:
winjer
Member
Not everything fits in caches. Especially in a CPU that only has 4MB+4MB of L3.
Zen2 has good branch prediction, but this does not negate memory accesses. And it does not negate cache misses. And an L3 cache miss in these consoles CPUs, is very expensive.
It doesn't matter much what the OS is doing in these consoles, because they have dedicated threads, it doesn't impact much the game that is running.
Although there is a good deal that programmers and a good compiler can do to avoid cache misses, these will still exist. And even the predicted accesses still have to occur.
But my previous point was not about the latency on the consoles. It was about the terrible latency that DF has on their 3600.
A normal 3600 has much lower latency than that. And this invalidates their benchmarks with this CPU.
They must be running it at such low speed and high timings, that it's underperforming by a very big margin.
LordOfChaos
Member
It's the GDDR
It's made to deliver wide bandwidth without caring too much about latency, which GPUs are perfectly fine with. But CPUs are more latency sensitive
Combine that with smaller caches and yeah
The only thing that sort of surprises me is that for many years, CPU upgrades barely mattered to gaming, but I guess now they do. I suppose that's also a function of moving on from Jaguar cores where everything had to do less to stay playable at all on consoles, things are still playable on these but may struggle for a locked 60 say.
Last edited:
winjer
Member
It's the GDDR
It's made to deliver wide bandwidth without caring too much about latency, which GPUs are perfectly fine with. But CPUs are more latency sensitive
Combine that with smaller caches and yeah
Dude, read my post again.
I'm talking about the latency of the their 3600. Not the 4800.
90ns is not a normal value for this CPU.
Last edited:
DenchDeckard
Moderated wildly
are you saying the 4090 is a minority card? it's sold boatloads. Thats a positive and also painful that NVIDIA managed to make us all upgrade to a 1500 pound card!
LazyParrot
Member
no $1600 GPU is ever going to be a majority card lmao
LordOfChaos
Member
Ok dude
GHG
Member
Steam hardware survey results says it represents ~0.26% of the market.
LordOfChaos
Member
It would be interesting to to see them pair it with as close to the consoles GPU as possible and see how games run, investigating the mystical "console optimization" angle vs just running the same hardware on PC