Insane Metal
Member
Didn't even read it but I don't think it does either.
-
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.
- Status
Didn't even read it but I don't think it does either.
LucidFlux
Member
On a serious note. If you have a genuine interest in learning a GPUs workload throughout the time it takes to render a single frame, how various parts of the GPU bottleneck each other and limit performance check out this Nvidia blog post:
developer.nvidia.com
SM is referring to Nvidia's Stream Multiprocessors, L2 Cach hits and so on.
No GPU is doing work at 100% efficiency, or calculating 100% of the time while rendering a single frame. So why wouldn't you run the GPU at full boost while it does have meaningful work to complete and step it down when it's idling anyway or is completing a task that it can finish within the frame target at a lower clock. Thereby saving on wasted power (generating less heat) and moving that power if needed to the CPU ala smartshift.
The Peak-Performance-Percentage Analysis Method for Optimizing Any GPU Workload | NVIDIA Technical Blog
This describes a performance triage method used to figure out the main performance limiters of a given GPU workload, using NVIDIA-specific hardware metrics.
developer.nvidia.com
SM is referring to Nvidia's Stream Multiprocessors, L2 Cach hits and so on.
No GPU is doing work at 100% efficiency, or calculating 100% of the time while rendering a single frame. So why wouldn't you run the GPU at full boost while it does have meaningful work to complete and step it down when it's idling anyway or is completing a task that it can finish within the frame target at a lower clock. Thereby saving on wasted power (generating less heat) and moving that power if needed to the CPU ala smartshift.
assurdum
Banned
You have a minimal idea of what you are talking about? The point of variable clocks it's to squeeze out more than a substained clocks, genius . Grow up for the fuck sake. He talks of basic grasp, unbelievable
Last edited:
GreyHand23
Member
1080p streams on Youtube really hurt people's impressions of game graphics. The compression is so bad that it's closer to 720p or 900p in reality, plus many people won't see HDR either. In some ways I like this because when I actually start up Demon's Souls for instance, I know it will look much better than the videos, but the average consumer doesn't really take these things into account. Watching Ratchet and Clank as it was streamed in the Gamescom show vs the 4k youtube version after was crazy how extreme of a difference there was so I agree that we should hold off a bit, unless it's obvious that the game wouldn't look good no matter how high the resolution is. Animations for instance won't be better just because the video is higher quality.
HeisenbergFX4
Gold Member
People are too caught up in clocks and is this image ray traced or not.
I havent followed the XSX near as much especially the last few days (mainly from not feeling well among other things) but this next gen is a massive upgrade.
Does Demon Souls have RT?
Who gives a shit have you seen the videos coming out for it?
Especially for pure console gamers the SSDs in these things will blow you away.
Stop fighting and just take it all in whats coming.
I havent followed the XSX near as much especially the last few days (mainly from not feeling well among other things) but this next gen is a massive upgrade.
Does Demon Souls have RT?
Who gives a shit have you seen the videos coming out for it?
Especially for pure console gamers the SSDs in these things will blow you away.
Stop fighting and just take it all in whats coming.
McHuj
Member
I don't think there would be much benefit from an Infinity Cache on PS5 or XSX. For their target performance levels, they have sufficient BW from DDR. Infinity Cache is something very likely utilized on the Pro models (if there are any). Infinity cache would allow Sony/MS to double the performance of the console (2x TF) while maintaining the current memory bus setup. RX6800XT already looks like what a first cut PS5Pro GPU would be.
Completely agreed. And it will be that way in just a few days.
But we are at the speculation stage, trying to find out about "secret sauces". It's inevitable.
Developers have hinted about PS5 being "special", "surprising", "most exciting hardware in the last 20 years".
So people want to see if there's something else other than the great IO speed and the dualsense.
Last edited:
HAL-01
Member
I see you’ve come up with your own little theory based on nothing but a poor understanding of computers
Varying clocks can be better than a “full 10tf at all times” because neither console runs at 10tf at all times
Teraflops are not a measure of sustained performance. It’s not a measure at all actually. It’s a theoretical maximum number. Similarly to how cars have a “top speed” that they’d never actually reach in real world use.
What prevents machines from reaching the higher end of their potential performance is 1. Power budget and 2. Thermals, along with a number of other variables.
The variable clocks solution of the ps5 allows the system to redirect more power to whichever processor has a heavier load at any given time. This way, higher performance is achieved by wasting less power than with a fixed solution.
Also a funny thing I’ve noticed, somehow people believe the PS5, with its 350 watt PSU, won’t be able to deliver enough power to maintain clocks on its 10tf gpu, but they see no problem with the XSX being able to maintain a locked 12tf on a 315 watt PSU. Just a funny thought
buenoblue
Member
Wow. Of course a 14TF sustained xbox would be better than a variable one that downclocks. Obviously we can't have a proper discussion if you don't have a basic grasp of how cpu and gpu works. Im out morons.
Nowcry
Member
However, I think that INF.Cache adapts very well to the vision that Sony has to achieve REYES, a processing without latencies and without high data transfer.
I think Sony cannot afford a 128MB cache and put the console at 399 Euros, but I think we will see some amount of ESRAM and some cache controlled by cache and coherence correctors, to try to approach that vision without having to sacrifice the 399 $ which I think was also his goal.
I'm very curious to see the PS5 Hotchip and see what Cerny has been thinking about.
I think the question you should ask him is:
If I sell you a GPU of 10 variable TFLOPS and I charge you $ 400 and I sell you a GPU of 10 fixed Tflops and I charge you $ 500.
Knowing that in the reviews they render absolutely the same without any difference, which one would you buy?
Thinking about the answer will make you understand what we are talking about.
Last edited:
oldergamer
Member
PS5 does not have infinity cache. This speculation is getting silly. If they had it they would have announced it, or been mentioned in the announcement from AMD
dotnotbot
Member
geordiemp
Member
Yes, there is no room for a 128 MB L3, we all know that.
What we dont know if what excatly is infinity cache ? Is it 128 MB L3 ?
If Ps5 has some form of shared L1 between arrays (takes up 4 mm2 on ps5) and say 8 or 16 MB L2 ? That is possible, do we call that infinity cache lite or something else ?
Ps5 wont have any L3, it might have shared L1 and L2 a bit bigger than 4MB if your lucky but too hard to calculate at this time.
So correct there is no large L3, but it is incorrect to say no infinity cache without defining what infinity cache is.
Last edited:
Dodkrake
Banned
Ok, shovelware was strong, but there are no Xbox exclusives. Remember that, I was addressing the claims of 20+ studios and Gears Tactics being a launch game (LOL)
Jebron
Member
Video is unavailable, what game was it?
dotnotbot
Member
AC Valhalla
Last edited:
RaySoft
Member
Isn't it kinda crazy implementing a huge cache, like Infinity Cache, without scrubbers or coherency?
I mean, if something changes all cache needs to be flushed since you can't pinpoint the affected data, effectively stalling everything else.
It only gets worse the bigger the cache is. I haven't seen any mentioning of these "intelligent" scrubbers in RDNA2?
I mean, if something changes all cache needs to be flushed since you can't pinpoint the affected data, effectively stalling everything else.
It only gets worse the bigger the cache is. I haven't seen any mentioning of these "intelligent" scrubbers in RDNA2?
geordiemp
Member
You dont even know what infinty cache is first. Its a marketing term and has not been defined yet,
Is it shared L1 cache amongst shader arrays ?
Is it a big L3 cache which also gives a bandwidth boost ?
Is it a shared L1 cache plus a bigger complimentary L2 or L3 ?
Shared L1 costs 0.09 mm2 per CU I recall. Bigger L2 and L3 are out of the question on ps5 by more than 8 MB IMO.
Last edited:
PSX
Member
damn series s will be pushing ps5 pretty hard within 12 months? what alternate universe have i entered
Good times
Last edited:
smallpotatosan
Member
this
Zadom
Member
An article about smartshift that may be out of date: (thanks geordiemp fir pointing that out)
AMD SmartShift is awesome. Too bad most gamers will never get to use it
(Update: The feature will be coming to the Sony PlayStation 5.) The AMD-only feature can give you "free" CPU and GPU performance by dynamically optimizing the power budget between the processors. Laptops with Nvidia or Intel hardware need not apply even though they make up the vast majority of...
www.notebookcheck.net
Last edited:
geordiemp
Member
Another good article:
AMD SmartShift is awesome. Too bad most gamers will never get to use it
(Update: The feature will be coming to the Sony PlayStation 5.) The AMD-only feature can give you "free" CPU and GPU performance by dynamically optimizing the power budget between the processors. Laptops with Nvidia or Intel hardware need not apply even though they make up the vast majority of...
Not really, do you think thats a good artucle, its very out of date..
If you watched the AMD RDNA2 presentation last week it explained the Big Navi has pervasive frequency gate control -
What that means is allot of the block logic of the die can have their own frequency.
Sound familiar ? Its called RDNA2 CUs.
RasAlGhoul
Member
A lot of people simply can’t enjoy this hobby.
saintjules
Member
Agreed. The loading times is actually what I'm looking forward to. What gets annoying is dying and then waiting to load back into a game. Knowing that it's almost a blink of an eye to respawn is game changing. I had to stop playing Control because it would take forever to get back into the game.
carloseduardocd
Member
Last edited:
MasterCornholio
Member
From the last video that they gave us. It's actually from performance mode.
Oddvintagechap
Member
Huh? Pixelopus is one dud from being closed. What are you even saying History doesnt matter at all. ZERO.
PaintTinJr
Member
Full CU utilisation in a synthetic workload (even for the tiny period it would get close) produces X heat, and requires Y power at fixed clock 1, and in this situation both X and Y are much larger than what can be cooled and the power that can be supplied in a console.
The current gen consoles predict real-maximum utilisations expected by end of a generation - less than half full from Cerny's "doing well if hitting 40%" quote - so adequate power and cooling is provided to meet those predictions, and will have console firmware shut the device down if the thermal range and power usage are exceeded - so something that on working retail hardware would get flagged as bugs by QA in development, so games don't release that trigger theses conditions, as these are constraints for programmers.
This is how the XsX is still designed, and the PS5 doesn't need to worry about that, because such code that didn't get caught in testing or certification testing will be handled more gracefully, according to Cerny. And that is because the system deterministically balances power versus utilisation with variable clock frequency.
Take the scenario where an algorithm that downclocks the PS5 GPU uses lots of matrix multiplications (that are a critical path calculation) and then the algorithm gets optimised - post launch in a patch.
The developer realises they can rearrange the algorithm and reduce the matrix multiplications by pre-calculating the parts that are constant into less multiplications - thereby reducing the CUs utilisation for the task.
The result on PS5 is that the algorithm can either remain at static performance and have another item fill the unused compute or the algorithm gets clock boosted and completes the critical path calculation earlier, thus allowing for the next task to start earlier. So in either situation a performance gain.
On the old fix clock design the algorithm just uses less power and performance remains static.
Last edited:
duhmetree
Member
I never claimed to be an expert but I have seen numerous people jump in, with a 'matter of fact' attitude, when simple 'researching' would prove their point wrong.
Last edited:
FranXico
Member
What use is a cache scrubber if your shared cache is tiny (or not shared at all). Those things are meant to push each other in order to really reap benefits from data locality.
Gudji
Member
MasterCornholio
Member
Appreciate the concern MC, wouldn't want anyone to get an enforced holiday for throwing around insults like 'retard' now, would we?
Or calling the other guy a midget. I'm seeing it go both ways and I don't want either of you to get into trouble for it.
Insane Metal
Member
FranXico
Member
Last edited:
azertydu91
Hard to Kill
He was talking about the one X not the series X.
RaySoft
Member
Caches usually aren't shared between logic before you reach level 3.
buenoblue
Member
What you descibe is the boost clock in all but name. What I would like to know is what is the base clock that games like assassin's creed can call upon, because it cant use 10.23TF and full cpu clock for sustained lengths of time. If it could sony wouldn't call it variable.
Some people's grasp of what the series x 'fixed clocks' means is laughable. Thinking it doesn't deviate at all. Thinking that downclocking when not utilized is what sony means by variable clocks. This is what the series x and every gpu and cpu has bin doing for 15 years. What sony describes is different. Ps5 has a power budget and thermal accoustics budget, if it exeeds those then power, thus clocks are dialled back untill those budgets are not exceeded any more.
Like I said before I would like to know the numbers on these situations. If the ps5 was more powerful and smaller I could understand this approach, but it's bigger and less powerful and still can't run at its peak performance constanly. Series x is smaller, more powerful, just as quiet yet never has power budget or cooling budget adjust it's power output.
I am getting a ps5 not a series x as I stated but that doesn't mean I shouldn't ask questions about it's design and approach.
kyliethicc
Member
Once again we see that 10 PS5 TFLOPs = 12 XSX TFLOPs. Because they're different custom architectures and each have unique strengths, weaknesses, and different tool kits, different limits / bottlenecks.
4K 60 on both... shocker. And PS5 version is better due to the better controller.
Keep trying that "3rd party games better on Xbox" narrative.
Mr Moose
Member
Mr. PlayStation
Member
Last edited:
FrankWza
Member
ive said this a few pages back but I’ll put it here in case it isn’t common knowledge. Demons Souls trailer is 1440p on the YouTube app on AppleTV. The webos version of YouTube will giveyou full 4K @60.
user1337
Member
Stop presuming for me.
I know the difference between WiFi 5 and 6 and wired.
I will be using WiFi 6.
I need the convenience of not having to hard wire my console to a router in a different room, while still being able to access as much of my broadband connection as possible. WiFi 6 will help with that and help minimise signals mixing with other WiFi 5 and below connections in my house and neighbours houses.
There is no mistake in using WiFi 5. It's all good. However, the only way technology grows is if lots of people start using the latest tech. Console is one way to spread new tech and make it common place (see dvd Vs cd, and Blu-ray Vs dvd). They all got the job done, but unless some company basically makes the newest version standard and sells millions, others will not make upgrades to use it.
Bojanglez
The Amiga Brotherhood
Nikana
Go Go Neo Rangers!
Someone needs to adapt this to replace "developers" with "teraflops"
This was done a long time ago.
ToadMan
Member
Yes the PS5 can use full clocks all the time if that is desirable.
Sony aren’t interested in your lack of comprehension when they’re describing how their system operates.
Shmunter
Member
Whats dynamic visuals and physics?
Last edited:
- Status