edit: Geometry Engine is definitely part of RDNA (see the white paper), so perhaps there's some Sony Sauce in there.
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
-
We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.
PlayStation 5 System Architecture Deep Dive |OT| Secret Agent Cerny
- Thread starter vestan
- Start date
- Status
edit: Geometry Engine is definitely part of RDNA (see the white paper), so perhaps there's some Sony Sauce in there.
I don't have a clear proper analogy formed in my head yet, but I will just rely on MC. Simple structure, like a game map, run highly inefficiently while highly complex scenes with dense, small workloads run much more efficiently. They also happen to be what is encountered most while actually playing your graphically intense AAA game.
In my brain worm riddled pink stuff it works like this: In this way, the GPU is able to operate at the higher efficiency when it is correlated with a matched load , ie. small, dense workloads, and is more inefficient as it tackles larger, simpler workloads. Efficiency in a transistor means that less power is dissipated within the device as heat, and more power is used to do the actual computations. At 2.2 GHz we are firmly in low RF frequencies and traditional methodologies of thinking break down. This is all of course, just how I internalize it bc that is how RF & u-wave transistors for different applications are used.
Alright... I figured it out.
Base on my understanding of it, I would say the GPU can go as low as whatever it's required to go as ould the CPU but not higher than 3.5Gh for the CPU or 2.23Ghz for the GPU.
Basically, what sony has done is figure out exactly he much power is needed to run their APU at those clocks. The build their cooling system around that specific power profile. This is why sony knows or believes the system can run at that clock pretty much all the time. But if things start getting hotter than planned, it ould simply cut back on the power it's sending to the APU, slight 2-5% downclock of either the CPU or GPU can shave off a sizeable amount of power draw.
Thanks.
Yes but it all depends on what you perceive to be the first event. Take it for what it was, just a middle of the road tech talk. Just a briefing. Maybe a bit frustrating but we just have to wait. You got info. You didn't get games. You'll get that now the tech stuff is out of the way. Do you really think a Wednesday morning online is the way to fully announce a console? I don't really know what else you were expecting. The reality is 700k and probably quite a few more will see it at a dedicated event in the summer, the tens of millions more are likely to first see PS5 in October/November via a media campaign, just like the This Is For The Players Campaign for PS4. Just feels like people are clutching at straws when the thing isn't really out of the traps yet.
100% there are more secrets to share.
Like I said, it was a technical deep dive. Many people expected something more. I did not. If anything, I have complained that it wasn't technical enough. Do 85 million people need to watch that and be confused at Sony's messaging? Fuck no. 84 million people won't make it through the first 15 minutes without thinking they are now being talked down to cause they don't know or care a thing about the message (not for them, obviously) that was being delivered! Personally, I loved it and have watched/listened 4 times.
Huh, that does sound like there's some unreleased sauce... We know they already added the cache scrubbers in as custom to only PS5, it's not impossible there are other modifications.
ya this was a fantastic watch. He essentially agrees with what most devs have been saying. the difference in 3rd parties will be resolutions and since theres about a 15% difference in gpu and targeting 4k it will be hardly noticeable. SSD game changer. Some cases he could see ps5 outperforming xbox
Anex could you go some more into detail about the cache scrubbers if it possible and what it could mean .
Was reading what you and faf were talking about but there was not much detail on how it really works.
If you read his post, then you've already read more than I can explain. The idea in general is to prevent cache flushes that can cause pipeline stalls.
Bingo. Using slower NVMe drives as backup drives would simply be a nice extra option with no real drawback.
Yes. I'm completely blind without seeing a benchmark/performance profile and knowing how often they encounter those kinds of flushes. What I can say is that if it was huge low hanging fruit, AMD would have already plucked it themselves.
At the most fundamental level it frees up cycles that were otherwise used for maintenance.
I agree that if it was huge AMD would have it .
Still with these being close boxes every extra hand helps lol.
lol this guy reminds me of the ps5 story i wrote mocking the guy spreading the conspiracy theory that sony sent out vega devkits to trick journalists. he played a minor role in the explosive ending, but its funny reading it now because it really does look like these insane set of events most probably transpired in real life. i really dont know why reiner, colin and klee would all say ps5 had more tflops.
Possibly not as it eats up space on-wafer and in a discrete GPU not entirely possible.
While it don't have as many TF as we would hope .
It still interesting to see that Sony went with more custom hardware compare to PS4.
Time will tell if that was right or wrong but i looking forward to games to see the differences even more so with first party .
It is really exciting that they went heavier into custom hardware this time versus the PS4. I'm excited to see how it pays off!
I doubt it takes up that much space. It's just an address snooper that updates a bit in the register file of the GPU.
Maybe it needs to somehow be coherent with the RAM configuration natively? I dunno. Just spitballing.
That could be part of it. A discrete GPU wouldn't necessarily know what was in main RAM or disk.
we went another day without a clarification from sony. to me, this is it. they couldnt even get BC right after potentially gimping the gpu just to get BC? this is so hard to believe.
i dont know. i hate to keep sounding alarm bells but all of this is very similar to what happened last gen. ESRAM vs SRAM on the die increasing die size and price and reducing CU count. nand costs biting them in the ass just like that fire that raised ddr3 prices. design decisions aimed at targeting the casual market ended up costing them the power advantage; kinect for MS, $399 casual friendly pricetag for sony. leaving them both with an expensive console that is not as powerful or feature rich as their main competitor.
while MS was touting HDR support for games released as far back as 2001, and native 4k upgrades for xbox one games, even ray tracing patches for their first party games. sony is struggling to prove standard bc for more than a 100 games.
and i dont think its just poor marketing either. seeing as how the github leaks were spot on, i wouldnt be surprised that the bloomberg article was too. maybe sony was caught off guard and now is struggling to respond in literally every way possible.
And now the Million dollar question.
Is Higher Rez = Better version?
So say the XSX would always run at 10% higher rez than the PS5 and they are both targeting 4K. But the PS5 always can match its framerates, can load in assets faster and has all-round better loading times.
Which ends up being the better version?
Is Higher Rez = Better version?
So say the XSX would always run at 10% higher rez than the PS5 and they are both targeting 4K. But the PS5 always can match its framerates, can load in assets faster and has all-round better loading times.
Which ends up being the better version?
I feel like this: If Sony stuck with 36CU for BC, worked with AMD for years on it (since 2014 it is known) and has 100 games MAYBE possible for BC at launch, this is the most colossal derp I could possibly imagine. Not just for Mark Cerny, but as well for is AMD collaborators. I see this as just not true. They would have abandoned fucking ship on this thing.
They really shouldn't have announced it till after getting some good PS5 press from having a good reveal event, soften the blow.
I could see it becoming the de facto multiplayer system if you can load into matches faster than anyone else (assuming crossplay).
Thanks for posting...that was great. NXGAMER really likes what Sony has done, and he thinks the removal of bottlenecks on the PS5 will be an advantage over the XSX and make it perform better at certain times/games.
Idle thought: if Sony is dependent on matching hardware configurations for BC, every future PS console has to be able to run at 2.23GHz.
I don't think it is accurate. I think the 100 is what has been tested with native clocks. We KNOW there are literally 2 other modes where clocks and hardware are adjusted for BC. I think the vast majority of games played and this is only a PR misunderstanding.
They really need a software solution before 2027, wonder if cloud gaming can help them out in the future.
Possibly not, as they only need to retain the same efficiency levels per clock at a minimum because the clock frequency itself will vary.
If that's the case then they probably should've mentioned that yesterday, poor marketing from the event.
100 times over, yes. I think the message is piss poor. Especially with no clarification today. Of course, I have been known to be wrong, but given what Mark Cerny said during his talk, for me at least, it was crystal clear he was discussing issues with running the old games at native (and variable) clocks.
I think that's an interesting question. It's going to be up to personal preference I think.
Quick question about these native clocks, are these 100 games targeting resolutions and frame rates higher that whats achievable on the Pro?
Is the Bloodbourne 60fps dream alive?
Of course all in my mind cannon:
Bloodbourne has a capped frame rate so without a patch, that will never improve. If you use HDMI 2.1 VRR, that may mitigate the frame pacing stutter issue.
Games with variable resolution (Battlefield) and unlocked frame rate modes (Tomb Raider, etc.) will natively benefit.
Every game should benefit from reduced loading times (tossing a coin to my Witcher)
There has been no discussion of any ML enhancements or otherwise (i.e. MS HDR remapping)
It would be nothing if not interesting that's for sure.
I think people are going to be very surprised at the end of the day. Especially when it becomes clear that the only advantage the XSXhasver the PS5 is a slight imperceptible resolution bump.
Possibly, although I'd say that depends on how much, if any, difference the dynamic latency input stuff on the XSX makes to input lag. If it's noticeable, then I don't think faster load times will make up for it.
In Sony's mind, they don't need to clarify because they haven't done their actual reveal yet. I'm 90% sure that's what's going on.
Really awful PR so far.
It seems Cerny and Co's goal is to push the experience forward rather than just pixels. Bespoke 3D audio, haptics, ssd and i/o throughput are very hard things to sell to people; and it takes balls to go for that over the higher numbers.
Would I have liked an extra flop and some 16Gbps RAM chips? Sure.. but even then it's not a case of ~10.3 vs 12.1, the vast difference in clocks will likely tighten the raw compute gap to 5-10% along with more efficient utilization, cache speed and rasterization. The small compromise at the top-end is worth everything else we're getting. My only major point of concern is VRS; although I would be amazed if it isn't in there as it seems to be inherent to the RDNA2 base architecture.
Sony's biggest mistakes here is promoting this GDC-esque talk on their consumer facing channels and doing it before a more generalised reveal. I personally liked the talk but Sony could have placed it better in their schedule and conveyed the nature of it better beforehand.
Would I have liked an extra flop and some 16Gbps RAM chips? Sure.. but even then it's not a case of ~10.3 vs 12.1, the vast difference in clocks will likely tighten the raw compute gap to 5-10% along with more efficient utilization, cache speed and rasterization. The small compromise at the top-end is worth everything else we're getting. My only major point of concern is VRS; although I would be amazed if it isn't in there as it seems to be inherent to the RDNA2 base architecture.
Sony's biggest mistakes here is promoting this GDC-esque talk on their consumer facing channels and doing it before a more generalised reveal. I personally liked the talk but Sony could have placed it better in their schedule and conveyed the nature of it better beforehand.
You can do all of this with a XSX SSD too. The XSX Velocity architecture is built to do exactly this. You use the higher throughput Ram combined with the SSD speeds you can stream a lot. Raw numbers matter obviously but again it depends on a lot on the overall architecture of not only the machine but also the game engines.
Yeah the way I took it was "it's real, but doesn't paint the whole picture. Glad we can move on from it"
- Status