but what happens to the cpu frequency then? surely it cant stay at 3.5 ghz. we know the cpu is probably 20-30w max based on the 4800u apu, how much can they really save by downgrading the cpu clocks? 5-10w?
-
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.
PlayStation 5 System Architecture Deep Dive |OT| Secret Agent Cerny
- Thread starter vestan
- Start date
- Status
You basically called out cerny, as if you knew better. You really think he would say 'most of the time' if he didn't mean it?
I love hearing stuff like this. Really seems that the PS5 is going to be the biggest jump in the way games are made since the mid-90s.
this is a very interesting and informative post. thanks.
i am watching the DF direct right now, and richard also mentioned how some smartphones and other devices use this kind of clock pushing based on load. ive just never seen it before. its weird this goes against everything sony used to believe.
it will be very interesting to see what happens come launch.
Honestly, no clue. There's a trade-off there for sure, but we would need the formula to determine the actual numbers. From what we've heard from developers though, it's not something to worry about it seems.
Its an AMD solution as it's used in their laptop APUs. Sony just dialed it up to 11.
As to why MS didn't just go with this? Design choices priorities. Note sony is using some die space to put staff in the chip that's not in the MSchip, hardware specifically made to make the most of their SSD solution and circumvent all the bottlenecks and that is why they have their 1-second loading as opposed to what is likely to be 8-second loading in the XSX.
Obviously, sony believes having a faster SSD is better than having a higher TF number. It has clearly been at the center of their entire console design.
One of the biggest data duplicating examples we have is Spider-man and it had only 20% of the install size used for duplications. So expect games to get x2-x4 larger while no data duplication will get them 20% smaller. 100GB will probably be next-gen's lower limit.
I really hope Sony will have a 1.65TB SKU, 825GB is just not enough for me. 825GB is 768GB and that's before the OS, the SSD data structure, preinstalled apps, space dedicated to quick resuming of games and space dedicated to virtual memory. I wouldn't be surprised if we will only get 600GB-650GB of free space for games. I really need a 1.65TB SKU, at launch.
You have it wrong he used those eg to show it's not like that .
It's something different that has not been done is this way before.
Exactly. Some people don't pay attention. He said the exact opposite of what he typed. It's a boost that's never been used before in devices. Not a typical boost as people see in regular cards/ phones.
i am so confused by this shit. lmao.
i did watch through the gpu section and he does say they capped it at 2.28 but they technically couldve gone with a higher clock.
i wonder if the 80 cu can hit 2.5 ghz on pc. that would be nuts. 25 tflops. all you need after that is a 7.5gbps ssd and you've got a ps5 pro at launch.
They build the system around the SSD so i wonder how that going to come into play with the bandwidth .
has any techhead talked about the memory bandwidth? Name of the poster is enough I'll check in the post history
It's 94% of the XSX BW/FLOP. It'll be fine. Notice people are comparing XSX to 2080 Ti and no one is crying about that difference.
Because Mr. Cerny said that when it drops, it's projected to be by "a couple percent". As long as the drop is less than 3%, PS5 will remain at 10 TF, not 9. It would take a ~7% speed drop to get them below 9.5TF, and a 13% drop to get them to 9TF.
No, rasterization does not scale with number of CUs. The RBEs are separate hardware, attached to the Shader Engine frontend in AMD terminology. The XSX and PS5 have the same number of them, but PS5's are clocked much higher. Its pixel fillrate will be above XSX.
I believe that AMD's raytracing solution runs on the TMUs, and that TMU performance scales with clocks and not just CU counts. In that case PS5 is likely to perform behind XSX due to the overall compute gap, but adding raytracing won't make the performance difference even larger.
This. Everyone still talks about the game even today. It's cuts deep into my heart.
EA, aren't you listening?!? *Cries*
Do we have any other examples of the typical data duplication cost? Also, just glidin' over the compression part, eh? Kraken is a REALLY good compressor, and it will essentially be free on PS5. Here's a blog post comparing some compressors.
User Banned (5 days): Hostility, history of similar behaviour
Stuff like this is why we can't take most gaming journalism seriously. Instead of doing research when the time is right. He is out here just speculating on what he thinks may happen with no type of proof. Top journalistic integrity here folks 🙌
where is your proof behind your claims my guy
Why are you fully guessing as an editor of a tech journalism website
Leave the guessing to keyboard warriors pls
I heard on average a 100 GB PS5 game without duplication will be equivalent to a 140 GB PS4 games. Other things they only compress by 50% on average the size of game on PS4, here it is 160%.
I don't care about who's on top christ, just if someone wrote about it
It does.
Thts not how it works. The Raster units aren't tied to the CUs they are tied to the SEs, of which both the PS5 and XSX has the same amount.
Beaten by Liabe Brave with an awesome post no less.
Wow, thanks for the great article! Here's an excerpt:
Note: zlib is what the XSX is using according to articles.
Believes fraud Klee, calls out legit journalist Tom Warren.
Checks out.
It would take only a 3% drop to take it below 10TF and into the 9TF area.
I in the show some first party Sony goodness with there SSD and other customise stuff .
XSX will have the edge in multiplatform titles nothing big .
I assumed 825 was after the OS And everything.. made the most sense to me at least
We pretty much do. The Microsoft renders of their SOC clearly show two SEs. There is no way that a single SE could support all 36 CUs on PS5. And GitHub--which has been pretty reliable--indicates identical RBE count on both chips.
I need Sony to invest in some survival horror games. I need a complete town, with lots of backtracking, and every home/building to be accessible.
Lol you have issues dude. If what he is doing with his statements is considered legit journalism then I'm terrible at my career.
i know right?
Jason Schreier on Twitter
“@thefuckyusay The narrative will for a long time (possibly forever) be that the Xbox Series X is more powerful than the PS5. Narratives matter!”
twitter.com
So the PS5 isn't as powerful as the XSX...I kind of expected that.
The XSX is the higher end model though right? Or has there been any info about the lower specced one getting canned? We've heard nothing about it yet.
The XSX is the higher end model though right? Or has there been any info about the lower specced one getting canned? We've heard nothing about it yet.
What intrigues me is the lack of talk about Adaptive/Variable Rate Shading and Machine Learning (DLSS has tangible benefits) from PS5's RDNA2 reveal. Given the smaller CU count and consequently, comparatively less powerful RT capability, these performance saving features would be great boons.
It would be quite disappointing if PS5 lacks these features.
It would be quite disappointing if PS5 lacks these features.
I am surprised that some people chose not to understand what Cerny is trying to achieve with the power load balancing between the GPU and CPU. When a game frame rate tank down below the intended FPS target, it's never because of both GPU and CPU. If at some point in time, the bottleneck is the CPU, the system will decrease the GPU frequency a bit and redirect this power to the CPU. If the bottleneck is the GPU, which I suspect is 99% of the case then, it will let it run at full 2.23GHz frequency while decreasing the power to the CPU. I find this really smart.
Crap. Back to being unhappy.
The context of Cerny's words suggests the case-by-case testing of those top 100 games is in relation to running BC titles at the non-native clocks.
But you're right - it's still ambiguous...
- Status