Prophecy2k
Veteran
Out of interest, why does having exactly double the number of CUs as the PS4 Pro mean "easy" BC?
Predict: Next gen console tech (9th iteration and 10th iteration edition) [2014 - 2017]
- Thread starter Shifty Geezer
- Start date
- Status
Disable half the gpu for BC mode?
It doesn't require half. We're talking napkins math easy though.
Scorpio does some weird stuff with it's 40 CUs, sometimes it can use all, other times it can only use 1/2.
Prophecy2k
Veteran
Yeah but double PS4 Pro would mean 4 x PS4 CU count, so 3/4 of the GPU would be enabled for BC-mode... I'm not sure how that makes BC "easier" than some other arbitrary % of the GPU disabled to facilitate BC?
Also, being likely Navi-based (with some future arch. features) wouldn't BC by the PS4Pro method essentially be a non-starter; since I imagine the PS5 GPU micro-architecture would be a much more significant departure from PS4 than the Pro was...no?
Also, being likely Navi-based (with some future arch. features) wouldn't BC by the PS4Pro method essentially be a non-starter; since I imagine the PS5 GPU micro-architecture would be a much more significant departure from PS4 than the Pro was...no?
All valid points. I think we may have pushed the mirror message too far. Today as I understand it, for native games, it will use 1/2 the GPU (which is effectively a PS4) and then with boost mode, use 4Pro clock speeds, without boost mode, use PS4 clock speeds.
So I guess we took the whole mirror image thing further than we need to. But then again...
For Scorpio, they didn't really deviate that far from 1X either, the technology is as I understand it, essentially xbox one, scaled up and profiled for 4K. Certain older SDKs require the GPU to shut off 1/2 as well. I'm not sure of the reasoning but this is the case. Clocks do not change however.
There's something there we need to take account of, but perhaps we're putting too much emphasis on it.
One thing that would need to be defined is what the duties main OS are, particularly for Sony.Additionally, I would like for Sony to move entire main OS to the secondary "background" chip, Make it beefier [or even X86 if needed to maintain compatibility with PS4s software... heck it can even be 2 regular Jaguar cores], place 4GB DDR3 ram next to it, and dedicate entire APU for gaming.
Microsoft's system has three operating systems, with one potentially being independent enough from the game partition to be physically running on a separate cluster.
The OS is still there for a lot of system functions that games are generally not expected to implement, or are forbidden by the platform from doing. Moving platform and system functions out of the APU can create significant overheads where there weren't before.
One thing to note about doubling the PS4 Pro is that this takes GCN outside of the maximum range of CUs, shader engines, ROPs(?), allowed without design change.
It's not impossible, as AMD noted with Vega there's no theoretical reason for it to be fixed. AMD didn't fell motivated to do so, but perhaps if paid it might bother.
I thought the idea of the mirroring configuration was to make it easy to keep a perfect cycle-exact BC for the difficult titles which would never get patched, while at the same time doubling makes it easy to have a balanced system without waste. Would something asymetrical work as well?
AMD was talking about increasing the CU's IPC for Vega/Navi, but it didn't seem to work for Vega (unless they meant DR fp16 is the ipc increase). So my wild assumption was that Navi could possible get either wider SM, or some other tweaks providing my (completely hypothetical) 20% more effective results per clock. Whichever way, shouldn't we expect improvements from Navi versus polaris/vega? I mean like frequency, or feeding SM more effectively, or new wider instructions, etc... Obviously if amd manages to clock GCN at nvidia's speed, it makes everything simpler.
The reason I wanted doubling was also to get a similar die size at 7nm. So maybe similar cost.
The two little pieces of info they gave about Navi were: next gen memory, and scalability. So maybe they no longer have a CU limit. Or maybe the new architecture have macro modules.
AMD was talking about increasing the CU's IPC for Vega/Navi, but it didn't seem to work for Vega (unless they meant DR fp16 is the ipc increase). So my wild assumption was that Navi could possible get either wider SM, or some other tweaks providing my (completely hypothetical) 20% more effective results per clock. Whichever way, shouldn't we expect improvements from Navi versus polaris/vega? I mean like frequency, or feeding SM more effectively, or new wider instructions, etc... Obviously if amd manages to clock GCN at nvidia's speed, it makes everything simpler.
The reason I wanted doubling was also to get a similar die size at 7nm. So maybe similar cost.
The two little pieces of info they gave about Navi were: next gen memory, and scalability. So maybe they no longer have a CU limit. Or maybe the new architecture have macro modules.
phoenix_chipset
Regular
Honestly? Sounds good to me lol. 16 jaguar cores at 3ghz or the higher side of 2 sounds good enough. And if someone says 16 cores will never be utilized, why talk about 8 cores with hyperthreading? BC should be a priority for ps5.
phoenix_chipset
Regular
Prophecy2k
Veteran
I'm not all that sure Jaguar cores can be clocked so high (i.e. 3GHz) within a reasonable power envelope.
Also, even at 3GHz a 4-core Jaguar module would almost certainly get bitch-slapped in performance by a 4-core Zen CCX.
I'm not sure CPU code in a gaming context is at the stage yet where single-threaded performance isn't important. I'm pretty sure the gains in IPC from Jaguar to Zen, as it pertains to game code, cannot be reasonably be overcome by simply adding more Jaguar cores at the same clockrate (that's assuming you can even match the clock speeds, which I doubt you can).
In which case, choosing a 16 core Jaguar CPU over an 8 core Zen for PS5, will simply amount to throwing away a true generational upgrade in CPU performance for the sake of BC... as much as BC would be nice, I just don't think it's all that important going into next-gen.
phoenix_chipset
Regular
For 100% compatibility? Like Wii U, Wii Ps2 etc. It needs the same exact architecture for that. Unless there's been some revolution that i'm unaware of.Why would moving to Ryzen present a problem for compatibility
You don't even need the same instruction set for some sort of compatibility e.g. Xbox one, but that's not what I mean when I say BC.
phoenix_chipset
Regular
Well the single thread improvements would come from the higher clocks. Not the biggest improvement in the world no, but it'd be something substantial at least over the base ps4. I too wonder if they could reach 3ghz, hence why I said the higher side of 2ghz.
Certainly it would place more burden on the developer ; having to work with more cores to compensate for the lack of single threaded performance to realize more complex game design. But BC should be the industry standard, until a point is reached where the hardware would be held back tremendously like Wii U. And certainly the Jaguar is not anywhere near as dated as espresso was.
But, they always could just slap a jaguar in there alongside a new zen cpu, it would require very little power to operate. It could double as the PS5's OS chip and leave the whole zen chip to games. Nintendo should have done that with the Wii U ; slap wii's cpu in there and have a more modern design as the main chip.
PC is 100% compatible on the CPU side. As long as instructions aren't present in the old architecture and missing in the new (without being emulated in microcode) then CPUs are interchangeable, which is why you can take a 5 year old PC on an i5 and swap out the mobo+CPU for Ryzen and run all the same software. By sticking with x86, CPU compatibility should be a given.
And are covered by copyrights, making it impossible for new suppliers to enter the market.
- Status