Next Generation Hardware Speculation with a Technical Spin [pre E3 2019]

Status
Not open for further replies.
Is there any chance at all that Sony and Toshiba had worked with AMD to do a 3D chip similar to the one in PS Vita? HBM2 perhaps on top of the APU? http://chipworksrealchips.blogspot.com/2012/07/sonys-ps-vita-uses-chip-on-chip-sip-3d.html

Cerny talked about terabytes per second solution they opted out for PS4. I don't remember that he gave reasons other than they wanted an easier development for the PS4. Now that we are in 7nm, and we have the HBCC solution, is this route for the PS5 practical now?
 
Cerny talked about terabytes per second solution they opted out for PS4. I don't remember that he gave reasons other than they wanted an easier development for the PS4. Now that we are in 7nm, and we have the HBCC solution, is this route for the PS5 practical now?
It was a choice between speed and capacity, as ever. 1 TB/s would have been a small scratchpad/VRAM type solution, requiring tiling and moving stuff in and out of RAM.
 
Writing on QLC is crazy slow when the SLC cache is full....

Or if you don't have a SLC cache at all, which I find likely for a console drive.

It really doesn't matter much. Write speed only has to match the speed of the the internet connection/the speed of your BR drive, whichever is slower.

I also think it's likely that they will will match the logical block size with the the drive page size, which on qlc is likely 64kB.
 
Writing on QLC is crazy slow when the SLC cache is full....
As @tunafish mentioned, write speeds are irrelevant when you're only writing when getting data from the internet connection (100MB/s tops), a Blu-ray drive (20MB/s tops) or recording gameplay (20MB/s tops if 4K H265).
 
It was a choice between speed and capacity, as ever. 1 TB/s would have been a small scratchpad/VRAM type solution, requiring tiling and moving stuff in and out of RAM.
Maybe this next generation of hardware is sufficient to calculate some things on the fly. If so we might see a lot more procedural approach.
 
Can someone with an account on Resetera please ask hmqgg and klobrille if the "Next-Gen" gpu and/or Zen 3 cpu in Anaconda is true or not?
 
I have a supposedly blazing fast m2 ssd, but in practice it does not work that way for games. If Sony has a custom solution that delivers real results in eliminating load times and streaming issues, I'd say that'll be the killer feature of next gen. I'd choose that over a better gpu any day.

After playing a little of GR: Wildlands (PC) and RDR2 (Xbox-X) today, I'm in a 100% agreement of this.
 
Can someone with an account on Resetera please ask hmqgg and klobrille if the "Next-Gen" gpu and/or Zen 3 cpu in Anaconda is true or not?

hmqgg has been asked.

What makes him/her an expert or insider on the Xbox-Next and/or PS5? Please don't tell me he/she got a prior E3 game leak right. I've seen plenty of those claiming to be insiders after E3 predictions.

Honestly, if Digital Foundry, IGN, AnandTech or even The Tech Report aren't whispering certain rumors, then everyone else (i.e., Reddit, ResetEra, etc.) are worthy of dump-truck loads of salt.
 
Last edited:
More baseless speculation:

- io chiplet with stacked dram on top of it
- cpu chiplet (zen 3)
- gpu chiplet ("next-gen")
 
PhysX goes open source.

That physics demo vid you linked to is next-gen, I agree, but it's also quite distracting regards the water IMO as it looks like gel, not water. Would be nice to have good water simulations though.
Unless I totally missed it, I have yet to see hardware accelerated PhysX running on non-nVidia hardware, though, even though it's been open source for half a year. Also, that video is from a shipped game from 2009 called Cryostasis, made by the developers of Chasm: The Rift and Carnivores. There are other examples, but that game always stood out to me because the PhysX effects change the entire look of the game from start to finish. The Mirrors Edge PhysX enhancements are also nice. That game had destructible cloth and tons of PhysX particles but it was more window dressing then a fully realized PhysX showcase.
 
Will we only really see the most substantial "next gen" effects of ssd baseline for exclusives for many years after next gen release?

Wouldn't third party games still be limited to PC's mix of different tiers of SSD and HDD and also more cross-gen titles of this gen and the next?

What's gonna happen to PC AAA games? Slower gameplay speed, terrible LoD/poppins and much worse load times or is this gonna be the firs time consoles lead the pack where if PC wants to "properly" play AAA games they'd need to buy higher end SSD assuming the next gen console's ssd are higher tier?
 
Will we only really see the most substantial "next gen" effects of ssd baseline for exclusives for many years after next gen release?

Wouldn't third party games still be limited to PC's mix of different tiers of SSD and HDD and also more cross-gen titles of this gen and the next?
Quite possibly. It depends on the economics of scaling ideas. History and common-financial-sense tells us the lowest common denominator will define feature usage.

However, console exclusives are often more plentiful, so possibly more than just a handful of titles if it really makes a difference. A dev wanting to use a feature can trust in tens of millions of consoles, versus only a few million PCs for high-end feature.

What's gonna happen to PC AAA games? Slower gameplay speed, terrible LoD/poppins and much worse load times or is this gonna be the firs time consoles lead the pack where if PC wants to "properly" play AAA games they'd need to buy higher end SSD assuming the next gen console's ssd are higher tier?
It's not the first time. In the 8 and 16 bit eras, PCs were junk! It's only since 3D where monster PCs could out-perform consoles, and only really by throwing money at the problem which 3D enables.
 
I was wondering if instead of going with an SSD, consoles might delve into the smartphone space and go with UFS instead. Samsung is already mass producing UFS 3.0 512 GB rated at seq reads of 2100 MB/s and random reads of 68K IOPS. 1 TB chips are due later this year. Hell even a 2.0 or 2.1 version would provide massive improvements over today’s consoles’ slow ass hard drives. I read on a website that claimed that the PS4 Pro and non pro are rated at 100 MB/s for seq reads and less than 100 IOPs (4K) for random reads.

Current estimated BOM on the galaxy 10+ runs about $50 for 8 GBs of RAM and 128 GB of UFS 2.1. Given that most of top end phones run with 2.1 UFS and account for 100s of million in unit sales annually, I imagine that smartphone manufacturers ever constant appetite for 1 upping or keeping up with Joneses might be able to drive higher performance larger memory chips at cheaper prices than SSDs in the near future.

https://www.micron.com/-/media/clie...vel_performance_in_embedded_systems.pdf?la=en

A white paper comparing nvme vs UFS.
 
Last edited:
Need to revise my calculations for next generation APUs, I was under the assumption that the non-cpu and gpu portions of the chip would take up a constant percentage of the die, but further research shows that this is not the case. Bus controllers + etc shrink in size too with each node progression. In fact, you can fit a 384 bit bus in less die space as the 256 bit bus on the Pro.

So my new conclusion is that non-cpu / gpu on the next gen console should only be around ~80mm2 (1.8x scaling over the 140mm2 in the ps4 pro) if they don't add extra bus controllers, caches, dark silicon. In a 350mm2 die there should be 270mm2 to use, assuming cpu and Rops take around 100mm2, there is ~170mm2 remaining, which should fit 64 Vega 7 sized CUs with 30mm2 to spare.

New estimates for me is that >330mm2 SOC should be big enough to accommodate 8 core Zen2, 64CUs (4 disabled), 64 Rops.

Another epiphany of the new conclusion is 512 bit bus for the GPU is not crazy out of the world. It's less die space expensive than the 256 bit bus in the base PS4.

So we're straight up limited in next generation by Navi clock speed and the 64 CU limit.

If there is no CU limit, you can do ~100 CUs in a xb1x sized 360mm2 die. :eek:
 
Are you basing that off of the Raven Ridge die? Where are you getting the I/O size for comparison? Do we have a foundation for comparing GDDR6 or at what frequencies* that need to be supported?

*higher data transfer rate paths need a somewhat more robust I/O, though the converse implication for HBM is fairly clear - 2GT/s vs GDDR's rates, which are several times that.

non-cpu / gpu

I prefer Intel's "uncore" term. ;)
 
Last edited:
What makes him/her an expert or insider on the Xbox-Next and/or PS5? Please don't tell me he/she got a prior E3 game leak right. I've seen plenty of those claiming to be insiders after E3 predictions.

Honestly, if Digital Foundry, IGN, AnandTech or even The Tech Report aren't whispering certain rumors, then everyone else (i.e., Reddit, ResetEra, etc.) are worthy of dump-truck loads of salt.
Verification from forum admins that he has legitimate ties to the industry which could provide him with inside knowledge. He also hinted at GPU decompression and SSD usage before those rumors were universally accepted.
 
Status
Not open for further replies.
Back
Top