Next Generation Hardware Speculation with a Technical Spin [2018]

Status
Not open for further replies.
Fair point I missed the Bloomberg reports regarding A12 at 7nm when I searched the process and wound up on the vague TSMC press release, I'm still sceptical given the poor yields Intel has suffered but there's no denying that the gap between 28nm introduction and adoption for PS4 would be similar for a 2020 PS5 at 7nm

Edit: I'm not going to get in to Intel 7nm vs TSMC 7nm (as I'm not qualified to discuss) but if as Intel claims that TSMC 7nm is more like their 10nm then that actually boost the chances of 2020 PS5 being TSMC 7nm given that yields on Intel 10nm have been keeping the lights on for years at this point
 
Last edited:
Fair point I missed the Bloomberg reports regarding A12 at 7nm when I searched the process and wound up on the vague TSMC press release, I'm still sceptical given the poor yields Intel has suffered but there's no denying that the gap between 28nm introduction and adoption for PS4 would be similar for a 2020 PS5 at 7nm

Intel is being extremely aggressive on some of their pitches, and have admitted needed up to sextuple patterning for their latest process. They’re doing a lot of cool things like cobalt, COAG, and SDG, but it may be too much at once.

28nm HVM to PS4 release -> 23 months.

7nm HVM to PS5 November 2020 release -> 30 months.
 
Going by this article HVM for 28nm started in October according to TSMC but further Xilinx state they were shipping their FPGA chips on HKMG (I think the console chips use HKMG?) since April 2011.

The article as a whole mirrors closely in words and timeline wise with what we've heard about 7nm recently so 2019 can't yet be ruled out.

https://www.eetimes.com/document.asp?doc_id=1260456

That doesn’t surprise me. FPGA vendors tend to be very low volume and extremely high margin. Single FPGAs can go for upwards of 20K.
 
Yup FPGAs are used for very price insensitive tasks such as High Volume Trading (a firm I worked with a few years back used them to make very low latency NICs) or defense applications
 
Starting to be fairly convinced that whatever incarnation of Zen the PS5 will have, it will support AVX512, even if it’s 256-bit wide paths double-cycled. The Sony engineer responsible for the znver1 commits is having more and more AVX512 related commits lately.
 
Fair point I missed the Bloomberg reports regarding A12 at 7nm when I searched the process and wound up on the vague TSMC press release, I'm still sceptical given the poor yields Intel has suffered but there's no denying that the gap between 28nm introduction and adoption for PS4 would be similar for a 2020 PS5 at 7nm

Edit: I'm not going to get in to Intel 7nm vs TSMC 7nm (as I'm not qualified to discuss) but if as Intel claims that TSMC 7nm is more like their 10nm then that actually boost the chances of 2020 PS5 being TSMC 7nm given that yields on Intel 10nm have been keeping the lights on for years at this point

Intel delayed volume production at 10 nm until 2019. Yields still aren’t good enough.
 
Intel delayed volume production at 10 nm until 2019. Yields still aren’t good enough.
”Yields” is a slippery concept. Yields for what product, what is the die size and overall complexity? Is the silicon functional but doesn’t scale well in frequency? Theoretically Intel could have 100% functional dies, but if they can’t scale to their frequency targets and perform worse than their immediate predecessors, they are going to be difficult to sell.

It’s a balancing act of complexity, market projections, competitive environment, ...
It’s no secret that Intel has been having issues with their next process node, but that doesn’t mean that TSMC or Samsung will face the same delays! There is a lot more than physics involved in whether it makes sense for intel to move their product stack to a new process, and those considerations don’t necessarily transfer.
 
”Yields” is a slippery concept. Yields for what product, what is the die size and overall complexity? Is the silicon functional but doesn’t scale well in frequency? Theoretically Intel could have 100% functional dies, but if they can’t scale to their frequency targets and perform worse than their immediate predecessors, they are going to be difficult to sell.
How about a tiny die, with only 2 active cores, its GPU disabled and worse clocks/power then its 14nm counterparts.....

so by bad yields, it looks like its on all fronts.......
 
How about a tiny die, with only 2 active cores, its GPU disabled and worse clocks/power then its 14nm counterparts.....

so by bad yields, it looks like its on all fronts.......
OK, picture this hypothetical scenario:
Sony plans to release a 15TF PS5 in time for the 2019 holidays shopping spree. However, AMD and TSMC can’t hit their power/clock targets for obscure reasons. They are 20% down on clocks at their target power draw! For Sony this becomes a choice of either postponing the launch, hoping that the issues will be sorted out in a year for a holiday 2020 launch, or they can increase the fan speeds a little and launch a 13 TF console on schedule. Arguably, in spite of missing their targets, they are still capable of launching a perfectly valid product, and they can even get goodwill for lowering noise or cheapened manufacturing once the problem is sorted.
Now imagine the same situation for Intel. There is no way in hell they can justify starting high volume manufacturing with such issues, they already have something better out there so they would fill warehouses with product that would have to be dumped. It makes more sense to crack the whip over the chip guys until they have sorted out the issues in a fashion that allows intel to maintain margins.

Now this was purely hypothetical, but should demonstrate that something that can be seen as a show stopping yield problem in one mass market sector can be totally acceptable in another.
 
Last edited:
How about a tiny die, with only 2 active cores, its GPU disabled and worse clocks/power then its 14nm counterparts.....

so by bad yields, it looks like its on all fronts.......

Yup. Intel bit off a lot. Cobalt interconnects, COAG, and SDG all at once, and they’ve admitted to needing sextuple patterning. They’ll be chewing for a while, it seems.
 
If we are guessing future hardware shouldn’t we be looking closer at what is required to make a noticeable difference to what we already have? What’s the point of baby steps in hardware? What are we aiming for? ...please don’t mention ray tracing in game...it’s not going to happen in any meaningful way for a long time.

What is the next big thing?
I believe we are going to need something about 2 - 3 more powerful than what we already have at a minimum.
That would be a good place to start.
 
That has been looked at over the past 1400 posts. There are choices between times and performances. Let's say for example that the next significant upgrade that's as notable as PS2 to PS3, say, cannot happen until 2025. Does that mean there shouldn't be a new console with a lesser advance before then? So this thread has largely looked at a reasonable release schedule and what tech will be available with 2018-2020, 2020 being 7 years after PS4 launched which would be as long a gen as PS3 > PS4 (which was considered overly long by many).

Incidentally, 2-3x more powerful than what we already have is XB1X. That's well below what a generational advance provides, which is broadly about 8-10x more powerful than the previous generation (power being impossible to quantify in single metric). I think the key point next-gen will be a deciding tech. One will be better compute and GPU driven rendering, freeing up a lot of CPU. Another will probably be notable CPU power rather than weak-source mobile cores. Third could be a glut of bandwidth. Even with a limited progression of raw GPU power, game should be able to push the envelope once again.

I think the next jaw-dropping next-gen visuals will be when we can get something like the Dice Star Wars raytracing demo. We'll probably progress towards that piecemeal though (look how graphics have massively improved this gen) and we'll probably never get a jaw-dropping next-gen leap ever again, unless some paradigm shift appears.
 
If we are guessing future hardware shouldn’t we be looking closer at what is required to make a noticeable difference to what we already have? What’s the point of baby steps in hardware? What are we aiming for? ...please don’t mention ray tracing in game...it’s not going to happen in any meaningful way for a long time.

What is the next big thing?
I believe we are going to need something about 2 - 3 more powerful than what we already have at a minimum.
That would be a good place to start.
I have often felt, that we've not been stepping back enough to look at the forest from the trees. Just building off what Shifty wrote in terms of technical discussion.
GPU side dispatch, more bandwidth, more compute.
If I told you that the next generation console will have the following features:
  • 4K60 with HDR
  • 4K30 with HDR
  • 1080p120 with HDR (same number of pixels per second as 4K30)
  • Freesync
  • Dolby Atmos
  • UHD Player
  • App Support
  • Backwards Compatibility, dating back 2 generations
  • Enhanced visuals for backwards compatible titles
  • Enhanced visuals for older titles
  • Newer and better services
People would say, yup. That's next gen console right there.

All of these are supported on Xbox One X today.
4K TVs haven't overtaken 1080p sets yet, and VR hasn't flown, so I'm not sure a company would base it's entire strategy either one.
Ray tracing is going to be that leap but it's not happening at $399 anytime soon.

So next gen will either do the above list better, or this generation is going to be extended for a long time.
 
Last edited:
I have often felt, that we've not been stepping back enough to look at the forest from the trees. Just building off what Shifty wrote in terms of technical discussion.
GPU side dispatch, more bandwidth, more compute.
If I told you that the next generation console will have the following features:
  • 4K60 with HDR
  • 4K30 with HDR
  • 1080p120 with HDR (same number of pixels per second as 4K30)
  • Freesync
  • Dolby Atmos
  • UHD Player
  • App Support
  • Backwards Compatibility, dating back 2 generations
  • Enhanced visuals for backwards compatible titles
  • Enhanced visuals for older titles
  • Newer and better services
People would say, yup. That's next gen console right there.

All of these are supported on Xbox One X today.
4K TVs haven't overtaken 1080p sets yet, and VR hasn't flown, so I'm not sure a company would base it's entire strategy either one.
Ray tracing is going to be that leap but it's not happening at $399 anytime soon.

So next gen will either do the above list better, or this generation is going to be extended for a long time.

HDMI 2.1 is a necessity IMO. It enables full chroma, 10/12 bit color depths, and greater than 60 hertz refresh rates in all combos at 4K. You have to sacrifice on multiple fronts to fit in the fastest 18Gb current HDMI connections. eARC is important for audio. Plus, VRR is baked in.

If a next gen console doesn’t have HDMI 2.1, it’s severely hobbled IMO.
 
HDMI 2.1 is a necessity IMO. It enables full chroma, 10/12 bit color depths, and greater than 60 hertz refresh rates in all combos at 4K. You have to sacrifice on multiple fronts to fit in the fastest 18Gb current HDMI connections. eARC is important for audio. Plus, VRR is baked in.

If a next gen console doesn’t have HDMI 2.1, it’s severely hobbled IMO.
They can release a refresh of the consoles to support HDMI 2.1 if it's a hardware issue.
It's been done before, it doesn't impose significant costs.

If it's not a hardware issue, and say it's possible to firmware upgrade, they could do that as well.
 
HDMI chipsets should have sampled late last year or early this year.

I would imagine TVs and AVRs with HDMI 2.1 will be announced at CES next year and then ship during 2019.

Even if a next-gen console launched in 2019, there should be no excuse for it not to have 2.1.
 
HDMI chipsets should have sampled late last year or early this year.

I would imagine TVs and AVRs with HDMI 2.1 will be announced at CES next year and then ship during 2019.

Even if a next-gen console launched in 2019, there should be no excuse for it not to have 2.1.
Samsung is already pushing updates for their Q series that enable VRR despite lacking 2.1, so I think the industry understands the importance to gamers.
 
Status
Not open for further replies.
Back
Top