Next Generation Hardware Speculation with a Technical Spin [2018]

Status
Not open for further replies.
I really do not view ARM as much of a challenge to work around. See Fortnite. UE and Unity are very well equipped for high performance ARM code. I wouldn't expect any serious issues for a switch to arm. Mobile gaming is only going to increase anyway. Not to mention Nintendo Switch. Also, developers not needing wide multi-threading is not a thing. They've needed multi-threading for a long time now.

Bigger issue is backwards compatibility to Jaguar and AMD not providing ARM cores.
 
K12 is dead. I don't see an Arm core fast enough to allow full BC being available in the next couple of years. And in terms of pure grunt, I'm not aware of any ARM offerings that can match the highly scalable 8 core, 16 thread Zen setup AMD would likely be able to offer console vendors in the 2019/2020 period.

The last official word about K12 from Lisa Su is that they were 'always looking at ARM', but there was no current ARM R&D, and that K12 would be available to semi-custom customers if they wanted it. Sorry, I can't find a link for the quote to the last part.

Since K12 development was 'highly leveraged' by Zen x86 development that it's not currently in development I find less of an issue.

Hypothetical scenario, but what if Microsoft wanted K12 for ARM servers and leveraged that R&D with console? I posted a link earlier about Microsoft server and XBOX silicon R&D capability being joint and how Microsoft was planning to lead with silicon innovation.

As far as other ARM big cores... ARM itself is going to have something big and next-generation for 7nm that its R&D team in Austin has been working on for years... and with expanded Softbank budget now. Computex in May is one to mark the calendar for that.
 
I meant an actual benchmark link, not just a spurious claim made off the cuff.

FWIW, transistors spent on x86/x64 ISA overhead is not 10% of a core's worth, but much less. I seem to recall claims of 1-2% or so of Intel's cell phone x86 CPUs from a number of years back, and those cores were tiny. So it would likely be much less for a modern desktop x64 core.

Well, in the video Keller mentioned 'inherent architectural efficiencies' of the ARM ISA that he thought were 'pretty cool', mentioned how that expanded performance, and that they took lessons from ARM ISA and applied them to Zen. So, in a sense, ARM is in Zen!

That's a pre-Zen 4 year old interview..
AMD dropped their ARM development years ago and Zen is pretty successful.

AMD isn't going to start working on ARM cores again just because Softbank is rich.

And Sony knows that getting devs working fast and efficiently on the new architecture is orders of magnitude more important than the whims of a japanese magnate.

According to the latest rumors, earlier devkits are already in the hands of some devs.
What do you think it's more likely to be inside those devkits? A special order SoC with an ARM K12 or a regular X370 motherboard with a Ryzen 1700 and a Vega 64?

Don't disagree with your reasoning, but by that reasoning Microsoft shouldn't be working on Windows 10 on ARM or making the major commitment to ARM servers, but yet those initiatives are taking place. So, there is another critical 'big picture' strategic dimension at play.

As far as what developers want.. I think the importance of that is overstated as when there is one these larger strategic factors defining hardware choices then the decision will just get 'pushed' onto developers. It's not like developers aren't already all over ARM with mobile?.. Switch? Potential console/mobile consolidation opportunity?.. I would think developers would be happy?

Sounds like Tim Sweeney is ready for mobile/console barriers to fall-- https://www.msn.com/en-gb/finance/t...ower-the-cross-platform-revolution/ar-BBKwYyW

I really do not view ARM as much of a challenge to work around. See Fortnite. UE and Unity are very well equipped for high performance ARM code. I wouldn't expect any serious issues for a switch to arm. Mobile gaming is only going to increase anyway. Not to mention Nintendo Switch. Also, developers not needing wide multi-threading is not a thing. They've needed multi-threading for a long time now.

Exactly, I just look at Fortnite and PUBG running on ARM and I see a trend which I extrapolate into the future. There are a whole lot of mobile devices compared to consoles and PC's..

Bigger issue is backwards compatibility to Jaguar and AMD not providing ARM cores.

Well, Microsoft has positioned itself to be free from hardware-based backwards-compatibility with UWP, and Sony doesn't seem to be to concerned with b.c at all.

Lisa Su said that semi-custom would be 'huge' for K12, and that it would still be available to semi-custom customers.
 
Last edited:
Well, in the video Keller mentioned 'inherent architectural efficiencies' of the ARM ISA that he thought were 'pretty cool', mentioned how that expanded performance, and that they took lessons from ARM ISA and applied them to Zen. So, in a sense, ARM is in Zen!
No.

Don't disagree with your reasoning, but by that reasoning Microsoft shouldn't be working on Windows 10 on ARM or making the major commitment to ARM servers, but yet those initiatives are taking place. So, there is another critical 'big picture' strategic dimension at play.
ARM is being used in servers and mobile and cars etc. Unless MS just want to be stuck with desktops and x86 servers, they need Windows on ARM.

I would think developers would be happy?
At best they won't care. The engines running in Switch have been last-gen ports. If Switch was x86 and could run the current x86 code, perhaps it'd be getting current-gen engine ports more?

Sounds like Tim Sweeney is ready for mobile/console barriers to fall-- https://www.msn.com/en-gb/finance/t...ower-the-cross-platform-revolution/ar-BBKwYyW

Exactly, I just look at Fortnite and PUBG running on ARM and I see a trend which I extrapolate into the future. There are a whole lot of mobile devices compared to consoles and PC's..
EA and Ubisoft etc have their own engines. Unity also has super-sucky performance on consoles, so yes, you can create an ARM game on it, but it's not going to get the sort of results of HZD if that's what your going for.

I don't think ARM would be a major barrier (as long as the implementation isn't a 48 core server processor), but it is another reason not to switch, and if the only reason to switch to ARM is because you can, it's pointless. It needs a price-performance advantage. As we keep repeating, despite the wide evidence you present, everything points so far to Zen being better, and the only come-back you have is a fictional future CPU with no price/performance data. Sony already made that mistake with a future Cell promise that never delivered. It'd be a brave (dumb?) console company who'd be designing their next-gen console now based on ARM, and rolling out ARM-based dev-kits, on the hope that their CPU of choice isn't flawed when they get hold of it.
 
AMD is likely to pitch an x86 cpu to all interested parties for the simple reason that they and intel are the only ones capable of supplying x86, neatly locking their customers in.
So they might offer a nice deal in order to achieve that.
 
AMD is likely to pitch an x86 cpu to all interested parties for the simple reason that they and intel are the only ones capable of supplying x86, neatly locking their customers in.
So they might offer a nice deal in order to achieve that.

More like x86 is their bread and butter so why pitch something else?

I don't think arm in a console is impossible but I don't think we'll see it happening until there is a arm cpu that can be shoe horned into a console with little effort just like a x86 cpu. And if such a part exists it will of course need some benefits over a x86 part to make sense for Sony or MS to switch.
 
So, in a sense, ARM is in Zen!
Well, I'm no Jim Keller (and I'll never be :D), but in a sense, (nearly) every x86 core since the Pentium Pro have been ARM-like, in that they translate or break down x86/x64 instructions into one or more 'micro-ops', and then execute those on a RISC-like core. (Incidentally, what was previously known as RISC-architectures like ARM also do the same nowadays... :D) Perhaps AMD cribbed bits of its ARM execution units for Ryzen's micro-op execution engines, I dunno. Who can say! Interviews like the one you linked are more PR than anything else really; all the real meat and potatoes of how the chips function on a low-level would be protected intellectual property/trade secrets and whatnot.

The engines running in Switch have been last-gen ports.
Switch runs DOOM, which isn't exactly last-gen. Also, Bethy is porting New Colossus to Switch, and one has to say that is a current-gen gaming engine...
 
My point regards Switch was the companies didn't invest in ARM ports of their engines. If it was that trivial to port them, why not? Why build a new variation of PS360 FIFA rather than port Frostbite? Going forwards, for Switch there's a market now, and maybe everything from Frostbite to IW Engine to everything Blizzard will get ARM support, but the 'it doesn't impact devs' comment seem to be fobbing off one of the concerns. Even if it's only an additional 5% hassle, why put that hassle onto your devs if the CPU change doesn't bring any benefits to offset that hassle?

To switch from x86 to ARM, ARM has to not be zero-impact relative to x86, but positive impact. At the moment the only thing we can say absolutely about an ARM choice is it will impact development and tools somewhat, especially for high-end first-party titles. Claims for cost, performance, etc., have zero current data. Do you choose ARM based on the argument, "it will impact devs a bit, but it might be cost effective."?
 
32GB or bust.

So many people are calling for 16-24GB of RAM. I don't see that as enough for a proper nextgen jump.

OS will take more than ever, GPU will ask for more (textures, render targets, meshes, buffers], CPU will handle more...

32GB or bust.
 
Last edited:
Estimated cost of 8GB of GDDR5 RAM in launch PS4 [16x 512MB chips] was $88 [APU was ~$100], so there is headroom for purchasing a lot of memory for PS5.

Samsung is already producing 16gbit [2GB] GDDR6 chips [on a undefined 10nm-like proccess node with reportedly 30% better yields than GDDR5], so 32GB is technologically already possible to be present in a future console [8 clamshell slots].

JEDEC standard supports up to 32gbit [4GB] chips, but nobody is making them yet.
 
Last edited:
AMD would be better off reviving its second X86 cpu line than going with ARM. ARM cores are getting bigger and bigger and independent manufacturers are pushing in that direction stronger than ARM itself (Apple initiated the move). The benefit over X86 are getting irrelevant, a lot comes down to OS, how the "whole" environment is benched (my belief is that hardware evaluation on any mobile OS still do not say the whole story versus MacOS and even less "matched to hardware" OS like Windows).
There is a market for tinier cores, from cheap laptop and tablets and who knows "PC gaming on the go" could take off somehow as hardware starts to deliver (for cheap that is it). It seems that AMD still can't afford to develop mutilple CPU lines, worse whereas zen is a massive improvement over Bulldozer it still suffers from (severe? I mean not easily alleviated) limitation, I'm speaking about the CCX scaling and how the L3 is interfaced with the cores on side and the local share of L3 of others (though AMD has not pushed past two for now). The architecture bears some resemblance in that regard with Cat cores.
I wonder what AMD's option are to improve that? Sticking to a lesser amount of L3 per CCX (say 4MB) and add another layer of Cache?
Put more cores in CCX while having the local share of 8MB L3 working t half speed (as in Cat cores) I assume that at such speed it could serve more clients, may be increase the L2 size to make up for the latency increase?). Anyway it gets OT and I believe that a CCX operating at a sane speed would already be a significant improvement over this gen of product especially the basic SKU.
 
But at what cost? Or waiting until 2023?
More is more but imo that is way too much. 8GB was awesome this gen, it was always a field in which consoles were significantly lagging behind PC, actually the 8GB came pretty late in the PS4 design. As for memory it fluctuates lets say often and enough to expect significant fluctuation throughout a console life cycle.
It is a little like FLOPS when this gen started even PS4 level of performance was not on the radar of most members who were day dreaming high p level of FLOPS disregarding power consumption, etc.
I think Sony should keep designing its system rationally, not subsidizing, the more you give the more some people expect out of rationality. The rational behind 4GB for the PS4 was lear, it is enough and we can have an UMA delivering enough bandwidth. As things evolved (and competition got known) it was simply wroth to deprive them from any selling point on raw hardware provess (yet I think the PS4 would have been fine and Sony would have fight a little more on price).
In that context the PS4 Pro, which Sony acknowledged having though about all along, made even more sense and that system would have gone with 8GB no matter what.

Imho companies have to get out a little of social media and "managing" (or so they think) people expectations, they are feeding an ever growing Leviathan that swallow whole whatever you throw at it and keep asking for more no matter what.
I hope the XB1X does not sell to well (good start) so it injects rationality back into some executives heads, simply put we can't keep going (gaming and on a more general manner) entering dick contest at any costs, without actual strategic thinking and a mid/long term vision and pretend it is a good business practice, we can afford to have crazy giants killing good practices and competitions on such premises, not too mention hiding the economics (but also the world) fundamentals to the People. As for MSFT they created imho a piece of kit that could prove a significant hurdle for them to overcome: one year late, tough to get something much better, costly, significant development costs, etc. Imho MSFT possible salvation was a "better", a more significant redesign of the XB1S. and fight Sony on price (hardware and subscription price and benefits). They would have left their options open to react to Sony next move which is a comfortable market leader which has never been under pressure to down price its system aggressively this gen and that even in MSFT strong market. Now they have a one year buffer (if Sony aggressively move to another system stinking to incremental upgrade that are in -line with process as well as AMD tech availability), and they are fighting with a performance advantage and also 25% price desadvantage against Sony offer.
 
But at what cost? Or waiting until 2023?
Let's hope the governments crack down on the ram price fixing soon. There are investigations now but with the speed of government it'll be a while til the manufacturers get the ol' slap on the wrist and prices come back down.
 
More is more but imo that is way too much. 8GB was awesome this gen, it was always a field in which consoles were significantly lagging behind PC, actually the 8GB came pretty late in the PS4 design.
And it came at the very start of design for Xbone.
 
More like x86 is their bread and butter so why pitch something else?
Exactly. And it neatly shuts nVidia out of the running. (Unless they partner up with intel for an x86 CPU. That doesn't sound like a budget friendly alternative though. :))
I don't think arm in a console is impossible but I don't think we'll see it happening until there is a arm cpu that can be shoe horned into a console with little effort just like a x86 cpu. And if such a part exists it will of course need some benefits over a x86 part to make sense for Sony or MS to switch.
An A75 stomps the Jaguars in the PS4 and XBOne. Ryzen performs better in absolute terms, but I have no reason to believe it would do better than an A75 in performance/watt.
The benefit to Microsoft and Sony to go ARM is that they could source different suppliers between generations or even roll their own. AMD will try to ensure that it doesn't happen by offering a decent deal. Which way they choose to go is probably less influenced by technical minutiae and more by corporate strategy.
 
Don't disagree with your reasoning, but by that reasoning Microsoft shouldn't be working on Windows 10 on ARM or making the major commitment to ARM servers, but yet those initiatives are taking place. So, there is another critical 'big picture' strategic dimension at play.
Microsoft is trying to get WIndows on ARM (which they will probably fail again) because they don't want to depend on Intel as much as they do right now. Even more considering how little competition Intel had on the x86 market until recently with Zen, which made Intel stagnate on core count in consumer products for almost 8 years.

This has nothing to do with consoles, Softbank or other external forces.


As far as what developers want.. I think the importance of that is overstated as when there is one these larger strategic factors defining hardware choices then the decision will just get 'pushed' onto developers. It's not like developers aren't already all over ARM with mobile?.. Switch? Potential console/mobile consolidation opportunity?.. I would think developers would be happy?

There are still very few multiplatform devs working on the Switch, and the ones who do aren't exactly taking their state-of-the-art engines to the console (see FIFA and other sports games).

The question still remains: in a cutthroat market for super important release-window game lineups, what do you think the earliest devkits for the PS5 (and eventually Xbox Two) will have, other than an existing high-performance multi-core x86 CPU together with a high-performance discrete GPU?


Switch runs DOOM, which isn't exactly last-gen. Also, Bethy is porting New Colossus to Switch, and one has to say that is a current-gen gaming engine...

That makes it a total of one (1) current-gen engine running on the console.
 
A75 stomps the Jaguars in the PS4 and XBOne. Ryzen performs better in absolute terms, but I have no reason to believe it would do better than an A75 in performance/watt.
Please look back at the xeon example I provided. Twice the performance per watt by droping the clock. Absolute core performance costs a lot in perf/watt, regardless of the ISA.

So unless you have apples-to-apples performance profiles, the perf/watt of a smartphone chip means nothing.
 
Please look back at the xeon example I provided. Twice the performance per watt by droping the clock. Absolute core performance costs a lot in perf/watt, regardless of the ISA.

So unless you have apples-to-apples performance profiles, the perf/watt of a smartphone chip means nothing.
Of course I don't. How would that even be possible?

Additionally, future SoCs would use future core implementations, in unknown numbers and at unknown positions in the frequency vs. power curves, running code that in at least one case would be completely new binaries, ......

If you believe that x86 will be competitive vs. ARM in a relatively low power multi core environment, well, no-one can really argue with that, the data just doesn't exist. My belief is that no, it will be at a certain disadvantage, but not greater than that other factors will dominate.

Where is Laurent when you need him.
 
Last edited:
Of course I don't. How would that even be possible?

Additionally, future SoCs would use future core implementations, in unknown numbers and at unknown positions in the frequency vs. power curves, running code that in at least one case would be completely new binaries, ......

If you believe that x86 will be competitive vs. ARM in a relatively low power multi core environment, well, no-one can really argue with that, the data just doesn't exist. My belief is that no, it will be at a certain disadvantage, but not greater than that other factors will dominate.

Where is Laurent when you need him.
Okay I agree, it depends what we consider an important margin. Lets say it would provides 10%, other factors will be dominant, I think 10% is negligible. But if it were twice the perf/w or per die area, it would be a very different discussion. There are no indication it would be any such dramatic differences.

Judging from threadripper/epyc benchmarks against intel, Zen looks very good as a middle ground between intel (king of absolute core performance) and current attempts at arm server chips (beating intel by a good margin in perf/watt, with lower power cores).

Die area and cost are much more difficult to estimate. We know little about large arm cores with desktop-class I/O.
 
Status
Not open for further replies.
Back
Top