Next Generation Hardware Speculation with a Technical Spin [2018]

Status
Not open for further replies.
Remember heat and leakage increase exponentially as the frequency goes up. You'll likely get more performance within a low thermal envelope (say 30-40W) with SMT than by marginally increasing clock speed.

Hypothetically speaking, if your limit is 35W, would you rather have 8c/16t @ 2.8GHz or 8c/8t @ 3.2ish?
 
Last edited:
Remember heat and leakage increase exponentially as the frequency goes up. You'll likely get more performance within a low thermal envelope (say 30-40W) with SMT than by marginally increasing clock speed.

Indeed. A few hundred Mhz @ a hypothetical 3GHz base-clock would only amount to about 10-15%. I think historically, we might find boosts up to 35% from SMT, although Phoenix's point is regarding the point of diminishing returns, in which case, it'd be difficult for non-developers to ascertain with any real certainty given the nature of targeted hardware development and the scope of limitations (budget & scalability of game design).

----

In the long run, given the issues surrounding fab nodes, it might be prudent to go wide early on, and then focus on clocks later on, not unlike how the half-gen updates went.
 
Last edited:
Indeed. A few hundred Mhz @ a hypothetical 3GHz base-clock would only amount to about 10-15%. I think historically, we might find boosts up to 35% from SMT, although Phoenix's point is regarding the point of diminishing returns, in which case, it'd be difficult for non-developers to ascertain with any real certainty given the nature of targeted hardware development and the scope of limitations (budget & scalability of game design).

----

In the long run, given the issues surrounding fab nodes, it might be prudent to go wide early on, and then focus on clocks later on, not unlike how the half-gen updates went.

I mean there are games right now, that (mainly) only use 2 cores on the ps4 and xbox one like kingdom come deliverance. Sure we can't foresee how thread scalability would play out exactly for this next gen, but the fact is simply having 6 cores for this gen didn't force developers to adopt heavy multithreading en masse.

I think it's important to note as well, that this shift didn't happen even though the jaguar cores are pretty darn weak individually. With zen, I can't imagine devs will have a fire lit under them since the cores are actually good this time.

So yes @shiznit I actually do think a 3.2ghz zen would be more widely beneficial than a 2.8ghz zen with smt, but that's just me.
 
With smt or hyperthreading there's more work being done by the cores which uses more power and therefore creates more heat. So in a console closed box, smt would result in lower clocks to compensate. A developer would have no such choice on console rather than making use of all threads or not.

The clocks wouldn't be vastly lower mind you, maybe 300 to 400 mhz or thereabouts but considering the state of software development I would personally say smt is still not worth it in a closed box assuming 8 cores are there.

Thanks, but I don't think I was clear enough. I was asking if developers could, in theory, choose the mode in which the CPU operates, per game e.g. 8c/16t @ 2.8Ghz or 8c/8t @3.2Ghz?

From what you wrote, it sounds like SMT only impacts clockspeeds when it's actually in use, and clockspeeds could remain unaffected if the hardware was still capable of SMT, but that function was disabled/not utilised.
 
Since this is the speculation thread and I have yet to post a prediction.....Speculation says that rumors (yes, there is a joke there)may become more solid in the near future so I wanted to take my shot now and have a little fun with it. I did try, mostly, to stay within the confines of real possibility. I may have over-stepped a bit. Once more unto the breach, Dear friends! Or at least, once again back to the well? I'm going with MS and Sony playing to their strengths. Sony clones the PS4 and MS goes with a high/low approach that leverages their network strengths and looks to the XOX and 360 for inspiration. I'll put them in price order.

Note: Originally I did research 7nm and 7+nm, die space, transistor density, etc, but eventually dumped the frequency and die space numbers I came up with as ultimately unreliable and not particularly relevant. The higher frequency vs more cores debate, how much difference would the decision ultimately make? EULA - As always, worth every penny you paid to read it. No refunds on time lost.

$250 - Xbox Net
The price is more important than the hardware here. How much do you want to undercut the PS5? Frankly I don't think there is enough information to make a decent hardware guess. Beyond the statements that there will be some level of local hardware work such that a simple Chromecast style device with a controller packaged in will not be enough. For now.

I do believe MS will want to be able to appeal on a reduced initial outlay. There will be a subscription requirement. Much like the dreaded "cell phone" contract. Something introductory like getting 3 months free for your first 12-month contract. The low end will include Xbox Game Pass and more expensive services include other publishers or perhaps even some ala carte' by publishers? That could make the publishers rather happy. I really expect all kinds of options for how you pay for your games on here. You just won't own any of them.

A test on the Xbox website to see if your ISP service can handle the connection requirements would go a long way to alleviating fears about your connection quality. Be even nicer if you could run it in the background and let if accumulate a few days worth of tests and tell you what kind of quality you can expect. For those who do not fear the "games as service" model where you own nothing and you incur on-going costs rather than larger initial outlays, this is your console. I think Sony will let others try this out and take the risks rather than jump in right away with a low end cloud based box.

1Q 2020
$400 Sony PS5


The early adopters are going to buy about as many as you can make anyway. Launch in the Spring and get your ducks in a row for a blowout 2020 Fall with 1st party titles having an extra year to look good and get out the door. Could make for some nice holiday bundles as well. It could have launched over the 2019 holiday, but why? Just because it can? PS4 gets one more holiday season in the sun. I said everyone is going back to the well and I meant it. Sony leverages their first-party lineup, overall popularity and goes with what brung em' in the last generation.

128GB/ 1.5TB SATA SSHD - user replaceable. Will continue to work, only slower of course, if the SSD degrades over time/ use. With RDR2 already hitting the 100GB mark, I felt the 128GB was a near minimum.

7nm APU - 6C/12T Zen 2 + 12TF Navi - The "secret sauce" is something minor. Nothing cataclysmic like AI upscaling, RT or Path Tracing.

8GB DDR4 for OS/ App use - This is too handy a trick and I expect MS to use it as well.
16GB GDDR6 256-bit

4K BR / Full BC at launch / FreeSync

Seems boring? Nothing sexy here? It is straight-forward and an update on the idea of the PS4/Pro. Sony sells on their software library, price and enough power to get the job done.

Fall 2020
$500 Xbox 2 RT
(or whatever made up word the marketing department comes up with)

Back to the well again. I think MS has realized that, for the time being, they are just not going to be able to go head to head with Sony in 1st party titles. The 360 era was their heyday. Maybe you get lucky again but why count on it? Scorpio got them some respect back after the Xbox One stuttered. Take some lessons from what worked and run with it. The Xbox Net gives you a price competitive option. Now they go after the high ground. This may backslide into early 2021, but only on fab problems I would expect.

If you are going to take the high ground, perhaps something a little more than just raw power would be useful. I toyed with both RT and AI up-scaling (DLSS kind of tech). MS has no small amount of experience with AI. AMD and MS worked on the RT together already (along with others of course). Since "up-scaling" can be a dirty word, and I fully expect to hear plenty of "Native 4K!" marketing early, I went with RT.
Marketing matters here. There is already developer support for Nvidia's RT tech so I see no reason why millions of potential console customers wouldn't be enough for developers to support it. Especially if you are selling "RT enhanced" as part of your marketing. Engine support shouldn't be long in coming. I can already see the tag lines - Xbox 2 RT - the best place to play "insert fps du jour here".

256GB NVMe SSD - M.2 user replaceable. If this fails/ degrades past a certain point, it must be replaced. I expect heavy users would view this similar to replacing your disc drive. They always fail at some point. This being a requirement will open up some serious possibilities.

2TB SATA 7200rpm Mechanical - user replaceable
8GB DDR 4 for OS/ App use

16GB GDDR6 256-bit

7nm+ (Partial EUV) APU - 8C/16T Zen 3 + 16TF Post-Navi + 4GB HBM3 - RT enabled (How much die space dedicated to what kind of hardware specifically? I have no friggin' idea. This is just the direction I would expect MS to go. Maybe it is dedicated silicon. Maybe it is a modification to the existing GPU that gives it more punch for RT work.)

4K BR / Full BC at launch / FreeSync

An unanswered question. When does MS launch the Xbox Net? I did not put a launch date because I was torn. I expect this could be ready by the time the PS5 launches. Do you launch along side or wait and do a dual launch with the 2 RT? E3 2019 should be very interesting. My gut says, if you can, launch it alongside the PS5. You would have a lower priced competitor from the start and time to work out your kinks before possible major sales in the 2020 holiday season. Then MS brings the marketing for the next year getting people to wait on their PS5 purchase because yours will be higher quality on all those 3rd party titles come the Fall.
With chip design costs skyrocketing and major players stating this could be the last console generation with a move to cloud computing staring it in the face, who says their will even be shrinks? 5nm may not be that far away but for a $500m+ design cost, how many would you have to sell at a reduced price to be worth it? And when? We could be stuck with these consoles until either the major players give up on the "thin-client" idea as unworkable for the foreseeable future and release new hardware or the "cloud" remains just out of reach, but close enough they all keep trying for an extended time period. Some types of games will obviously translate better than others here. I feel I did go a bit overboard on the specs. For the Xbox almost certainly. But with the future shrinks uncertain it felt better to launch a more powerful device now since you are already going after a premium market with the $500 price point.
 
Since this is the speculation thread and I have yet to post a prediction.....
Some random unorganized thoughts on the Xbox front.

Once the technology is available, there is generally no reason to delay production unless your design isn't finalized. The only other option is cost, but not a lot changes in costs to manufacture in a single year. So likely they would be willing to absorb some cost to release at the same time as their competitor because falling behind a year, even 1/2 year could be detrimental to the generation.

That being said, on the MS front, everything they want has already been deployed years in advance. The APIs (DX12, and DXR, and DirectML), the services (BC and Gamepass), and feature sets (free sync, 120fps, M/Kb, accessibility, etc), and the OS have already been deployed. Any licensing issues and all that jazz has been done up front.

So they are effectively waiting mainly on the hardware front.

They will either launch at the same month as Sony, or they will launch before them. It's unlikely they'll launch 1/2 a year behind. Just my thoughts there. They have been learning from their mistakes big time. Going the RT route could trip them up and fall behind. Going the 4K route is a path they already know and have learned tons about (using XBO as their test bed).

They used XBO live game code at 4K and used that to profile the 1X processor. They were effective in maximizing an effective 4K setup and have largely met their performance targets on 3P and 1P titles. You could easily look at their next generation device as being Xbox One X 4K code being profiled for possibly 60 fps ? / even more ultra features?

They can start profiling RT code as well theoretically as game code for DXR games are coming out within the next 2 years, but not likely cause they aren't designed for 1X.
 
Thanks, but I don't think I was clear enough. I was asking if developers could, in theory, choose the mode in which the CPU operates, per game e.g. 8c/16t @ 2.8Ghz or 8c/8t @3.2Ghz?

From what you wrote, it sounds like SMT only impacts clockspeeds when it's actually in use, and clockspeeds could remain unaffected if the hardware was still capable of SMT, but that function was disabled/not utilised.
I don't know really, I guess it's possible. On PC if you disable multithreading in the bios, in theory that would allow the boost clocks to reach higher levels on average than with smt on, but i've never really looked into it.

But yeah if that wouldn't present any problems that'd be the best solution.
 
I mean there are games right now, that (mainly) only use 2 cores on the ps4 and xbox one like kingdom come deliverance. Sure we can't foresee how thread scalability would play out exactly for this next gen, but the fact is simply having 6 cores for this gen didn't force developers to adopt heavy multithreading en masse.

I think it's important to note as well, that this shift didn't happen even though the jaguar cores are pretty darn weak individually. With zen, I can't imagine devs will have a fire lit under them since the cores are actually good this time.

So yes @shiznit I actually do think a 3.2ghz zen would be more widely beneficial than a 2.8ghz zen with smt, but that's just me.
Of course, there are always outliers (CS:GO I'm looking at you). Cryengine was never really updated to a job-based system. It was my impression that most console engines are now 80-90% "jobified" and run 1000s of jobs per frame, going as wide as possible. This means less regard for low-level optimization, focus on breaking up the work into the smallest possible units, and letting the CPU sort it out. I believe the TLOU Remastered and Killzone presentations talk about this. In that scenario SMT seems to make a lot of sense, especially with GDDR6 latency in play (not a dev so I'm guessing here).
 
Yup Sony and Microsoft will know if they need smt and if the next consoles have it it's because there devs have said that it's useful. So it will be interesting to see what they go for higher clock or smt.
 
Since this is the speculation thread and I have yet to post a prediction.....Speculation says that rumors (yes, there is a joke there)may become more solid in the near future so I wanted to take my shot now and have a little fun with it. I did try, mostly, to stay within the confines of real possibility. I may have over-stepped a bit. Once more unto the breach, Dear friends! Or at least, once again back to the well? I'm going with MS and Sony playing to their strengths. Sony clones the PS4 and MS goes with a high/low approach that leverages their network strengths and looks to the XOX and 360 for inspiration. I'll put them in price order.

Note: Originally I did research 7nm and 7+nm, die space, transistor density, etc, but eventually dumped the frequency and die space numbers I came up with as ultimately unreliable and not particularly relevant. The higher frequency vs more cores debate, how much difference would the decision ultimately make? EULA - As always, worth every penny you paid to read it. No refunds on time lost.

$250 - Xbox Net
The price is more important than the hardware here. How much do you want to undercut the PS5? Frankly I don't think there is enough information to make a decent hardware guess. Beyond the statements that there will be some level of local hardware work such that a simple Chromecast style device with a controller packaged in will not be enough. For now.

I do believe MS will want to be able to appeal on a reduced initial outlay. There will be a subscription requirement. Much like the dreaded "cell phone" contract. Something introductory like getting 3 months free for your first 12-month contract. The low end will include Xbox Game Pass and more expensive services include other publishers or perhaps even some ala carte' by publishers? That could make the publishers rather happy. I really expect all kinds of options for how you pay for your games on here. You just won't own any of them.

A test on the Xbox website to see if your ISP service can handle the connection requirements would go a long way to alleviating fears about your connection quality. Be even nicer if you could run it in the background and let if accumulate a few days worth of tests and tell you what kind of quality you can expect. For those who do not fear the "games as service" model where you own nothing and you incur on-going costs rather than larger initial outlays, this is your console. I think Sony will let others try this out and take the risks rather than jump in right away with a low end cloud based box.

1Q 2020
$400 Sony PS5


The early adopters are going to buy about as many as you can make anyway. Launch in the Spring and get your ducks in a row for a blowout 2020 Fall with 1st party titles having an extra year to look good and get out the door. Could make for some nice holiday bundles as well. It could have launched over the 2019 holiday, but why? Just because it can? PS4 gets one more holiday season in the sun. I said everyone is going back to the well and I meant it. Sony leverages their first-party lineup, overall popularity and goes with what brung em' in the last generation.

128GB/ 1.5TB SATA SSHD - user replaceable. Will continue to work, only slower of course, if the SSD degrades over time/ use. With RDR2 already hitting the 100GB mark, I felt the 128GB was a near minimum.

7nm APU - 6C/12T Zen 2 + 12TF Navi - The "secret sauce" is something minor. Nothing cataclysmic like AI upscaling, RT or Path Tracing.

8GB DDR4 for OS/ App use - This is too handy a trick and I expect MS to use it as well.
16GB GDDR6 256-bit

4K BR / Full BC at launch / FreeSync

Seems boring? Nothing sexy here? It is straight-forward and an update on the idea of the PS4/Pro. Sony sells on their software library, price and enough power to get the job done.

Fall 2020
$500 Xbox 2 RT
(or whatever made up word the marketing department comes up with)

Back to the well again. I think MS has realized that, for the time being, they are just not going to be able to go head to head with Sony in 1st party titles. The 360 era was their heyday. Maybe you get lucky again but why count on it? Scorpio got them some respect back after the Xbox One stuttered. Take some lessons from what worked and run with it. The Xbox Net gives you a price competitive option. Now they go after the high ground. This may backslide into early 2021, but only on fab problems I would expect.

If you are going to take the high ground, perhaps something a little more than just raw power would be useful. I toyed with both RT and AI up-scaling (DLSS kind of tech). MS has no small amount of experience with AI. AMD and MS worked on the RT together already (along with others of course). Since "up-scaling" can be a dirty word, and I fully expect to hear plenty of "Native 4K!" marketing early, I went with RT.
Marketing matters here. There is already developer support for Nvidia's RT tech so I see no reason why millions of potential console customers wouldn't be enough for developers to support it. Especially if you are selling "RT enhanced" as part of your marketing. Engine support shouldn't be long in coming. I can already see the tag lines - Xbox 2 RT - the best place to play "insert fps du jour here".

256GB NVMe SSD - M.2 user replaceable. If this fails/ degrades past a certain point, it must be replaced. I expect heavy users would view this similar to replacing your disc drive. They always fail at some point. This being a requirement will open up some serious possibilities.

2TB SATA 7200rpm Mechanical - user replaceable
8GB DDR 4 for OS/ App use

16GB GDDR6 256-bit

7nm+ (Partial EUV) APU - 8C/16T Zen 3 + 16TF Post-Navi + 4GB HBM3 - RT enabled (How much die space dedicated to what kind of hardware specifically? I have no friggin' idea. This is just the direction I would expect MS to go. Maybe it is dedicated silicon. Maybe it is a modification to the existing GPU that gives it more punch for RT work.)

4K BR / Full BC at launch / FreeSync

An unanswered question. When does MS launch the Xbox Net? I did not put a launch date because I was torn. I expect this could be ready by the time the PS5 launches. Do you launch along side or wait and do a dual launch with the 2 RT? E3 2019 should be very interesting. My gut says, if you can, launch it alongside the PS5. You would have a lower priced competitor from the start and time to work out your kinks before possible major sales in the 2020 holiday season. Then MS brings the marketing for the next year getting people to wait on their PS5 purchase because yours will be higher quality on all those 3rd party titles come the Fall.
With chip design costs skyrocketing and major players stating this could be the last console generation with a move to cloud computing staring it in the face, who says their will even be shrinks? 5nm may not be that far away but for a $500m+ design cost, how many would you have to sell at a reduced price to be worth it? And when? We could be stuck with these consoles until either the major players give up on the "thin-client" idea as unworkable for the foreseeable future and release new hardware or the "cloud" remains just out of reach, but close enough they all keep trying for an extended time period. Some types of games will obviously translate better than others here. I feel I did go a bit overboard on the specs. For the Xbox almost certainly. But with the future shrinks uncertain it felt better to launch a more powerful device now since you are already going after a premium market with the $500 price point.

I hope Sony also joins the RT/hybrid bandwagon so RT/hybrid becomes the actual standard and as fast as possible rather than one still on just rasterization.

MS would be crazy to give up their current strategy of having the lowest cost of entry and the most premium console but there's still the Sony wildcard. Sony is rarely having the best gen launch specs.

As for streaming, wouldn't it be wiser for MS to invest more in current infrastructure such as XBL, One S, One X, etc. for the streaming part of Xbox RX? In fact should also consider all major distributions including Windows, iOS and Android, too.

If MS really need to have a new box, maybe they can even go lower and repackage some of the One S/X to have the next gen XRX controllers?

Is MS rumored to be cooking a new tech only available for new and unreleased hardware? Maybe this can also do the whole Cloud thing better compared to entrance of current gen if MS is still serious with such.
 
As for streaming, wouldn't it be wiser for MS to invest more in current infrastructure such as XBL, One S, One X, etc. for the streaming part of Xbox RX? In fact should also consider all major distributions including Windows, iOS and Android, too.

If MS really need to have a new box, maybe they can even go lower and repackage some of the One S/X to have the next gen XRX controllers?

Yup, I don't see why they'd release another new box when they already have 3 on the market with sufficient power to stream. If my XZ2 can Remote Play perfectly, I don't see why a ~1.3TF console couldn't.

I'm firmly of the opinion that MS's next generation strategy should be the following:
- Release 7nm versions of both existing consoles. No ODD in the base model, and maybe an SSD instead of HDD.
- Focus on making the streaming service work with the XBoxOne and up.
- Make the XBoxOneX the 1080p30 console for the Scarlet platform.
- Make the XBoxTwo the 4K60 console for the Scarlet platform.

That way, they don't really have to pay much attention to what Sony are doing, because MS will already have a next generation console on the market. And their streaming box for the next generation could play games locally too, rather than just being an overly expensive Firestick.
 
I'm inclined to believe so. Considering that Xbox One already has this (to a degree), they're late to the game if they don't have it for ps5.
 
Thought this was shelved as a concept for the time being.
They had trouble enabling it on Vega, but the belief is that all the broken features of Vega will be fixed for Navi. Nvidia is doing the same concept with mesh shaders which already has a Vulkan extension, so they’ll be behind the curve as iroboto pointed out. I guess the interesting thing is that the patent is dated 2014, which means Sony’s seemingly had an active role in AMD’s GPU roadmap since PS4 was conceptualized.
 
They had trouble enabling it on Vega, but the belief is that all the broken features of Vega will be fixed for Navi. Nvidia is doing the same concept with mesh shaders which already has a Vulkan extension, so they’ll be behind the curve as iroboto pointed out. I guess the interesting thing is that the patent is dated 2014, which means Sony’s seemingly had an active role in AMD’s GPU roadmap since PS4 was conceptualized.

Now that is interesting. Sparks a question: It was my understanding that going to "primitive shaders", or a rose by any other name, would mean a massive load off the CPU. Would that mean one would go with a much smaller CPU and a heavier GPU mm budget on an APU? Or am I conflating "primitive shaders" with a different concept?
 
Now that is interesting. Sparks a question: It was my understanding that going to "primitive shaders", or a rose by any other name, would mean a massive load off the CPU. Would that mean one would go with a much smaller CPU and a heavier GPU mm budget on an APU? Or am I conflating "primitive shaders" with a different concept?
No, that’s how it has been advertised to me. Make the CPU more than just a draw call machine. What’s unclear to me is how much burden can be lifted off the CPU, and how this affects GPU simulation workload.

One would think with more cycles free, the CPU would be able to dedicate those cycles to more world events, AI behaviors, etc.
 
Thank you both. Though I fear the answers from sebbbi are a bit beyond my understanding. It feels just a touch out of reach for my knowledge.

Could have sworn I saw a graph of CPU usage that depicted an incredible percentage of usage that going to primitive shaders would eliminate or vastly reduce. Sadly I cannot find it. Probably wasn't correct anyway. Or I was just misunderstanding it.
 
The idea of GPU side dispatch as moving forward into next generation imo. is a win for developers and reinforces the idea of having small CPU and large GPU. Which is why I don't think we'll need a large CPU for next generation even though there is a large desire for higher FPS or better world interactions.
 
Status
Not open for further replies.
Back
Top