megreotsugua
Banned
Pro versions will happen. It's just a matter of when.
The problem with production will settle down eventually.
The problem with production will settle down eventually.
Fabs are under construction. They will be online by 2023 atleast near the end of it.No chance, silicon is in massive shortage and will be for sometime to come. Just not logistically possible to produce another console when the original is still in short supply.
This reads like a fanfic someone wrote for the next gen spec thread. Brings back weird memories from early 2020....
"This year Sony is doing a redesign for cost. Next year, I think there is evidence there could be a slightly slimmer console, or at least a 6nm chip-based cost reduction — although, I can't confirm it. 2023 is the year, then, that it makes sense to bring in a stronger unit — this would be after RDNA 3 is out on PC, and near enough to RDNA 4 for some updates to be considered from that architecture, like Sony used some of 2017's Vega in the 2016 PS4 Pro."
...
"I think you are looking at something 50% to 100% stronger. They aim for a $700 price point, though? I mean, I could see them really making something crazy powerful. It all comes down to higher resolutions and frame rates, I think."
...
"I expect [Sony] to at least double GPU performance, or even higher if they charged $700 and waited till late 2023. However, I wouldn't be sure they will go that higher. [I would expect them to] add more RAM, effectively increase bandwidth by at least 100% (this could just be an addition of an infinity cache), increase CPU performance by 20% plus," says Tom. "This would be enough for easy 4K/120fps with current graphics, and 8K/60fps in many games if Sony wants to market that for their new TVs." Remember, though: "games get more complex, games get harder to run. It is a moving target."
...
"Though, things have changed since last gen — a PS5 Pro that's 2-3x the performance of a PS5 Base would definitely cost more, due to TSMC 5nm costing more than 7nm," Tom reasons. "So, in two years the iteration will happen. Again, two years is a long time — what games were you playing two years ago?"
...
This reads like a fanfic someone wrote for the next gen spec thread. Brings back weird memories from early 2020.
People are not paying up front for their expensive phones though, they pay monthly to their current carrier. If carriers said ok that iPhone/galaxy whatever will be 999$ up front, most people would laugh and say nope. Carriers make it easy to pay off the phones over time with various deals and such, this is not the case with consoles as of now. MS has a pay off plan but Sony and Nintendo do not.
It does read like first half of the infamous "next-gen speculation thread" with all the clueless "engineers" spamming 16-24 core CPU, 24-32GB HBM3, 20-24TF GPU etc. No one will make a 700$ console to begin with, if anything Sony and MS would want to make even cheaper consoles than they already are, and most importantly the technology doesn't progress at such pace anymore, 2x increase across the board probably won't even happen in PS6/XS2, let alone mid-gen refresh (if they happen).
Let's be honest, the only limiting factor in current consoles are the GPUs, both raster and RT performance, so if by any chance we could get a newer revisions made in 5nm process, I believe it's the GPU where most if not all the advancements would go, more CUs and RDNA3 architecture, so that you won't have to make a compromise between 4K, 60 FPS, and RT anymore, but get the best possible experience, that would make a selling point/marketing sense.
Like you stated MS lets you pay $35 a month for a Series X and GP.People are not paying up front for their expensive phones though, they pay monthly to their current carrier. If carriers said ok that iPhone/galaxy whatever will be 999$ up front, most people would laugh and say nope. Carriers make it easy to pay off the phones over time with various deals and such, this is not the case with consoles as of now. MS has a pay off plan but Sony and Nintendo do not.
we already have games like doom and metro drop down to 1080p on these consoles under load. They are more powerful but clearly not capable of 4k 60 fps.There's no need for a pro model unlike the last gen where the consoles were very underpowered
I don't think we are getting Pro consoles this generation no matter how many people are acting like it's inevitable.
dude people still cant buy the ps5 and xsx. This gen is gonna be a long one. And these consoles are really good. Why are we even talking about new ones already
There's no need for a pro model unlike the last gen where the consoles were very underpowered
I was kinda hoping there would be no PRO consoles this time... Feels pointless and against the console lifestyle of "buy and forget".
And these consoles are really powerful and "hot"
I agree. Doesn't feel like Pro consoles are needed this time. Both feel powerful enough.I don't think there is going to be a mid gen refresh this generation. I know that a lot of people assume there will be because there was one previous gen, but that was still the exception rather than the rule for home consoles.
Here is why I don't think there will be one:
- There's no marketing hook. Both the PS4 Pro and XoneX were sold with the promise that they would deliver 4k gaming. Now whether they actually achieved this is up for debate, but the PS5 and XSX already claim to support 4k gaming. So what's the marketing hook going to be? There's no resolution bump, games already run at a solid framerate.
- PS4 Pro and X one X continued to be outsold by the standard models after they came out. Most consumers still went for the cheaper options
- You make you the people who won't upgrade feel like 2nd rate customers because they're getting a lesser experience. One of the benefits of buying a game on a console is knowing that you're, more or less, playing the best version of the game outside of playing it on a higher end gaming PC.
I don't think we'll see another upgrade. We will probably see a redesign of both consoles, especially of the PS5.
A simple revision will suffice, just like previous generationsThese consoles released with way too small hard drives, the refreshes need to rectify that.
These consoles released with way too small hard drives, the refreshes need to rectify that.
https://www.truetrophies.com/n19956/ps5-pro-moores-law-is-dead-interview
Man they have got to be joking. Most of the folks who want these systems today can't get one. So if these rumors are true we are looking at 2023 when we are not really going to get fully realized next gen games until 2022. What do you guys think ---could they be referring to redesigns of the current systems to make them smaller/cheaper or full on upgrades?
My series X is def already full and I have basically no games.I don't know about that anymore tbh. Except for the Series S, feels like Series X and even PS5's SSDs are big enough, at least for a while.
PS5 for example there is compression helping certain games get down to 20 GB in file size. Then on the Xbox side of things there's selective parts of some games you can uninstall cutting down on install sizes.
My series X is def already full and I have basically no games.
and won't charge you for the upgradesI’m more inclined to upgrade to series x pro because I feel like MS would do a much better job upgrading the games to use the extra power.
No chance, silicon is in massive shortage and will be for sometime to come. Just not logistically possible to produce another console when the original is still in short supply.
DF's Rich leadbetter has said that silicon/die shrinks are not going to reduce in price like they have done in prior gens.
However I dont see why smaller manufacturing (3nm) processes cant be the same price as 7nm if they are the same size chip.
So, a 3nm 310mm² SOC with roughly 24 billion transistors, should cost the same as a 7nm 310mm² 12 billion transistor SOC.
If this turns out to be what happens, pro variants with double the GPU is possible at $499 when 3nm is available to sony/ms.
The only issue with pro consoles this gen is that next gen may seem disappointing in comparison. If all goes well and 1nm is available to sony/ms in 2027, we would be looking at a 40-50tflop PS6 with a 310mm SOC, maybe skipping the pro would make the PS6/XSX2 have a bigger impact.
Moving from crappy 8 core notebook CPUs and mechanical hard drives was a big leap. Next gen i dont think the leap will be as big, I expect, 24-32gb gddr7 @1 - 2tb/s bandwidth, 30gb/s ssd bandwidth and about double the CPU performance. I wonder if that would enable games that could not work on PS5/XSX.
Given the jump from base PS4/One to Pro/One X, is it fair to say a mid gen refresh could mean PS5 Pro and Series X Pro at 20 TF by 2023/24?
Just watched this; I still think the 2023 date is too aggressive. Maybe that was a target pre-COVID, but now? I wouldn't be surprised if both Microsoft and Sony hold off for mid-gen refreshes until 2024 simply because of the increased load new systems would bring to game developers in optimizing their games for their features in time.
He seems to have two different perspectives of a PS5 Pro but I don't see the $700 option being too likely. Of late Sony seems to be wanting to take the Nintendo approach: cheaper-to-produce hardware sold at cost and at a profit as soon as possible. They're on record saying they're watching how the Switch OLED performs, and we all know how that (well, IMO) isn't really doing too much more in justifying the price, and will be sold at a slight profit at the very least.
That's the model it seems Sony wants to emulate going forward, so how does a $700 PS5 Pro fit into that? Well, IMO it doesn't, and that's even considering they sell it at a profit, because it would still necessitate additional production costs that Sony might not want to go for. I'm starting to see why they had such an internal pushback against doing a PS4 Pro; it effectively cut out some of their production capacity for the more profitable base PS4, and increased their overall production volumes (they had to satisfy demand for PS4 and now add in demand for PS4 Pro on top of that). That's extra money they'd probably rather had not spent.
Which is why I agree more with his idea on a new PS5 that takes up the original disc model's price bracket. They could use the 6nm process for both this PS5 and a slimmer model that just takes the old design and moves it to the 6nm process, with that one becoming the new $399 (maybe $349?) entry model, probably discless in order to save on further production costs. Meanwhile that other PS5 "Pro" would be the same basic architecture: 36 CUs (although the other 4 could potentially be activated), 64 ROPs, etc. Probably even the same SSD spec, but from there make some changes.
A while ago I did some speculations on mid-gen refreshes and for PS5's I talked about RAM capacity staying the same but the speed increasing, and a slight TF bump-up. I've since had some changes of thought there. HeisenbergFX4 just linked this , and I think it hints (potentially) at what an actual PS5 "Pro" ends up being like for Sony, if you combine it with the other stuff I just talked about. I think they're going to keep RAM capacity the same (16 GB), but they might go for really fast GDDR6X; back a while ago I was thinking 16 Gbps but now I think they could go with 20 Gbps. That'd give them 640 GB/s of bandwidth on the same 256-bit bus.
For a second I was thinking they'd add IC, but I strongly doubt that now. Why? Because then Sony'd have to add that to the base PS5 model as well, or risk splitting up production capacity between, effectively, two APU design, one without IC and one with it, which might defeat the purpose of where I think they're trending. They could more easily get that bandwidth increase by swapping out RAM modules between the two models, which wouldn't impact actual APU wafer production capacities or splinter them.
The basic bandwidth-per-TF figure for PS5 as-is, is around 43.6 GB/per TF. Sony would probably want to at least maintain that ratio for a PS5 "Pro", so if the bandwidth increases to 640, you could achieve that while getting the TF performance to close to 14 TF; 13.824 TF to be precise. Which means they could either keep the number of active CUs in a PS5 "Pro" at 36 but bump the clock up to 3 GHz (gets you 13.824 TF), or increase the active count of CUs to 40 (the full chip) while increasing the clock to 2.7 GHz (also gets you 13.824 TF). The latter is the safer option insofar as RDNA 2 (these would still be RDNA 2 GPUs btw) because IIRC some folks have overclocked their cards and gotten close to 2.7 GHz already, and that's therefore something which could be a lot more feasible on 6nm process.
However, if you go back to that link HeisenbergFX4 posted, if Sony are trying to focus more on machine learning and RT performance going forward, both would probably benefit more from higher clocks, although ML would probably benefit more from a wider net of active CUs, of the two. 2.7 GHz would get them 420 billion BVH intersections per second, but only if all 40 CUs are active. They could get that same figure with the higher 3 GHz clock but keeping to 36 active CUs.
The same figures hold true for their texel fillrate figures. However, a design with only 36 CUs but higher clocks also benefits pixel fillrate here in a way not true with the 40 CU design, because in both cases your ROPs will still be at 64. Again though, going by that provided link and considering base PS5 isn't really hurting for pixel fillrates anyway, what bump they get there from the 40 CU design (172.8 Gpixels/sec) is probably a good enough tradeoff.
Meanwhile in terms of any extra hardware features, they may bring some things related to certain hardware acceleration and machine learning in RDNA 3 into their design, but only if they're cheap enough, and can reasonably serve both an unchanged PS5 spec and the "Pro" spec. The reason I rule out IC is, for one, it'd increase die size considerably (6nm doesn't offer any die reduction over 7nm), and secondly capacity-to-costs benefit is probably reduced for a Sony that wants to use the same APU design on 6nm process with different clock settings for various components. Instead of paying 100% of the time for an IC cost regardless of the PS5 model, they can pay say only 20% of the time for the faster GDDR6X they'd need for PS5 "Pro" units, still get the bandwidth increase benefit, and get the additional flexibility that comes with all of that.
So there you go, that's your PS5 "Pro"; 2024 (most likely, slight chance of 2023 tho), $499, 6nm, 13.824 TF @ 2.7 GHz, 432 Gtexels/sec, 172.8 Gpixels/sec, 16 GB GDDR6X, 640 GB/s, 5.5 GB/s SSD likely at 1 TB capacity. Meanwhile, you get your PS5 "full" revision, possibly 2023, $399 with all the same specs as base PS5 but lower power consumption, and discless (they would still keep disc PS5 on 7nm for the interim, then shift the disc edition to the PS5 "Pro" while the revised PS5 remains discless-only). If Sony could manage to bring PS5 "Pro" to market in 2023, then they could ditch 7nm that much sooner across the board.
As for his Microsoft stuff, well he had virtually nothing on Series X refresh other than that it's in fact coming. The truth is I don't even know what they could do for a Series X refresh IMHO anymore. I did post some specs in the past but they weren't realistic because I severely undershot some things like ROP counts and active CUs they'd need to run 4x Series S instances on one system. Even without changing the clock speed of the GPU, you'd be getting over 37 TF with that type of setup, which is well past mid-gen refresh territory and closer to what 10th-gen hardware (on the lower end) would be targeting, IMO. So I think they might just aim for something that can stimulate 2x Series S instances but at a spec where 1440p60 is more of a true consistent baseline, so probably around 6 TF each like his Series S refresh was hinting (I'll get to that in a minute).
I think Microsoft is going to try making a discless Series X and one at a lower price of $399 by 2023 or early 2024. They wouldn't need to do much to match a PS5 "Pro" in basic raw specs if they also shift to 6nm; simply increase the clock to 2.1 GHz and they already get 13.9 TF (essentially 14 TF) right there. Removing the disc drive would help in keeping thermals they'd want even with having a larger APU, and it's not like 2.1 GHz would be stressing an RDNA 2 design (in fact it's still lower than the Game Clock of RDNA 2 GPUs). However, this would still be a Series X with 16 GB; there's room for MS to increase the capacity to 20 GB tho they would save this for a $499 model possibly also still keeping the disc drive, and that would be the one with the increased clock on the GPU, while keeping the same 1.825 GHz clock on the cheaper discless model.
So that's what Microsoft could do for a Series X refresh: $399 discless with same specs as base Series X, and a "Series X-2", $499 model on 6nm with disc drive, 52 CUs, 13.97 TF @ 2.1 GHz, 20 GB GDDR6 @ 16 Gbps for 640 GB/s, and a slot for M.2 SSD storage (NVMe 2.0, PCIe Gen 4, 8 GB/s support) while shipping with a 1 TB drive @ 4 GB/s and having legacy support for the 2.4 GB/s expansion cards. Now, they could be cheeky and push the GPU clocks higher than 2.1 GHz, but by that point cooling would need big reinvestment again and probably wouldn't be worth the cost. Power is not even an issue with base Series X (or base PS5, for that matter), but I think all devs would prefer for a more standard RAM pool so capacity increase to 20 GB would be appreciated.
In terms of certain performance metrics, a $499 Series X refresh like this would be getting 134.4 Gpixels/sec, notably lower than PS5 "Pro" and even base PS5, but if things are moving away from the traditional rasterization pipeline anyway, this metric starts to fall to the wayside of importance. BVH intersection rates (for RT) would increase to 436.8 G-BVH intersections/sec, and texel fillrate would increase to 436.8 Gtexels/sec. Bandwidth performance would be in parity between PS5 "Pro" and Series X-2, but if something like cache scrubbers are still not in the Series design (and unless AMD standardized cache scrubbers in RDNA 3, they wouldn't be) then Microsoft probably would go for slightly faster GDDR6 modules, say 18 Gbps, to bump their bandwidth up to 720 GB/s, but that's all relative. In any case tho as you could see, Microsoft would have a RAM capacity advantage, which could be beneficial for their design (it'd also allow them to better service 2x Series S instances since Series S uses 10 GB of RAM).
Finally on to his Series S stuff...I kinda don't see this happening. If Microsoft wants to push Series S as a secondary console to PS5 and/or PC (or Switch) gamers, I don't see how a more powerful Series S that still sits at $299 (or did he say $349 in his video?) accomplishes that, especially if Sony pushes a PS5 revamp into $349 territory. It would probably also be too expensive for a Switch gamer wanting one as a secondary platform, though maybe have some benefit to PC gamers wanting an entry-level console...provided they would want a console at all, particularly a Series S of any type if they already have GamePass on PC and Microsoft's revamp of the Microsoft Store is genuine and big-time.
That's why to me it makes more sense they move Series S to 6nm too (this part as he said in the vid), but get a smaller Series S with same specifications (but potentially increase SSD capacity) to $249 or even $199, while they spin that design off for themselves and AMD in the PC space. There's too much potential for Microsoft to put some of that 6nm Series S production into a new Surface product, meanwhile AMD could spin it into an entry-level APU package they license out to OEM laptop and desktop makers. All of that translating to more profits justifying mass-scale production of Series S SoCs on 6nm combined with production of Series X SoCs on the 6nm process too (some going towards a discless Series X, others towards an up-spec'd $499 Series X). Provided wafer capacities could be fulfilled for all of that, of course.
You wouldn't need 1nm for 40-50 TF 10th-gen consoles unless you want notably lower TDP and die size footprint for them compared to PS5 and Series X. You can get 40-50 TF performance at similar TDPs of 9th-gen consoles (which increased their overall TDPs over 8th-gen btw) on 3nm EUV, with the benefit that 3nm EUV would be incredibly mature by 10th-gen with much larger capacities and less congestion from manufacturers (i.e Apple) fighting for them compared to 1nm.
If in terms of just raw numbers yes in some ways 10th-gen will not be as big a leap as 9th was from 8th. However, it's more about development of new technologies for ML, various types of hardware acceleration and image upscaling that will be the big improvement areas. With those seeing gains you won't need as much raw "generic" compute power as we're seeing GPUs seemingly need currently. I also think GDDR7 will probably be skipped in favor of something HBM3-derived if HBM3 can get to market on time (between 2022 and early 2024 at latest), because at that point you get that bandwidth at lower power consumption, better latency and wider bus widths.
But I also think just more of a power increase of any kind is not going to be enough for 10th-gen.
I think this might hint towards a direction they're taking with a PS5 mid-gen update, but it sounds like they want to invest more in hardware-agnostic (i.e it wouldn't need to rely solely on extra hardware in a PS5 Pro) approaches for bettering RT and ML performance across the PS5 family of devices.
A pretty similar approach in a lot of ways to what Microsoft has already been doing for a while, tbh.
It's certainly possible, but given Sony's trajectory with design trends and the Nintendo-like production model they seemingly want to adapt, quite unlikely. I honestly think some of those other patents were in reference to cloud network server implementations, not a mid-gen refresh. At most, probably concepts of chiplet-like designs for a 10th-gen PlayStation that are still pretty early.
That is an option.Would rather they skip a PS5 Pro and just do a slim in 2023.
Then just launch PS6 in 2026 or 2027.
PS5 slim could still keep pricing up like this:That is an option.
And weve seen that before. Slim model is smaller and cheaper to buy. Or they do mid gen refresh and try to keep pricing up. At best the base model gets cheaper, but not slimmed.
Sucker?Sucker born every generation I guess.
Just watched this; I still think the 2023 date is too aggressive. Maybe that was a target pre-COVID, but now? I wouldn't be surprised if both Microsoft and Sony hold off for mid-gen refreshes until 2024 simply because of the increased load new systems would bring to game developers in optimizing their games for their features in time.
He seems to have two different perspectives of a PS5 Pro but I don't see the $700 option being too likely. Of late Sony seems to be wanting to take the Nintendo approach: cheaper-to-produce hardware sold at cost and at a profit as soon as possible. They're on record saying they're watching how the Switch OLED performs, and we all know how that (well, IMO) isn't really doing too much more in justifying the price, and will be sold at a slight profit at the very least.
That's the model it seems Sony wants to emulate going forward, so how does a $700 PS5 Pro fit into that? Well, IMO it doesn't, and that's even considering they sell it at a profit, because it would still necessitate additional production costs that Sony might not want to go for. I'm starting to see why they had such an internal pushback against doing a PS4 Pro; it effectively cut out some of their production capacity for the more profitable base PS4, and increased their overall production volumes (they had to satisfy demand for PS4 and now add in demand for PS4 Pro on top of that). That's extra money they'd probably rather had not spent.
Which is why I agree more with his idea on a new PS5 that takes up the original disc model's price bracket. They could use the 6nm process for both this PS5 and a slimmer model that just takes the old design and moves it to the 6nm process, with that one becoming the new $399 (maybe $349?) entry model, probably discless in order to save on further production costs. Meanwhile that other PS5 "Pro" would be the same basic architecture: 36 CUs (although the other 4 could potentially be activated), 64 ROPs, etc. Probably even the same SSD spec, but from there make some changes.
A while ago I did some speculations on mid-gen refreshes and for PS5's I talked about RAM capacity staying the same but the speed increasing, and a slight TF bump-up. I've since had some changes of thought there. HeisenbergFX4 just linked this , and I think it hints (potentially) at what an actual PS5 "Pro" ends up being like for Sony, if you combine it with the other stuff I just talked about. I think they're going to keep RAM capacity the same (16 GB), but they might go for really fast GDDR6X; back a while ago I was thinking 16 Gbps but now I think they could go with 20 Gbps. That'd give them 640 GB/s of bandwidth on the same 256-bit bus.
For a second I was thinking they'd add IC, but I strongly doubt that now. Why? Because then Sony'd have to add that to the base PS5 model as well, or risk splitting up production capacity between, effectively, two APU design, one without IC and one with it, which might defeat the purpose of where I think they're trending. They could more easily get that bandwidth increase by swapping out RAM modules between the two models, which wouldn't impact actual APU wafer production capacities or splinter them.
The basic bandwidth-per-TF figure for PS5 as-is, is around 43.6 GB/per TF. Sony would probably want to at least maintain that ratio for a PS5 "Pro", so if the bandwidth increases to 640, you could achieve that while getting the TF performance to close to 14 TF; 13.824 TF to be precise. Which means they could either keep the number of active CUs in a PS5 "Pro" at 36 but bump the clock up to 3 GHz (gets you 13.824 TF), or increase the active count of CUs to 40 (the full chip) while increasing the clock to 2.7 GHz (also gets you 13.824 TF). The latter is the safer option insofar as RDNA 2 (these would still be RDNA 2 GPUs btw) because IIRC some folks have overclocked their cards and gotten close to 2.7 GHz already, and that's therefore something which could be a lot more feasible on 6nm process.
However, if you go back to that link HeisenbergFX4 posted, if Sony are trying to focus more on machine learning and RT performance going forward, both would probably benefit more from higher clocks, although ML would probably benefit more from a wider net of active CUs, of the two. 2.7 GHz would get them 420 billion BVH intersections per second, but only if all 40 CUs are active. They could get that same figure with the higher 3 GHz clock but keeping to 36 active CUs.
The same figures hold true for their texel fillrate figures. However, a design with only 36 CUs but higher clocks also benefits pixel fillrate here in a way not true with the 40 CU design, because in both cases your ROPs will still be at 64. Again though, going by that provided link and considering base PS5 isn't really hurting for pixel fillrates anyway, what bump they get there from the 40 CU design (172.8 Gpixels/sec) is probably a good enough tradeoff.
Meanwhile in terms of any extra hardware features, they may bring some things related to certain hardware acceleration and machine learning in RDNA 3 into their design, but only if they're cheap enough, and can reasonably serve both an unchanged PS5 spec and the "Pro" spec. The reason I rule out IC is, for one, it'd increase die size considerably (6nm doesn't offer any die reduction over 7nm), and secondly capacity-to-costs benefit is probably reduced for a Sony that wants to use the same APU design on 6nm process with different clock settings for various components. Instead of paying 100% of the time for an IC cost regardless of the PS5 model, they can pay say only 20% of the time for the faster GDDR6X they'd need for PS5 "Pro" units, still get the bandwidth increase benefit, and get the additional flexibility that comes with all of that.
So there you go, that's your PS5 "Pro"; 2024 (most likely, slight chance of 2023 tho), $499, 6nm, 13.824 TF @ 2.7 GHz, 432 Gtexels/sec, 172.8 Gpixels/sec, 16 GB GDDR6X, 640 GB/s, 5.5 GB/s SSD likely at 1 TB capacity. Meanwhile, you get your PS5 "full" revision, possibly 2023, $399 with all the same specs as base PS5 but lower power consumption, and discless (they would still keep disc PS5 on 7nm for the interim, then shift the disc edition to the PS5 "Pro" while the revised PS5 remains discless-only). If Sony could manage to bring PS5 "Pro" to market in 2023, then they could ditch 7nm that much sooner across the board.
As for his Microsoft stuff, well he had virtually nothing on Series X refresh other than that it's in fact coming. The truth is I don't even know what they could do for a Series X refresh IMHO anymore. I did post some specs in the past but they weren't realistic because I severely undershot some things like ROP counts and active CUs they'd need to run 4x Series S instances on one system. Even without changing the clock speed of the GPU, you'd be getting over 37 TF with that type of setup, which is well past mid-gen refresh territory and closer to what 10th-gen hardware (on the lower end) would be targeting, IMO. So I think they might just aim for something that can stimulate 2x Series S instances but at a spec where 1440p60 is more of a true consistent baseline, so probably around 6 TF each like his Series S refresh was hinting (I'll get to that in a minute).
I think Microsoft is going to try making a discless Series X and one at a lower price of $399 by 2023 or early 2024. They wouldn't need to do much to match a PS5 "Pro" in basic raw specs if they also shift to 6nm; simply increase the clock to 2.1 GHz and they already get 13.9 TF (essentially 14 TF) right there. Removing the disc drive would help in keeping thermals they'd want even with having a larger APU, and it's not like 2.1 GHz would be stressing an RDNA 2 design (in fact it's still lower than the Game Clock of RDNA 2 GPUs). However, this would still be a Series X with 16 GB; there's room for MS to increase the capacity to 20 GB tho they would save this for a $499 model possibly also still keeping the disc drive, and that would be the one with the increased clock on the GPU, while keeping the same 1.825 GHz clock on the cheaper discless model.
So that's what Microsoft could do for a Series X refresh: $399 discless with same specs as base Series X, and a "Series X-2", $499 model on 6nm with disc drive, 52 CUs, 13.97 TF @ 2.1 GHz, 20 GB GDDR6 @ 16 Gbps for 640 GB/s, and a slot for M.2 SSD storage (NVMe 2.0, PCIe Gen 4, 8 GB/s support) while shipping with a 1 TB drive @ 4 GB/s and having legacy support for the 2.4 GB/s expansion cards. Now, they could be cheeky and push the GPU clocks higher than 2.1 GHz, but by that point cooling would need big reinvestment again and probably wouldn't be worth the cost. Power is not even an issue with base Series X (or base PS5, for that matter), but I think all devs would prefer for a more standard RAM pool so capacity increase to 20 GB would be appreciated.
In terms of certain performance metrics, a $499 Series X refresh like this would be getting 134.4 Gpixels/sec, notably lower than PS5 "Pro" and even base PS5, but if things are moving away from the traditional rasterization pipeline anyway, this metric starts to fall to the wayside of importance. BVH intersection rates (for RT) would increase to 436.8 G-BVH intersections/sec, and texel fillrate would increase to 436.8 Gtexels/sec. Bandwidth performance would be in parity between PS5 "Pro" and Series X-2, but if something like cache scrubbers are still not in the Series design (and unless AMD standardized cache scrubbers in RDNA 3, they wouldn't be) then Microsoft probably would go for slightly faster GDDR6 modules, say 18 Gbps, to bump their bandwidth up to 720 GB/s, but that's all relative. In any case tho as you could see, Microsoft would have a RAM capacity advantage, which could be beneficial for their design (it'd also allow them to better service 2x Series S instances since Series S uses 10 GB of RAM).
Finally on to his Series S stuff...I kinda don't see this happening. If Microsoft wants to push Series S as a secondary console to PS5 and/or PC (or Switch) gamers, I don't see how a more powerful Series S that still sits at $299 (or did he say $349 in his video?) accomplishes that, especially if Sony pushes a PS5 revamp into $349 territory. It would probably also be too expensive for a Switch gamer wanting one as a secondary platform, though maybe have some benefit to PC gamers wanting an entry-level console...provided they would want a console at all, particularly a Series S of any type if they already have GamePass on PC and Microsoft's revamp of the Microsoft Store is genuine and big-time.
That's why to me it makes more sense they move Series S to 6nm too (this part as he said in the vid), but get a smaller Series S with same specifications (but potentially increase SSD capacity) to $249 or even $199, while they spin that design off for themselves and AMD in the PC space. There's too much potential for Microsoft to put some of that 6nm Series S production into a new Surface product, meanwhile AMD could spin it into an entry-level APU package they license out to OEM laptop and desktop makers. All of that translating to more profits justifying mass-scale production of Series S SoCs on 6nm combined with production of Series X SoCs on the 6nm process too (some going towards a discless Series X, others towards an up-spec'd $499 Series X). Provided wafer capacities could be fulfilled for all of that, of course.
You wouldn't need 1nm for 40-50 TF 10th-gen consoles unless you want notably lower TDP and die size footprint for them compared to PS5 and Series X. You can get 40-50 TF performance at similar TDPs of 9th-gen consoles (which increased their overall TDPs over 8th-gen btw) on 3nm EUV, with the benefit that 3nm EUV would be incredibly mature by 10th-gen with much larger capacities and less congestion from manufacturers (i.e Apple) fighting for them compared to 1nm.
If in terms of just raw numbers yes in some ways 10th-gen will not be as big a leap as 9th was from 8th. However, it's more about development of new technologies for ML, various types of hardware acceleration and image upscaling that will be the big improvement areas. With those seeing gains you won't need as much raw "generic" compute power as we're seeing GPUs seemingly need currently. I also think GDDR7 will probably be skipped in favor of something HBM3-derived if HBM3 can get to market on time (between 2022 and early 2024 at latest), because at that point you get that bandwidth at lower power consumption, better latency and wider bus widths.
But I also think just more of a power increase of any kind is not going to be enough for 10th-gen.
I think this might hint towards a direction they're taking with a PS5 mid-gen update, but it sounds like they want to invest more in hardware-agnostic (i.e it wouldn't need to rely solely on extra hardware in a PS5 Pro) approaches for bettering RT and ML performance across the PS5 family of devices.
A pretty similar approach in a lot of ways to what Microsoft has already been doing for a while, tbh.
It's certainly possible, but given Sony's trajectory with design trends and the Nintendo-like production model they seemingly want to adapt, quite unlikely. I honestly think some of those other patents were in reference to cloud network server implementations, not a mid-gen refresh. At most, probably concepts of chiplet-like designs for a 10th-gen PlayStation that are still pretty early.
Whatever happened to squeezing every last inch of power from a system, designing around that features set for years to punch above its weight.