Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
Sony announces PlayStation 5 specs
https://techreport.com/news/3469318/playstation-5-specs-reveal/
Like the next Xbox, the PS5 will have eight Zen 2 cores. It will clock them a little lower, at 3.5 GHz rather than 3.8 GHz for the Xbox. But CPUs are so heavy on turbo that the nominal clock speed is fairly dubious. Given that both will have the same CPU cores, if there's a performance difference, it will be due to the amount of power that the CPU is allowed to burn.
On the memory side of things, both the PS5 and the next Xbox have 16 GB of GDDR6. The two consoles handle it differently, however. Sony does the simple, obvious thing: four identical 64-bit channels with 4 GB each, for 448 GB/sec of bandwidth. You get the same bandwidth regardless of how you use it.
As best as I can tell, the Xbox has a 320-bit memory bus, which breaks down unevenly as three 64-bit channels with 4 GB each and two with 2 GB each. That means that if you stay inside of 10 GB, you can get up to 560 GB/sec of memory bandwidth. Go over that, and you only use three of the channels, for 336 GB/sec of bandwidth. That's a lot like the awful shenanigans that Nvidia used to play before the class action lawsuit over the GTX 970 got them to stop. AMD hasn't done that in the past, and I'd bet that Microsoft pushed for it against AMD's advice.
There are a lot of circumstances where, if you really, carefully optimize things specifically for the new Xbox, it will win on memory bandwidth. In spite of that, I'm going to say that the PS5 has the better memory subsystem, due to its simplicity. Programmers hate having to manually put things into different memory pools if there's no good reason for it, especially when the two memory pools will trip over each other because of a lot of shared bandwidth. Apparently Microsoft didn't learn its lesson with the ESRAM debacle on the Xbox One.
As for the GPU itself, both use AMD's upcoming RDNA2 architecture, now branded as Navi 2X. The next Xbox has 52 compute units at 1.825 GHz, for about 12.1 TFLOPS. The PS5 has 36 compute units at 2.23 GHz, for about 10.3 TFLOPS. So on raw performance, the next Xbox wins.
If you read my link above, Sony has some spin as to why their approach is better. Their spin is garbage and they know it. It basically translates to "ours isn't as good, but we don't want you to realize that". More compute units clocked lower wins, because you can get the same performance in less power, in part because the lower clock speed allows lower voltages. By relying on such a high clock speed, there's also a significant risk that TSMC just can't manufacture it to be able to run at the intended clock speed with acceptable yields, and that TFLOPS gap widens further when they have to clock the chip lower. Sony's approach does have the advantage of a smaller die being cheaper to build, which is why they did it that way.
The two vendors are both offering a custom PCI-E over NVMe SSD, but beyond that, they're taking substantially different approaches. Sony is going to fit the PS5 with an SSD that offers high end performance. It would probably be the fastest M.2 drive on the market if it were available today. And they're telling developers that you can assume that the PS5 will have this much performance and it's fine to make a game that doesn't work properly if paired with a slower SSD. Sony will allow you to add your own SSD, but if it's slower than the one you're using, some games might not work at all.
Microsoft, on the other hand, is using a much slower SSD, and possibly only PCI-E 3.0. Meanwhile, Microsoft is saying that you can add your own SSD for more capacity and it will just work. That is, Microsoft isn't allowing game developers to assume high end SSD performance. They've surely got some sort of minimum specs, but they might be set such that every single PCI-E over NVMe SSD ever made is good enough.
I'm going to call that a win for Microsoft. There just isn't much to gain from games assuming that the SSD can transfer data at 5 GB/sec rather than "only" 2 GB/sec. Having real SSD speed rather than the speed of a hard drive, or worse, an optical drive like Blu-Ray, is a huge deal. But even 1 GB/sec sequential and 20k IOPS is really a lot of performance. If that's not enough for a game, the problem is with the developers, not the storage system.
In contrast, the advantage to Microsoft's approach is huge: if you want to add capacity, you can get it more cheaply. And I mean a lot more cheaply, not just a little. If you want to add another 1 TB of capacity with specs analogous to the next Xbox, at current New Egg prices, you can get a Crucial P1 or a Western Digital Blue SN550 for $115. For a PCI-E 4.0 drive of the same capacity, prices start at $200. That right there eats up the PS5's cost advantage from a smaller die and then some.
So on net, Microsoft made better decisions than Sony on the SSD, while Sony made better decisions on the memory bus. For the GPU, Microsoft will give you a better GPU, but it also costs them more, and that's going to get passed on to you. If you want to add more storage than the default ~1 TB that the consoles will come with, however, that's going to make the next Xbox cheaper overall. I'm going to call that a net win for Microsoft on the hardware side of things, but it's not a complete slaughter like the PS4 over the Xbox One or the Xbox 360 over the PS3.
Like the next Xbox, the PS5 will have eight Zen 2 cores. It will clock them a little lower, at 3.5 GHz rather than 3.8 GHz for the Xbox. But CPUs are so heavy on turbo that the nominal clock speed is fairly dubious. Given that both will have the same CPU cores, if there's a performance difference, it will be due to the amount of power that the CPU is allowed to burn.
On the memory side of things, both the PS5 and the next Xbox have 16 GB of GDDR6. The two consoles handle it differently, however. Sony does the simple, obvious thing: four identical 64-bit channels with 4 GB each, for 448 GB/sec of bandwidth. You get the same bandwidth regardless of how you use it.
As best as I can tell, the Xbox has a 320-bit memory bus, which breaks down unevenly as three 64-bit channels with 4 GB each and two with 2 GB each. That means that if you stay inside of 10 GB, you can get up to 560 GB/sec of memory bandwidth. Go over that, and you only use three of the channels, for 336 GB/sec of bandwidth. That's a lot like the awful shenanigans that Nvidia used to play before the class action lawsuit over the GTX 970 got them to stop. AMD hasn't done that in the past, and I'd bet that Microsoft pushed for it against AMD's advice.
There are a lot of circumstances where, if you really, carefully optimize things specifically for the new Xbox, it will win on memory bandwidth. In spite of that, I'm going to say that the PS5 has the better memory subsystem, due to its simplicity. Programmers hate having to manually put things into different memory pools if there's no good reason for it, especially when the two memory pools will trip over each other because of a lot of shared bandwidth. Apparently Microsoft didn't learn its lesson with the ESRAM debacle on the Xbox One.
As for the GPU itself, both use AMD's upcoming RDNA2 architecture, now branded as Navi 2X. The next Xbox has 52 compute units at 1.825 GHz, for about 12.1 TFLOPS. The PS5 has 36 compute units at 2.23 GHz, for about 10.3 TFLOPS. So on raw performance, the next Xbox wins.
If you read my link above, Sony has some spin as to why their approach is better. Their spin is garbage and they know it. It basically translates to "ours isn't as good, but we don't want you to realize that". More compute units clocked lower wins, because you can get the same performance in less power, in part because the lower clock speed allows lower voltages. By relying on such a high clock speed, there's also a significant risk that TSMC just can't manufacture it to be able to run at the intended clock speed with acceptable yields, and that TFLOPS gap widens further when they have to clock the chip lower. Sony's approach does have the advantage of a smaller die being cheaper to build, which is why they did it that way.
The two vendors are both offering a custom PCI-E over NVMe SSD, but beyond that, they're taking substantially different approaches. Sony is going to fit the PS5 with an SSD that offers high end performance. It would probably be the fastest M.2 drive on the market if it were available today. And they're telling developers that you can assume that the PS5 will have this much performance and it's fine to make a game that doesn't work properly if paired with a slower SSD. Sony will allow you to add your own SSD, but if it's slower than the one you're using, some games might not work at all.
Microsoft, on the other hand, is using a much slower SSD, and possibly only PCI-E 3.0. Meanwhile, Microsoft is saying that you can add your own SSD for more capacity and it will just work. That is, Microsoft isn't allowing game developers to assume high end SSD performance. They've surely got some sort of minimum specs, but they might be set such that every single PCI-E over NVMe SSD ever made is good enough.
I'm going to call that a win for Microsoft. There just isn't much to gain from games assuming that the SSD can transfer data at 5 GB/sec rather than "only" 2 GB/sec. Having real SSD speed rather than the speed of a hard drive, or worse, an optical drive like Blu-Ray, is a huge deal. But even 1 GB/sec sequential and 20k IOPS is really a lot of performance. If that's not enough for a game, the problem is with the developers, not the storage system.
In contrast, the advantage to Microsoft's approach is huge: if you want to add capacity, you can get it more cheaply. And I mean a lot more cheaply, not just a little. If you want to add another 1 TB of capacity with specs analogous to the next Xbox, at current New Egg prices, you can get a Crucial P1 or a Western Digital Blue SN550 for $115. For a PCI-E 4.0 drive of the same capacity, prices start at $200. That right there eats up the PS5's cost advantage from a smaller die and then some.
So on net, Microsoft made better decisions than Sony on the SSD, while Sony made better decisions on the memory bus. For the GPU, Microsoft will give you a better GPU, but it also costs them more, and that's going to get passed on to you. If you want to add more storage than the default ~1 TB that the consoles will come with, however, that's going to make the next Xbox cheaper overall. I'm going to call that a net win for Microsoft on the hardware side of things, but it's not a complete slaughter like the PS4 over the Xbox One or the Xbox 360 over the PS3.
Comments
Sony is using NVMe standard to my knowledge.
I call that a win for Sony if I’m understanding it correctly
Im still worried about TDP/noise/size on these consoles
My GTX 970 would have fantastic performance until it tried to use that last bit of memory on a slower bus and then it would shit the bed immediately. If that's the expected results of the new XBox, then they're in for some lawsuits, stock drops and sales losses.
cool. Until you have to manage them. I suppose it will be up to the APIs and how much micromanaging Devs have to do to make it work like it should.
I tend to thing the simplistic approach was good enough there, but maybe it will make a difference, especially if you really are trying to push 4K 60Hz without a lot of other shenanigans.
my opinion is it’s more about the content than the hardware (otherwise Nintendo wouldn’t even exist any longer), and making Devs happy means they make content for your box. Sony kinda learned the lesson about complicated architectures with the PS3; while MS is still trying to woo them over with more power. It didn’t really move the needle with the XB1X and my opinion is it won’t here either, but time will tell.
The problem with mismatched channels is that you just can't do that. If you don't use more than 10 GB, then you can only use the first 2 GB of each of the channels with 4 GB of physical memory. That way, you can keep everything evenly split among the channels and everything is good.
The trouble is that the last 6 GB uses three of the same channels as the first 10 GB, while not using the other two. If you try to access that 6 GB, you're using some of the memory channels that you'd need for the first 10 GB, so you can't access parts of that first 10 GB at the same time. If you know that 6 GB of your data will be accessed relatively rarely, you could stick it in the last 6 GB and it wouldn't be that big of a deal. But having to count up how many bytes will be used how often in a typical situation is quite a nuisance.
If it were only having to count things up like that on a GPU, it wouldn't really be that bad. GPUs typically don't allocate and free memory anywhere nearly as often as CPUs. But this is a shared memory pool that will be used for the CPU, too. Now you've forced every single programmer who does anything at all to care about the split, likely including some who aren't used to optimizing code at all, not just the few who handle the GPU.
If Microsoft wanted a 320-bit memory bus, then the reasonable amounts of total memory capacity are 10 GB, 20 GB, 40 GB, or possibly 5 GB (as I'm not sure of the capacity of the smallest GDDR6 chips). Even if we dismiss 40 GB or 5 GB as far away from what they wanted, the reasonable choices are 10 GB or 20 GB. Microsoft decided 10 GB wasn't enough, but they weren't willing to pay for 20 GB, so they instead made a mess.
Perhaps the simplest way to handle it is to pretend that the whole system only has 10 GB of memory and the other 6 GB is only attached via duct tape or something. That is, treat it like a GeForce GTX 550 Ti, GeForce GTX 660, or whatever.
where have I heard that before....
Error: No Keyboard Detected!
Press F1 to continue......
I mentioned the GTX 970 because it led to a class action lawsuit, but it did something different. It had 4 GB of memory, with properly matched channels so that each 64-bit memory channel had 1 GB. The problem is that it disabled some of the ROPs and possibly other stuff that was tied to 0.5 GB of memory so that you couldn't get the full use of it.
This isn't that good of a comparison, but imagine a quad core CPU where one of the cores simply has no access to the L3 cache.
So it's upgrade or play old games forever.
거북이는 목을 내밀 때 안 움직입니다
거북이는 목을 내밀 때 안 움직입니다
With CPU caches, there is no magic fix to have a larger data set cached in a smaller cache without ever having the wrong data cached there. But with striping data across multiple memory channels, there is a basically perfect solution: stripe everything across all of the channels evenly. That's why nearly every CPU and GPU ever made has used that solution, at least up to OEMs misconfiguring systems or servers with workloads that are known to need enormous memory capacity but not a ton of bandwidth. For that matter, RAID arrays do about the same thing and for the same reasons.
Now, you could argue that it's likely to not be that bad in the real world, as even 336 GB/sec theoretical in the worst case is still a lot of bandwidth. This isn't going to make the next Xbox slower than a console that has much less bandwidth than that even in the best case. But it's still a dumb mistake. You're spending all that die space and motherboard traces to have a 320-bit GDDR6 memory bus without being able to use the whole thing all of the time.
거북이는 목을 내밀 때 안 움직입니다
the XB1SX is roughly equal to a 2080Ti (which costs $1k+), and the PS5 is roughly a 2080 (which costs $700) —- and those prices are just the GPU, you still need the rest of the system.
consoles are expected to be at or under $600, but no pricing has been defined yet
if that puts it in any perspective,
More competitive game pricing, since you don’t have to pay license fees to the PC maker
Much more flexibility in how much you choose to pay for your rig
You can get/build a much faster PC than exists for consoles
But the console has counters to those:
A console, historically, has always been less expensive than a similarly performing PC
No need to fool with drivers or DIY upgrades
The software is vetted much more strictly and typically runs well
KB/M vs Controller .... I won’t really call that a pro or con either way as it’s a matter of personal preference.
https://devblogs.microsoft.com/directx/announcing-directx-12-ultimate/
Microsoft finally unified the API and, indirectly, hardware standards between PC and Console.
It's not exactly write once, run anywhere, but it's pretty darn close. It means that a Dev can publish for both XSX and Windows 10 with very little friction.
The XB1 wasn't far off, as it also adapted DirectX as an API, but now MS is pushing harder on hardware unification.
This, combined with XBGP bridging across the console and PC, is a pretty big deal and something Sony has absolutely no way to counter.
That being said, it can cut both ways. Microsoft obviously wants you to be ubiquitously in their ecosystem, if it's on a console or PC running their OS. And this API unification along with XBGP (and to a degree, cross-platform play), could very well lead to that.
It could also lead to people saying "Why do I need an XBox when I can just run it on my PC". And that would lead to Sony running away with the Console race.
Maybe that's what Microsoft is going for, and the console isn't really what MS is using to drive their business model; it's just a means to an end. Lose the console battle, but win the gaming war. If MS can collect licensing on games published for PC just as easily, they already collect analytics and telemetry from the OS which pays, then the cost of R&D and support of a console hardware platform is just dead weight, as you have dozens of PC OEMs and hardware manufacturers out there doing that work for free right now.
I can't tell if MS here is about to catch lightning in a bottle, or shoot themselves in the foot.
Godfred's Tomb Trailer: https://youtu.be/-nsXGddj_4w
Original Skyrim: https://www.nexusmods.com/skyrim/mods/109547
Serph toze kindly has started a walk-through. https://youtu.be/UIelCK-lldo
Right now, the Xbox One X is the fastest game console ever made. Let's suppose that the PS5 and the next Xbox both launch late this year. Now many more games for the Xbox One X do you think will launch in 2023 or later?
Sure, you may get some here or there, but they have recently published games for the 3DS, Vita and Wii. So it's not entirely unheard of in either circle.