Topic: General Xbox Series X|S Thread

@BAMozzy This chart doesn’t take into account the AMD power shift bottleneck that will affect during CPU intensive workloads in low latency 60/120fps scenarios.

@AJDarkstar AGAIN you are factoring the system as a whole and NOT just the GPU.

The GPU of the PS3 maybe inferior, but the CPU (the Cell) helps offset some of this. The cell CPU in the Playstation 3 helps take some of the load off the GPU by farming it out to the SPU’s. These SPU’s can then handle various bits and bobs that the GPU cannot (due to not having the power to do so). Understanding how to maximize the Cell’s SPU is key to getting the best out of the console – although typically needs to be an exclusive title to really benefit. Certain titles such as Alan Wake for the XBox 360 did use the CPU for certain effects too however – but overall the console doesn’t rely on it to the degree that the PS3 does.

The GPU in the PS3 was not as powerful as the Xbox 360's but because of the design of the CPU, certain tasks could be offloaded to it and take up some of that 'slack' so to speak but in terms of 'RAW POWER' the Xbox 360 GPU was more powerful than the PS3.

A Console isn't 'just' the GPU but a whole system including CPU, RAM, Bandwidth etc.

@Senua That chart was just to illustrate the 'relative' Flops of consoles over the generations ONLY. That is ONLY looking at the GPU and its theoretical maximum Floating point operations per second and doesn't factor in efficiency gains either or latency problems that allows newer GPU's to get closer to their theoretical maximum.

FLOPs are calculated based on the number of shaders and the number of FP32 operations each can do per second and as the frequency is number of cycles it runs at per second, the 'speed' has a big impact on this.

However, NO GPU runs at 100% efficiency - in other words, there is cycles where a floating point operation is NOT being done - either due to latency (waiting for instruction) or efficiency (waiting on other operations to complete) so a 12TF Series X GPU cannot technically do 12 trillion Floating Point Operations per second and neither can the PS5 do 10.2 trillion Floating Point operations - its the theoretical MAX.

Improvements in architecture, scheduling, bandwidth etc all contribute to a GPU performing more Floating Point operations - which is why an older 4TF GPU won't perform as well as a newer 4TF GPU. Both can have the exact same shaders, the exact same frequency so will give the exact same theoretical TFlop rating but because there is less latency and better overall efficiency, they can actually do more Floating Point operations per second - less idle time, less waiting for instructions or data from other parts of the system.

Its why you can't directly compare the 'Flops' between AMD and nVidia and why you can't compare the Flops between older and new GPUs because the architecture and efficiency is 'different'. This is a theoretical 'maximum' but no GPU is 100% efficient. You can say the Series X is 2x as powerful (based on Theoretical maximums) but the reality is you would need more than 2x 6TF GCN GPU's to deliver the same performance. You may well need an 18TF GCN GPU to match the performance of a 12TF RDNA2.0 GPU and the next gen may well do the same with just 10TF because its more efficient, gets closer to being 100% efficient. Old GPU's may well be 50% efficient (in other words, a 10TF theoretical maximum can only do 5trillion floating point operations - so 5TF in real world performance and to get 10TF real world, would need a 20TF GPU) and a newer one may be 66.66% efficient (so 10TF delivers 7TF in real world performance and would need a 15TF GPU to get 10TF of 'real' world performance).

This is based on FP32 instructions too - if they could use FP16 entirely, the Series X has a theoretical 24.2TF maximum GPU.

That chart doesn't need to factor in AMD's Smart Shift - that doesn't affect the the theoretical maximum whatsoever. A GPU running static at 12TF doesn't mean you are getting 12TF ALL the time It can be sat idle doing NOTHING for half a frame, not processing any floating point operations at all. You often see with Digital Foundry analysis where the GPU is only running at 75% meaning its spending a lot of cycles doing nothing. Smart Shift is designed to use that 'idle' time to bump up CPU speeds and then when the CPU has sent instructions to the GPU to draw the frame, send the power to the GPU and can switch the power, speed up either CPU or GPU multiple times per frame - essentially making it more efficient with its power consumption. There will be frames that may only be using 70% of the capacity of the GPU because its running at full speed all the time and then for a few frames be at 99% because of lots of action and effects happening all at once. On PS5, that GPU could be running at 99-100% all the time, drop the speed so its 99% capacity to save power and reduce temperatures - boosting up when needed - even several times during a single frame. It happens in less than 2ms and a frame at 60fps is 16.6ms so it could boost the CPU to do all the physics, AI and draw calls to the GPU and then boost the GPU to render that image. you don't start rendering the image at the 'start' of the next frame as the CPU has to process any inputs from the controller, any AI, Physics, impact/collisions etc and then tell the GPU what to draw. All of that is what makes up a frame time. That's why dropping resolution below a certain point has NO effect on frame rate because the CPU tasks are still taking time and cannot reduce the frame time any further...

If you need 'hype' - MS should have used this LOL

Sure, it will take a while but the Series X/S game library will definitely be mind-blowing. Sequels to beloved franchised, lots of promising new games, 23 first-party studios and the best console version of third-party games... It's incredible.

Xbox fans again beating the marketing team at it’s own game.

👀

@Senua Those are the profiles from the early PS5 dev kits produced before Sony got smart-shift working. On final PS5 hardware and current dev kits shifts happen hundreds of times the second depending on demand rather than having a whole game with a fixed profile.

@Ryall Considering the tweet is dated very recently, I don’t know whom to believe until they atleast send the hardware for a final transparent Digitalfoundry breakdown analysis. But anyways this proves Xbox hardware team’s statement on how they have initially tested with AMD power shift but have later decided to go for a locked balanced consistent predictable performance profile instead of peak boosted numbers for ease of dev optimisation.

@Senua Digital foundry discussed the profile is used in early dev kit on April 2. The relevant part of the video is between 4:30 to 6:00 minutes. The tweet is recent but it’s quoting something the person was told on discord earlier. That said I doubt every developer replaces all their dev kit every time so Sony brings out a new one.

https://m.youtube.com/watch?v=KfM_nTTxftE&t=929s

It’s really sad to see some people are still living in a state of denial. Almost seems like they are living in a parallel universe.

Time will tell but it's obvious that for the sake of stable performance, developers have to leave some room on PS5 so the frame rate doesn't tank awfully. That figures give us an idea of how PS5 compares to Series X and it doesn't look good for Sony's new console.

SERIES X
CPU clock speed 3.8 GHz (3.66 with SMT)

PS5
up to 3.5GHz (??? with SMT)

SERIES X
GPU 52 CUs at 1.825GHz (sustained)

PS5
36 CUs at 2.23GHz (advertised)
36 CUs at 1.2GHz when CPU is full-powered

For PS5 GPU to run at 2.23GHz as advertised and deliver the promised 10 TFLOPS, the CPU has to run at 2.0GHz. Series X CPU runs at 3.8GHz and that's above the maximum that PS5 CPU could ever achieve.

I will update this post accordingly:
https://www.purexbox.com/forums/xbox-series-x/series_x_specif...

By now it should be no brainer that Series X will be more certain to deliver a sustained locked performance and that devs will be having an easier time to optimise for a more predictable profile.

The thing is, optimisation isn't necessarily going to be affected by Smart Shift. A developer will know the 'stress' points in their games - whether its a stress point to the CPU (AI, Physics, a lot of Objects to call in etc etc), whether its a stress point to the GPU or both and can literally run that 'stress' point over and over again, making adjustments to see how the performance metrics change. Its not like they have to pick a PS5 profile and stick with it, optimise around that and maybe keep going back and forth with different profiles to see what one suits them best - select a 'high GPU/Low CPU' profile, a Balanced Profile or a 'Low GPU/High CPU Profile.

What Smart Shift does is shift power back and forth to boost either as required and can do this multiple times during a frame. For example, it could boost the CPU whilst its calculating Physics and preparing to tell the GPU what it needs to render and then boost the GPU when its starts to render the frame so essentially, you get 3.5GHz and 2.23GHz during that frame time.

Because its fluctuating based on load, all a developer needs to do is optimise based around the 'highest' stress points for the system as a whole - pretty much how they would if optimising on Series X. If they aren't getting the frame rate, they can tell whether its a CPU or GPU intensive and optimise accordingly to reduce the frame time - save a few 'ms' here or there to hit the 16.6ms (for 60fps). Its generally easier to optimise the Graphics side - reduce a setting here or there, drop the resolution as changes to AI, Physics etc affect the game-play. There are some things they can do to reduce the CPU demand, like reduce the amount of Objects (usually more distant ones) as that reduces the amount of draw calls the CPU has to make. Dirt 5 devs will reduce the crowd density for example for higher frame rates as they have far less time for the CPU & GPU to process and render that frame.

From what I heard, the early dev kits were having to use Profiles because the Smart-Shift functionality wasn't working but is now and will be on the consumer model. Whether you have a static profile or not, the difference is negligible. Games don't require a 'static' GPU or CPU as the frames are very different - you barely need any GPU to look at the ground or sky compared to a scene with many objects and a lot of action. The CPU too can vary depending on what's going on so its only going to 'hit' the peak utilisation of either on very rare occasion - unless you are pushing for a frame rate at higher visual settings than the GPU can cope with. A well optimised game that runs at a locked 60fps for example regardless of the complexity of the scene will have 'overheads' that maybe small at the most intensive parts but large if you look at the sky for example. On a Static Profile, you may see that the GPU varies from say 60% to 95% utilisation and on a variable profile, it may stay around 95% utilisation BUT the frequency varies from say 1800-2.23GHz - saving power consumption and reduce both heat and noise generation. Same can be said where the CPU is concerned too.

Its different approaches to the design. Its still not going to make a 36CU GPU out perform a 52CU GPU but may make it a more power efficient system. We will have to see whether or not the long term choices of these designs were the right choice, when Devs really start pushing Physics, AI etc - really pushing CPU heavy game designs after years of having to make do with streamlining, how that impacts the PS5 more so - if indeed it does. They could still boost the CPU to power through that at the start of the 'frame time' and then transfer power over to the GPU to boost that to finish the render in time. We don't know whether the console can run at Max and if it had to, for how long. Whether power from other areas, like the Audio component factors in as well to the equation - so if the Audio isn't being pushed hard, can the lower power draw of that affect the CPU/GPU speeds? Power seems to be the limiting factor (so to speak) so if the power to other areas of the APU is not being used, that power could affect the amount of power available to both CPU and GPU to enable them to run faster.

I get the impression devs are not having to choose between a 'high' GPU performance or High CPU - that the console will adapt automatically to the needs of the game 'up to' those limits and as frames differ in the amount of GPU/CPU resources needed, its not a 'constant' anyway.

What I have found more interesting though was that DF did a test with a 5700 36CU's @2100Mhz and 5700 XT 40CU's @1890Mhz and found that the XT performed slightly 'better' overall despite identical TF's. I know its not as different as 52 vs 36 and the speed isn't as different either - but also in their tests, they found that the power draw ramps up more to 'overclock' and the 'benefits' reduce. It takes more and more power to each time to raise a GPU by 100Mhz and the %age difference decreases too. In other words, going from 1.9Ghz to 2.0Ghz may see a 10% boost in frame rates, but only a 6% boost going from 2.0Ghz to 2.1Ghz. Going from 1.9 to 2.0 doesn't require as much 'extra' power as going from 2.0 to 2.1Ghz - at least in the PC space (although I have no reason to believe that the Console space is different). I do wonder if the aspects directly related to speed of GPU will be affected by the variability too. More Speed can mean an extra bounce or two of a Ray traced Ray - more cycles per second, more bounces - more available Shaders, more Rays. And lets not forget 1825Mhz is NOT slow by any stretch of the imagination - that's more than most GPU's on the market run at stock...

Anyway, not long now and we will see what Sony and MS's design choices translate to in games - how they compare. I am sure there will be a difference but I want to see more than just the first few months of releases - when devs are fully up to speed on BOTH systems as they may have more ease optimising on one so the differences may not be a 'true' representation of the difference in hardware. Out of the gate, you would expect the Series X to have the advantage because of the CPU/GPU but whether they are taking full advantage of it (or the PS5 for that matter) or not, we will have to see over the longer term

@BAMozzy Can Smart Shift absolutely confirm the CPU/GPU both will be at maximum clock speed for a sustained time period in low latency high fidelity requirement scenarios?

That scenario will be impossible to hold up, otherwise Mark Cerny would have been straightforward instead of cryptic and elusive about the benefits of the shared power pool for the CPU and GPU of PS5 in addition to controlling the heat. He wasn't clear about the TFLOPs either (I watched his boring explanation).

@BlueOcean Yes definitely or else why would you need Smart Shift in the first place when Series X could manage to be more (~130%) silent and cooler than even One X with a locked profile. Let’s see how it turns out, variable clock speeds have never been recommendable in PC hardware space.