This is an interview with rendering engineer at Crytech, Ali Salehi, talking about PS5 and XSX's specs. Very interesting info. It's in Persian. I'm Persian and I'll translate some important parts for you step by step. Stay tuned!
@JayDubcity16 https://vigiato.net/p/85342 

@TigerCA123
@JustAnLED
@xMBGx
@LiquidTitan
@Puertorock77_

"The developers are saying PS5 is the easiest console they have ever coded on to reach its peak performance. Software wise, coding for PS5 is extremely simple and has so many abilties that make the devs so free. In total I can say PS5 is a better console."
Wait for the rest.

Q: What is the importance of Tflops and does higher Tflops mean a more powerful console?
A: Tflops shows that this processor, in the most optimized and ideal condition, will perform in that rate. The number of Tflops is in ideal condition and theoretical.

Many factors must work together and each part feeds another and gives the result of one part to the other. If any of these factors doesn't work fast enough then it causes in lower performance in another part.

A good example of this situation has happened before. With PS3. PS3 had much higher flops than 360 because of its SPU. But in practice because of its complications and memory bottleneck and other problems it never reached its peak of performance on paper.

That's why you can't care much about these numbers.
But if all parts can work efficiently in Xbox Series X alongside its GPU, It can hit that number in practice, Which doesn't seem so possible. Beside all these, there's a software part too.

The thing we saw in PCs, was DX12 and Vulkan. Without change in hardware, with change in the architecture of the software you can use the hardware better.
This is applicable to the consoles too. Sony runs PS5 on its own OS but Microsoft uses a custom version of Windows

For Xbox Series X. The two are so different. Since Sony has made a software particularly for PS5 It'll obviously give more capabilities to devs than Microsoft which basically brings PC DX to its console.

Q: If I'm getting it right, Tflops if the factor of efficiency in different parts of GPU, am I wrong? What do these floating points mean at all? Please explain it like we're 5.

A: The real problem is the first person who made these public so now they need explanation [ :) ].
This technical information shouldn't mean much to casual users.

The GPU may have 20 different parts. CUs are just one part of it. They do the processings. Meanwhile IF all the other parts are in their best condition, not being limited, without memory bottlenecks, and the CUs get as much as the data they need in a second,

Then the CUs are capable of doing 12T floating-point operations in a second. So in an ideal world where we remove all limits it's possible. But in reality, it's just NOT.

A good example about Xbox Series X hardware is its RAM.
Microsoft has made the RAM two parts. The same mistake they made with Xbox One. One part of RAM has high bandwidth and the other is low. And definitely coding for this could be a little challenging.

Because the the total amount of things we want to put in the fast part is so much that it may cause problems. And if we want to support 4k it will be another whole story.
So there will be somethings that will hold the gpu off.

Q: You talked about CUs. PS5 has 36 CUs and Xbox Series has 56 that 4 of them are reserved and 52 are given to devs. What's the difference?
A: The main difference is that CUs frequencies in PS5 is a lot more and work at higher frequencies.

Therefore, despite the huge difference in CU count, They don't have that much of Tflops difference. An Interesting thing that one of IGN's journalists said. He said XSX is like a V8 motor and PS5 is like a V6 one turbo boosted to the max for best efficiency.

Raising the clock speed has some benefits like in memory, rasterizer, and every other part of the gpu that its efficiency depends on clock speed, things that's not related to CU count or Tflops, will work faster too. So the remaining parts of the GPU will work better

Than XSX. This will make the console work mostly on the 10.28 Tflops. But in XSX, since the other parts of the gpu work slower due to the lower clock speed, it actually works a lot at lower Tflops most often and reaches 12 only at ideal situations.

Q: Won't this difference in CU count show itself at the end of next gen when devs have become more familiar with XSX's hardware?
A: No. Because usually Playstations API generally gives you more freedom.

That's why at the end of each gen you see more mind-blowing results from Sony's consoles. For example in 7th gen, in the beginning common games ran so bad on PS3 but in the end Uncharted 3 and TLOU were coming out of it.

I think it'll happen in next gen as well. But in higher resolutions probably PS5 will face problems and XSX will show more pixel counts.

Q: Sony says lower CU count makes it easier to do the tasks. What does this claim mean?
A: Using all CUs at a time is costly. Because when CUs want to execute a code, they need resources inside GPU that are limited.

If the GPU fails to distribute the resources to all CUs, it'll have to leave some CUs behind. For example It may use only 20 CUs instead of 52 [that's just an example]. Because it might not have enough resources for the rest at all.

And Sony with being aware of this, used higher clock speeds instead of CU count to lower the cost as well. The more understandable instance of this topic is about CPUs.
It was a long time that AMD had high core count CPUs.

While even higher core count CPUs from Intel itself weren't working so much better. Quad-core CPUs, or 8 core CPUs but with much more performance per core, were actually doing better in running games.

Obviously a 16 or 32 core CPU has higher Tflops, But less core CPU performs better in games. Because it's difficult for devs to use all of those cores and prefer fewer but faster cores.

Q: Can SMT be a game changer for XSX in the last years of the next gen?
A: Technically hyper threading existed since Pentium 4 in desktops and considered each physical core as two virtual cores and helped performance in most cases.

The SMT in XSX allows devs to decide to choose between hyper threading or not using it and use higher clock speeds. And that's exactly how you say. It's not quite clear how to make a logical decision. So SMT might be used at the end of the next gen.

Q: Can you say what do you mean by "not quite clear"?
A: I mean it needs a proper analysis of the code. So this isn't something to be known by everyone right now. Right now there are far more important concerns to know the hardware of these consoles better.

And probably firstly the devs go without SMT and use it another time.

Q: There are 3328 Shaders in XSX's GPU, What's a Shader, What does it do, and what does 3328 shaders mean?

A: When CUs want to run a code, they do it through some units called "Wavefronts". Multiplying the number of CUs in Wavefronts is the number of shaders. But it really doesn't matter and all I said about CUs are applied here as well. There are still,

Some limits that don't allow all of them to be useable and a high number of them doesn't necessarily mean good.
Another important thing must be considered and Mark Cerny mentioned it too, CU or Tflops isn't necessarily the same on different

Architectures, meaning you can't just compare the Tflops and say which is better. So you can't trust these numbers that easily.

Q: Comparisons between Android and IOS phones have been heated up alongside consoles discussions, and It is mentioned in the discussions that Android phones have less performance with higher RAM. Is it somehow correct about consoles too?

A: The software stack that sits on top of the hardware determines everything. As with updating the driver the performance rises instantly, That's the case here too.
Sony always had the better software because Microsoft has to use Windows. So it's correct.

Q: Microsoft emphasized that the frequencies in XSX will be the same under any circumstances. But Sony doesn't have this approach instead they give a set energy budget to use it variably depending on the situation. What are the differences and which is better for devs?

A: What Sony has done seems more logical because according to the processing workload it decides that in a certain moment the GPU frequency must be higher, or the CPU must be higher. For example, in loading moment only CPU is engaged and GPU is not used,

Or in another scene it's a close-up of the face of a character, here GPU is much more engaged, On the other hand, the fact that XSX's cooling system keeps the frequency stable and prevent from throttling is very good. But the freedom that Sony has given here is a big thing.

Q: doesn't this freedom make it more difficult for devs?
A: actually no. Because we're already doing such things engine side. For example Dynamic Resolution Scaling which some games use, measures different factors and finds out how much the GPU is under pressure now?

And how much should the resolution get lowered so the FPS remains on the goal rate. So connecting these to each other is so easy.

Q: The Geometry Engine that Sony talks about, what is its usage exactly?
A: I don't believe it'll be used so much until two years in next gen. Probably we'll see its impact in the second wave of PS5's games. But in the beginning it won't have big impacts.

Q: XSX's chipset is 7nm. And we know the lower the number goes, the better chipset gets. Will you go deeper into nanometers and transistor counts?
A: nanometer derease means more transistors, and controlling their heat in bigger counts and smaller space.

It's just a better technology of production and the nanometer doesn't matter. What matters is transistor count.

Q: The peak speed of PS5's SSD is 8-9 GB/s. Now that we've achieved such a speed, aside from games loadings and more details, what other things'll happen?
A: The first thing is that happens is loading screens are gone from games.

Second, Microsoft already showed the ability to stop games and resuming them with running multiple games and jumping from one to the other under 5-6 seconds [Quick Resume]. This'll be under 1 second in PS5. Another expected thing to happen is changes in games menus.

When there's no loading anymore, naturally there's no waiting as well and you don't need to watch a video so the game can load in the background.

Q: Meanwhile how will the games be on PC? Since SSD is an option for a PC user.
A: It's always been the consoles that determine what is the standard. So devs will make their games based on consoles and If someone plays on PC and doesn't have SSD, will either have to

wait long behind loadings or reconsider buying an SSD.

Q: As a programmer and developer, which console do you find better working with and coding into yourself? PS5 or XSX?
A: Without a moment of hesitation and definitely PS5.

As a programmer I say PS5 is much better and I don't think you can find a programmer that

Could name one advantage that XSX has over PS5.
They have to use DX and Windows for Xbox, which are years old, But Sony makes the APIs and software from ground up the way they want for each new console they make. Beneficial to both us and them.

"Because there's only one way to do a job, and it's the best way possible."

I'm finally done with this translation. I hope you enjoyed.