Digital Foundry Article Technical Discussion [2024]

I love how PCs are archaic designs... but every generation consoles become more like PCs. What the hell is a console these days if not a PC with a unified memory architecture... which btw.. already exist outside of consoles.

I think this is always bound to happen. In the past,consoles are different simply because PC back in the day can't do what they want to do efficiently. If there are some new idea which don't exist on PC then maybe we'll see them in a new console, but unfortunately it becomes quite rare these days. Cost is another issue. It's now much more expensive to make something not from existing commodity hardware.
So in a way one can see PC as a testing ground for technologies to be used in consoles. If something is good and cheap you'll likely see them in new consoles. So I guess if things go this way we probably will see consoles with shorter life cycle.
 
should Microsoft go with PowerPC once again then?

Why Microsoft switched from Intel to Power PC for the Xbox 360 | MVG


We know how that ended up.

The issue is two fold:

1) Current x86 architecture is bullshit because it has so much legacy crap hampering the designs for supposed compatibility, whereas M1 showed 4 years ago that it is possible to emulate x86 fast enough so that it can be complete ditched. Every year trillions of transistors are wasted on garbage x86 silicon, those transistors will never even receive any current as close to nobody (0,0001%) is going to run 16 bit BS code for example

2) imagine two factories having to work together to create something. Both are built bigger and bigger and bigger, but, they are separated by an ocean. So the parts have to travel by boat. Over the years the factories were made even bigger and they use more and more ships, but the sea remains a bottleneck. So instead they try and create speedboats to transfer the parts between the factories which is a giant waste of energy and there will always be situations where one factory has to wait for parts of the other one.
So they build big storage facilities to store the parts so that the ocean is less of a bottleneck and they think they are really smart. It is the biggest factory combination of the world and no other single factory has the same production output!

However, factories that were wise enough to build them closer together, do not need to wait for any boats nor need they create and maintain large storage shelters, they can produce in 1 line.
So while the smaller factories might have only half the output of the big ocean separated factory, they only use 1/10th of the energy.

Of one were to scale up the single factory to the energy usage of the ocean separated factory, it would have 500% the output.
 
I think for many years what consoles should have only custom hardware as that was before 8th gen. Yes that hardware is harder to program, but results will be better.
Where would a platform holder go for this "custom hardware"? Is there someone in the world that is making better chips than AMD, Nvidia or Intel?
Or would they do this in house?

It's not possible today, the complexity and costs would be way too high. At most they can make some features to implement in other architectures, like what Sony did with the checkboard hardware in PS4 pro (and the ML architecture in the PS5 pro?)
 
The issue is two fold:

1) Current x86 architecture is bullshit because it has so much legacy crap hampering the designs for supposed compatibility, whereas M1 showed 4 years ago that it is possible to emulate x86 fast enough so that it can be complete ditched. Every year trillions of transistors are wasted on garbage x86 silicon, those transistors will never even receive any current as close to nobody (0,0001%) is going to run 16 bit BS code for example
"Back when AMD first announced its intentions to extend the x86 ISA to 64-bits I asked Fred Weber, AMD's old CTO, whether it really made sense to extend x86 or if Intel made the right move with Itanium and its brand new ISA. His response made sense at the time, but I didn't quite understand the magnitude of what he was saying.

Fred said that the overhead of maintaining x86 compatibility was negligible, at the time around 10% of the die was the x86 decoder and that percentage would only shrink over time. We're now at around 8x the transistor count of the K8 processor that Fred was talking about back then and the cost of maintaining x86 backwards compatibility has shrunk to a very small number."


For some reason I didn't see any benchmarks in your post?
 
Where would a platform holder go for this "custom hardware"? Is there someone in the world that is making better chips than AMD, Nvidia or Intel?
Or would they do this in house?

It's not possible today, the complexity and costs would be way too high. At most they can make some features to implement in other architectures, like what Sony did with the checkboard hardware in PS4 pro (and the ML architecture in the PS5 pro?)
I know that is not possible today. Just said that was better. And yes AMD, Nvidia or Intel can make custom hardware if there will be customer who buys. ATI made some custo chips for Gamecube and XBox 360, AMD made custom chip for Wii U. That was interestng time. And now that is as XboxKING said, cheap cost effective parts in efficient designs, nothing more.
 
The issue is two fold:
One, that you haven't presented any actual data to support your rhetoric, and two, that you haven't presented any actual data to support your rhetoric. Now I know technically that's only one issue, but given it's such a doozy I thought it worth mentioning twice.

This is a technical discussion. Please provide the evidence in support of the argument. Just repeating the same argument without moving it forwards contributes nothing to the discussion and just grinds the thread down into a loop.
 
One, that you haven't presented any actual data to support your rhetoric, and two, that you haven't presented any actual data to support your rhetoric. Now I know technically that's only one issue, but given it's such a doozy I thought it worth mentioning twice.

This is a technical discussion. Please provide the evidence in support of the argument. Just repeating the same argument without moving it forwards contributes nothing to the discussion and just grinds the thread down into a loop.

here are some CPU benchmarks
 
I know that is not possible today. Just said that was better. And yes AMD, Nvidia or Intel can make custom hardware if there will be customer who buys. ATI made some custo chips for Gamecube and XBox 360, AMD made custom chip for Wii U. That was interestng time. And now that is as XboxKING said, cheap cost effective parts in efficient designs, nothing more.
Depends what you mean by "custom" chips. Because consoles chips are already custom, as in the platform owners can choose to take or leave some features and add in others.

They are already pretty efficient for 2020 consoles, it's just that Nvidia was very smart at implementing Ray tracing hardware and ai upscaling when they did, at that point the console chips were almost finalized and there couldn't be big changes. The best thing sony and Microsoft can do is implementing those features, the rest is probably unnecessary.

Arm cores are a possibility, but other than that, there isn't much that they can do.
 

here are some CPU benchmarks

Please timestamp the specific benchmarks you're referring to which support your argument that ARM based (or some other CPU architecture) can be faster in gaming than the best x86 based CPU's. Because I'm not seeing anything to support such a claim in there.
 
Please timestamp the specific benchmarks you're referring to which support your argument that ARM based (or some other CPU architecture) can be faster in gaming than the best x86 based CPU's. Because I'm not seeing anything to support such a claim in there.
Unfortunately, it gets even worse:


from 17minutes on the M3 max video
 
Last edited:
I notice there are no AMD CPUs in that benchmark, only the Intel 13980HX, which is already 2 process nodes behind and beaten by Intel's own Meteor Lake, released last year.

They don’t stand a chance either, unless they are allowed to use 300-500% the energy then they might get double digit percentage advantages:

X86 main issue however is, M1, M2 and M3 were basically the same chip but with slightly modified nodes,
M4 and thus M4 pro and max (don’t even dare think about M4 ultra) is the actual ‘new’ chip
 
"Back when AMD first announced its intentions to extend the x86 ISA to 64-bits I asked Fred Weber, AMD's old CTO, whether it really made sense to extend x86 or if Intel made the right move with Itanium and its brand new ISA. His response made sense at the time, but I didn't quite understand the magnitude of what he was saying.

Fred said that the overhead of maintaining x86 compatibility was negligible, at the time around 10% of the die was the x86 decoder and that percentage would only shrink over time. We're now at around 8x the transistor count of the K8 processor that Fred was talking about back then and the cost of maintaining x86 backwards compatibility has shrunk to a very small number."


For some reason I didn't see any benchmarks in your post?
The quote you have is almost a quarter century old btw
 
That quote reminded me of something i've always wondered, If you remade the pentium 166mmx using todays best process how big would it be
 
Back
Top