Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]

Status
Not open for further replies.
Well, allegedly X1 has some hw acceleration for some DX12 api calls and MS bought Havok. If there is some benefit for some simple data structures transformations or specific VLIW operations it's not unlikely.
 
"CPU Built-in with DirectX 12.X, DXR, DirectML and Havok instructions into the chip."

The credibility is dwindling.... o_O
Its already like that for Xbox One X. This is a terrible wording though.
They integrated DX12 into the chip so that the CPU doesn't have to unpack the API calls for the GPU IIRC.

Since DXR and DirectML are part of 12 now this is a straight forward update to a process they are continuing forward. Worded poorly since it tries to separate DXR and DirectML from DX12, as both are part of the API and listed under the latest shader model for support.

Not sure about Havok though.
 
"they integrated dx12 into the chip"

I don't understand what that means, what additional instructions could possibly be so application-specific on a cpu?
 
"they integrated dx12 into the chip"

I don't understand what that means, what additional instructions could possibly be so application-specific on a cpu?
Not additional instructions; less work rather.

Normally the CPU is responsible to take and API call and convert it to calls for the GPU to operate on.

Thus more calls and the higher level the API abstraction and you begin to become CPU bound.

If you move the decoding to The GPU side of things the CPU has less work to do.

edit:
To help developers get the most out of Scorpio, Microsoft has moved the GPU command processor—the hardware that takes instructions from the CPU and funnels them through to the graphics core—over to a DirectX 12-optimised solution. In theory, this will massively improve draw call performance on Scorpio, and cut down on CPU overheads.

"We essentially moved Direct3D 12," Microsoft told Digital Foundry. "We built that into the command processor of the GPU and what that means is that, for all the high frequency API invocations that the games do, they'll all natively implemented in the logic of the command processor—and what this means is that our communication from the game to the GPU is super-efficient."

So what does this all mean? What kind of performance boosts can we expect in DX12 games like Gears of War 4, Halo 5: Guardians, Battlefield 1, and Forza Horizon 3 while running on Project Scorpio?

Microsoft told Digital Foundry's Richard Leadbetter that Project Scorpio's DirectX 12 hardware implementation should cut CPU workloads by half on DirectX 12 games.

"Roll call instructions on the CPU that would typically require thousands of instructions are now reduced to just 11," Leadbetter said in DF's exclusive Scorpio reveal.

"State changes that are also heavy on CPU are now cut down to just 9. Microsoft tells me that this custom hardware automation should cut CPU workload by half on titles built on DirectX 12 renderers."

https://arstechnica.com/gaming/2017/04/Xbox-scorpio-hardware-specs/
 
Last edited:
Not additional instructions; less work rather.
The wording of your post makes it sound like the CPU was enhanced. Instead, the GPU had hardware DX12 enhancements, moving the work from the CPU to the GPU. So how does that tie in with a new SOC having a CPU that has DX12 built in? Surely MS will again place the DX12 hardware in the GPU command processor. Similarly, what Havok enhancements fit on the CPU when the last demos were showing compute based Havok?


What instructions on the CPU aid Havok physics and DirectML? It seems a very odd choice. If we want to give the leak the benefit of the doubt, perhaps we're just seeing a normal, improved vector unit with instructions that can be used for X,Y,Z tasks like ML and Physics, and the leaker has picked up on some key words from some presentation/documentation? Instructions for the purpose of DML, DXR and Havok doesn't sound at all plausible.
 
"they integrated dx12 into the chip"

I don't understand what that means, what additional instructions could possibly be so application-specific on a cpu?

Although the rumor is BS, however, certain dx12 instructions sets being integrated into the CPU isn't all that new - maybe for AMD CPU's though. I think starting with Intel's Skylake-S processors have certain dx12 instructions sets on improving physics and other GPU related functions. IIRC, AMD CPUs during that time couldn't compete (besides clocks/cores) with Intel CPUs during physics benchmarks because of the instruction sets. Mind you, it's been awhile and my memory maybe fuzzy on certain details.
 
Last edited:
new SOC having a CPU that has DX12 built in?
I can only assume this is what the leaker implied. I've only been looking at the SOC.
I can only think that the leaker is implying that the API called to DXR/DML/12 and havok are unpacked on the command processor as opposed to the CPU.
Which is plausible, but the way it was worded in the leaks was very poor.
At least that's how I interpreted it.
 
That's an interpretation that better fits what you expect than what's written. Why group those things with the CPU rather than the GPU or the SOC? Rationalising nonsensical pastebins is of course a hobby, but authentic leaks probably won't require us to join the dots in different ways to find a workable interpretation. ;)
 
That's an interpretation that better fits what you expect than what's written. Why group those things with the CPU rather than the GPU or the SOC? Rationalising nonsensical pastebins is of course a hobby, but authentic leaks probably won't require us to join the dots in different ways to find a workable interpretation. ;)
haha fair enough.
 
Note I'm not saying it's not true. Could be true with the leaker not technical enough to present it accurately. But I'm totally with MrFox that credibility is dwindling with each peculiar notion that appears in a rumoured spec. Looking at just the numbers and ideas, what's the plausibility like? Seems pretty reasonable to me, assuming hardware features for DX12 on the SOC as you suggest. 52 CUs (56 CU chip, 4 for redundancy) at 1.6 GHz and 275 W total? What lithography is going to enable that? And then the little things like two HDMI ports.
 
Note I'm not saying it's not true. Could be true with the leaker not technical enough to present it accurately. But I'm totally with MrFox that credibility is dwindling with each peculiar notion that appears in a rumoured spec. Looking at just the numbers and ideas, what's the plausibility like? Seems pretty reasonable to me, assuming hardware features for DX12 on the SOC as you suggest. 52 CUs (56 CU chip, 4 for redundancy) at 1.6 GHz and 275 W total? What lithography is going to enable that? And then the little things like two HDMI ports.
I generally don't trust any leaks; most of them are made with information perusing forums and like you suggest it's their interpretation of what they read likely.
I just felt that this particular point isn't something that would be considered 'shocking' or perhaps noteworthy in a way that would make a rumour less credible as there is precedence for it.

I suspect Scarlett will be an evolution of the work done with Scorpio, so these types of features do speak to seem in alignment with me.
 
Exclusive Xbox one DX12 enhancements to jag based cores made sense since that was an EOL arch where AMD had no intentions of continued iterations with new features in the PC space.

Exclusive Scarlett based enhancements revolving around DX12/ML makes no sense since those enhancements should be beneficial in the PC space.

Excluding the features from Sony doesn’t require excluding the same features in the PC especially since they would aid in the competition against Intel.

The use of adjectives like custom or enhanced seem like a continuation of what we saw from mid gen jargon.
 
I call bulloks just because nobody at this point can forecast a so precise final gpu frequency based on an ES of a not yet released architecture

bullock-comic.png
 
They describe the 24 GB as 18 GB for games, 2 GB for cache, and 4 GB for OS. The 560 GB/s value has to be for the entire RAM pool as otherwise it'd have the bandwidth with the game partition. eg.

24GB GDDR6 ( Samsung 12x2GB ), 18GB GDDR6 (@ 560GB/s) + 2GB GDDR6 Cache + 32GB~64GB SSD vRAM ( Up-to 84GB for Game ) and Dedicated 4GB GDDR6 for OS ( Native 4K60/120FPS Dashboard )
 
Status
Not open for further replies.
Back
Top