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

Status
Not open for further replies.
And at this point (probably up until late Q2 2020 I'd say) the PS5 can also get an improved cooling solution together with improved VRM to be able to push the clocks up.

No one is suggesting Sony was using a 40 CU chip up until late 2019 and now they'll adopt a 60 CU model.
What we have right now is the following info:
- Devs telling journalists and insiders that the PS5 and XBSX are close in performance
- Github leaks: Oberon with 36 active CUs (tested @ 2GHz back in June 2019) and XBSX with 56 active CUs
- Aquarius-something SoC measurements: ~315mm^2 PS5, ~380mm^2 XBSX
- Phil Spencer claiming XBSX is 2x XBoneX.

How can all of these be true?
In the sense that you're asking; they cannot all be true. But the reality is that we're missing data. Context in particular. A bit like how sarcasm is more difficult to read in words than it is to hear it directly from the person. You can only determine truth with information that is usable, and a fair amount of the information isn't useful and even true information can be wrong. Could you imagine an improperly coded test in which the test had bugs that results in poor results? It's not like AMD comes out and said, yup this is a bug free leak boys, go at it. There's so much unknown, I don't place a lot of stock in any of it.

There are some things that we know that are universally true though, and that we have multiple methodologies in producing a number; like measuring the die size of a chip. Knowing how dense and how many transistors can be packed per mm2. These are things we can tangibly work with, use stable ground to test areas with uncertainty. I'm sure you know a great deal many more things than I do, and I think leveraging that background of yours, bucket information that is useful to funnel possibilities.

1 - All the journalists and insiders think that 9.2 TFLOPs is close to 12 TFLOPs (which IMHO is a downright stupid assumption, 30% was never close in hardware performance, it's way more than RX 5700XT vs. RTX2080 Super and no one considers those close. Besides, the "close" or "similar specs" wording probably came from devs themselves and those would definitely not consider a 30% gap as "close").

2 - All the journalists and insiders that claimed close performance were given false information (hardly, given the track record of some of them).

3 - The PS5 is using an off-chip ray-tracing accelerator that levels things between both consoles.

4 - The PS5 SoC is planned to use more than 36 CUs out of a total of 40 present in the silicon (the PS4 Pro also has 40 total CUs), together with very aggressive clocks. I'd be wary of this option, if it wasn't for the 1 year-old "proof" already showing a whopping 2GHz.

5 - XBSX is actually closer to 9.2 TFLOPs than it is to 12 TFLOPs, maybe due to low GPU clocks around 1.4GHz.

6 - ?
The only usable piece of information here is point 4. And that's only because of you latter statement of looking at some shotty 1 year old proof. I mean I could argue that MS is not using an AMD solution either. Error Proofing is a typical developers sigh moment right?
Assume you have a flag in which you use to test Ray Tracing: what should the results say if you choose no? or choose yes?
Does No = Blank? (The reason people believe Sony is using custom hardware)
Yes = you make it run a benchmark for ray tracing

Okay so if "yes"
What if it doesn't support ray tracing?
Do you write out the result of 0 for a division by 0 catch statement?
or or do they just have MRay/Sec but with no value in front of it? (this is the assumed case for Xbox, a metric with no value)
Do you leave the field blank? (People assume this is the case for Sony)
or is the test not run at all in which would beg to ask why bother with the flag anyway.
or there was never a benchmark run for Ray Tracing and it was nothing else other than checking some instructions and whether they are available yet or not.

Everyone is so convinced that the ray tracing test worked on that 1 year old shotty benchmark, but no one would be able to reasonably tell me how XSX scored on it because there are no values. But they seem to be 100% confident about the CUs and clock speeds, and it being a AMD RT based solution. (not saying it isn't, just there is reason to leave doubt)

Far too many ifs. It's just easier to wait for better information and discuss speculation in context with what we do know, instead of what we don't. Fun to speculate though, but not worth coming back to this thread for every hour.

I mean shit.. everyone is calling it regression testing. But do people know what it means? Simple Google here:

Regression Test Selection
  • Instead of re-executing the entire test suite, it is better to select part of the test suite to be run
  • Test cases selected can be categorized as 1) Reusable Test Cases 2) Obsolete Test Cases.
  • Re-usable Test cases can be used in succeeding regression cycles.
  • Obsolete Test Cases can't be used in succeeding cycles.
Selecting test cases for regression testing
It was found from industry data that a good number of the defects reported by customers were due to last minute bug fixes creating side effects and hence selecting the Test Case for regression testing is an art and not that easy. Effective Regression Tests can be done by selecting the following test cases -

  • Test cases which have frequent defects
  • Functionalities which are more visible to the users
  • Test cases which verify core features of the product
  • Test cases of Functionalities which has undergone more and recent changes
  • All Integration Test Cases
  • All Complex Test Cases
  • Boundary value test cases
  • A sample of Successful test cases
  • A sample of Failure test cases
So we know those tests are probably real. But do you know what tests they were trying to do? If the answer is no, then you can't proceed on this being 'everything'.
 
Has an actual credible insider/journnalist given a concrete TF number for PS5? We have had steady leaks of test hardware with codenmanes that are giving exact numbers that show 36CU's @1.8GHz and then 2.0GHz and it's being disputed by using statements from vague insiders and a journalsist that gave some general statmenets about power levels.

Has any of the "insiders" disputed PS5 having 36CU's? Yes/no?
 
Has an actual credible insider/journnalist given a concrete TF number for PS5? We have had steady leaks of test hardware with codenmanes that are giving exact numbers that show 36CU's @1.8GHz and then 2.0GHz and it's being disputed by using statements from vague insiders and a journalsist that gave some general statmenets about power levels.

Has any of the "insiders" disputed PS5 having 36CU's? Yes/no?

Klee disputed the number.
 
Has an actual credible insider/journnalist given a concrete TF number for PS5? We have had steady leaks of test hardware with codenmanes that are giving exact numbers that show 36CU's @1.8GHz and then 2.0GHz and it's being disputed by using statements from vague insiders and a journalsist that gave some general statmenets about power levels.

Has any of the "insiders" disputed PS5 having 36CU's? Yes/no?

I doubt if any game software devs have any concrete info outside of some first party devs. The hardware data dumps seems to be coming from people working in the hardware production/testing/verification field.
Devs seem to giving figures derived from dev kits that lack final hardware. Probably where all the confusion lies.
 
Or a more direct reading is required where "Sony is faster than MS" simply means it has higher MHz, would that make more sense of what's allegedly known?
 
I doubt if any game software devs have any concrete info outside of some first party devs. The hardware data dumps seems to be coming from people working in the hardware production/testing/verification field.
Devs seem to giving figures derived from dev kits that lack final hardware. Probably where all the confusion lies.

Right but this is why I believe the hardware data dumps over the "insiders". Unless the hardware data dumps/leaks were some extremely elaborate orcastrated hoax then there is no reason not to just take it at face value. With the insiders there could be misinterpration/misunderstanding at mutlple levels. This is even if you believe they have legit info to begin with.
 
Or a more direct reading is required where "Sony is faster than MS" simply means it has higher MHz, would that make more sense of what's allegedly known?
Yes it could make more sense.
TBH I think XB1 teached MS the lesson to rather bet big and not fall behind like before, and Sony might have underestimated MS's bet this time around on how far they're willing to push it.
As a consequence, if there will be Pro/X midgen updates, MS has less room to improve.
 
Like I said before Xbone had less bandwidth, 1/2 ROPs, 1/2 shader engines, in addition to the compute deficit.

How often was Xbox One fillrate limited, or limited by the gpu front-end? Serious question. I'm assuming it was usually alu limited or bandwidth limited.
 
Yes it could make more sense.
TBH I think XB1 teached MS the lesson to rather bet big and not fall behind like before, and Sony might have underestimated MS's bet this time around on how far they're willing to push it.
As a consequence, if there will be Pro/X midgen updates, MS has less room to improve.
After Scorpio 6tf GPU it would have being really stupid for them to underestimate MS will to aim high numbers.

It was very probably that they would aim for at least 12tf for next gen. Since 2017 their main selling point is power power power...and they would have released a 8tf machine in 2020 ? Ridiculous. And most people don't care about all that gcn tflops != rdna tflops...thing.
 
I'm still a proponent of the 320-bit bus with mixed capacity chips for XBSX. I think it makes sense in a console where software will be written according to a known memory configuration (as opposed to PC) and that the limited capacity of full bandwidth memory won't be an issue given the 3GB OS reservation and the need to store non GPU-related data structures. It'll be interesting to see if they do actually use this configuration and Sony chooses one of: 320-bit bus with 20GB of memory, 256-bit bus with 16 GB of memory, 256-bit, 16 + 4GB DDR4, something else.

I wanted to try to visually conceptualize my thinking regarding the XBSX vs Lockhart memory model. Noting ahead of time that I might be out of my depth a bit here.

XBSSX

upload_2020-1-10_15-54-21.png

Lockhart

upload_2020-1-10_15-56-16.png
 
After Scorpio 6tf GPU it would have being really stupid for them to underestimate MS will to aim high numbers.

It was very probably that they would aim for at least 12tf for next gen. Since 2017 their main selling point is power power power...and they would have released a 8tf machine in 2020 ? Ridiculous. And most people don't care about all that gcn tflops != rdna tflops...thing.
True enough, but Sony might have thought around 10 TFLOPS being realistic based on their own goals and possibly seeing XBX as overreaction on how big you should push because XB1 fell so badly behind, and be satisfied with what they'd think would be about performance parity.
 
My guess is the "power" of each console is a conclusion of the price point they decided to target rather than any marketing perspectives relative to the competition. You decide, do you want to be a $400, $500 or $600 console and then the performance you get is a consequence of that. They're not going to say $400 is the best price point for our consumers, but let's make a $500 console because we really want to smash the other guy.
 
Status
Not open for further replies.
Back
Top