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

Status
Not open for further replies.
That was also a few days after Phil had tweeted that he took Project Scarlett home with him. So, maybe related, maybe not?

Regards,
SB
Not related. Except he is statistically 100% on next gen reveals when he is attending a major event and tweeting about it :)
 
If doing nothing but MADDs, 9.2 TF of RDNA will be exactly the same as 9.2 TF of GCN, but in real workloads RDNA can do more per trillion operations.
I'm getting flashbacks to the Intel vs PowerPC processor performance arguments.
 
What if retail PS5 has 40/40 CUs enabled, and clocks pushed to 2.15ghz to hit a 11TF target?

To me that seems pretty out there...
38 CU for 10.2 TF ... [emoji6]
2,1 ghz....

Then 4 operative modes:
(A)Ps5: 38x2,1ghz
(B)Ps4pro-boost: 36x1,82ghz
(C)Ps4-pro:36x0,91ghz
(D)Ps4:18x0,8ghz
 
I think ps5 most of time will operate in mode (B) so stressing HW less (and beeing not so loud) specially first couple of years of console life ... I think Sony could not escape BC troubles in other ways than this....

Can you have 9 WGP in one SE and 10 in the other? Unbalanced SEs?
Well don't know
What I know they need -for PR reasons- to be a little above the two digit TF mark....

Well did the math 40CU@2.0 ghz give this "magical" 10.20 TF number... no redundancy may just come from utilizing the defective/underperforming chips for another model... they may build also a kind of "ps4-pro" or from cutted chips 36@0,91 ghs.... so they can dismiss the 16nm process... this second console will just have same performances ad MS secondary console...
 
Last edited by a moderator:
38 CU for 10.2 TF ...
emoji6.png

2,1 ghz....

Then 4 operative modes:
(A)Ps5: 38x2,1ghz
(B)Ps4pro-boost: 36x1,82ghz
(C)Ps4-pro:36x0,91ghz
(D)Ps4:18x0,8ghz

Yes, I've been thinking about different operation modes for a while, hence my "bet" in the friendly wager thread. Although my idea would be a bit more "fluid" with developers able to mix and match with CPU clock like I think they do on the Nintendo Switch? That and VRR should help to extract as much performance as possible from a baseline "weaker" specs.
 
Is it too far fetched to think that PS5 APU has 3 blocks of 18 CU's each for a total of 54 CU's and the 18/36 BC test was done with one block disabled? Maybe the last block contains RT hardware that did not show in test? Still strange the BC test was also done with 2 Ghz, but it needs to be tested at this speed if Sony have chosen so. Maybe I am just posting bullshit.....o_O
 
No. TF is just about MADDs and nothing else. Correct me if i'm wrong.
I don't know any specifics about the instructions, but all I've seen points to RDNA doing more per cycle, such as the transcendental units. RDNA can do more SINE ops per second than GCN, therefore it can do more floating point operations per second. The TF figure is only the fastest number of flops possible, but that doesn't represent true workloads. However, there's no naming convention to differentiate between best-case flop rate and real-world flop rate. Kinda like the difference between horse-power and brake-horse-power. It's the difference between comparing PS2's triangle draw rate with OXBs - sure it can do way more triangles per second, but to get the same results needs lots more triangles, so in the end that number isn't representative. You'd want a count of how many lit, shaded triangles you can perform a second to compare machines.

That's where benchmarks come in, and where this strange NaviFLOP has come from as a way to compare numbers that otherwise can't be meaningfully compared. Instead of using peak rates on specific, unrealistic workloads (see Cell's SPUs), we want a way to compare usable work, but that just doesn't, and probably can't, exist, leaving all these awkward paper specs to compare and try to understand what's going on.

Short answer is, processors are too complex to be described with a number, but human beings can't help themselves and will keep talking about flops. ;)
 
Which is exactly what this number is telling.
It is NOT a measure of overall performance, so if some people use it as that, it is their mistake, not mine.

I'm listening...

No. TF is just about MADDs and nothing else. Correct me if i'm wrong.

Yes, it CAN. It depends on the workloads and their implementations. Which brings us to noise an uncertainty. To prevent this, the TF number has been introduced, to have a measure that does not depend on any uncertain factors.

Again. The impression NAVI > GCN comes probably mainly from improved rasterization performance, resulting in much better game fps.
Why is a NAVI CU larger than GCN CU? More Simds? Nope. More LDS? Nope. Twice the ROPs? Yes. (IIRC)
So i'm sorry to inform you, but there is no indication about higher compute peak performance, aside the things i've said above.
If i'm wrong let me know, but you need to make a specific point.
Doesn't matter what tflops means, in context of speculative thread like this tflops mean power in game that we can compare that's why navi is like 1.25-1.3x per flop comparing to xbox one x (5700xt vs rx580 game performance in 4k on avarage) even though it's only simplistic help multiplayer
 
Few days and we will know more... I don't think Sony will let spread for too long the 9.2 tf rumor... sure we are (bit) above 10 with ps5...
 
Few days and we will know more... I don't think Sony will let spread for too long the 9.2 tf rumor... sure we are (bit) above 10 with ps5...

Honestly, if PS5 is a 36CU machine the final TF number could start with 8. I'm quite skeptical of them reaching 2.0 GHz for retail box (with good enough yields).
 
Few days and we will know more... I don't think Sony will let spread for too long the 9.2 tf rumor... sure we are (bit) above 10 with ps5...
I think keep the expectation low will be better idea, so you either getting expected or surprised numbers, not disappointed.
 
9.2 TF with much lower bandwidth next to 12 TF with much higher bandwidth is disgustingly low, can't believe Sony would prioritize BC and compromise the CU count, the same amount of CUs as PS4 Pro on their nextgen system..shake my head in full disappointment. The more I read each day the more I think Klee got it all wrong or just not up to date. I wonder how fast a 12 TF RDNA XsX compares to a 2080, maybe on equal footing if not beating it most of time in a console space.

The interesting thing was from the Day where Microsoft official annouced the new Xbox X Series , Klee going out the Discussion on Resetera Forum , and dont answer any Question, he said that he maybe come back later to the Forum when Sony make a official Statement about Ps5 and the Specs behind. So that show me that he dont know the real Ps5 Specs, because he was surprised that MS pushing the powerful 12 Tflops route , or the Ps5 Specs are not final , or he has fooling the people with wrong Infos , like Fud or damage Control. If he is not an insider , a normal User on Resetera , it doesent matter what Microsoft has published (Specs etc.) so why did he leave the discussion?

Strange.....
 
Last edited:
The interesting thing was from the Day where Microsoft official annouced the new Xbox X Series , Klee going out the Discussion on Resetera Forum , and dont answer any Question, he said that he maybe come back later to the Forum when Sony make a official Statement about Ps5 and the Specs behind. So that show me that he dont know the real Ps5 Specs, because he was surprised that MS pushing the powerful 12 Tflops route , or the Ps5 Specs are not final , or he has fooling the people with wrong Infos , like Fud or damage Control. If he is not an insider , a normal User on Resetera , it doesent matter what Microsoft has published (Specs etc.) so why did he leave the discussion?

Strange.....

He left because the place was becoming toxic. He mentioned multiple times that the XBSX reveal did not change what had said earlier and that it was inline with what he had been told and was expecting. Specifically he said both where double digit TF and both where quite close to each other performance wise.
 
He left because the place was becoming toxic. He mentioned multiple times that the XBSX reveal did not change what had said earlier and that it was inline with what he had been told and was expecting. Specifically he said both where double digit TF and both where quite close to each other performance wise.

I dont believe him , but this is OT here. ;)
 
You are not wrong, TF to TF, RDNA is no different from GCN, it's just that RDNA can extract more rasterization performance per TF than GCN.

I think a better way of putting it is that RDNA more efficiently uses (or makes better utilization) of the compute resourses (in terms of FLOPs) that it has in...and this is important...typical gaming workloads than GCN.

However, in other cases where software was already getting really good utilization of compute resources, there is little to no change in compute efficiency WRT FLOPs. Hence, AMD using GCN for professional compute heavy workloads would likely see little to no difference if they were instead using RDNA.

Regards,
SB
 
Last edited:
Status
Not open for further replies.
Back
Top