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

Status
Not open for further replies.
I know, thats why I said Scarlett setup from video does not provide enough for 12TF. Sure it does for 16Gbps, but this doesnt help in Flute case (PS5).

I cannot see them going below 50GB/s per TF, because PS4Pro was bw limited at 218GB/s.

This is how things would look in rumoured cases :

Flute bench - 256bit bus and 18Gbps and 16Gbps chips with 13TF GPU

576GB/s - 44GB/s per TF
512GB/s - 39GB/s per TF

For reference PS4 and PS4Pro

176GB/s - 95GB/s per TF
218GB/s - 52GB/s per TF

Now, bw per TF is going down with time, but I doubt its gone that much. PS4 Pro was bw limited with 50+, so I dont see how low 40s would be acceptable (especially considering Zen2 and RT are on die this time around).

+1
 
Can anyone tell me why would MS go wide (64CU) and slow (1.4GHZ) if that clock is comfortably below RDNA sweet spot (which I assume will be even higher in RDNA2)?
Could work out more effective to bin chips for Lockhart that way. Although then Lockhart ends up with a pretty big APU just with a lot more disabled CU's compared to having it be a different chip.
Maybe having more CU's works out better for RT compared to having everything run faster with less CU's.

The norm is to disable CU's for yield purposes, so if they haven't gone that route, will have a good reason. (guess obvious statement)
 
Could work out more effective to bin chips for Lockhart that way. Although then Lockhart ends up with a pretty big APU just with a lot more disabled CU's compared to having it be a different chip.
Maybe having more CU's works out better for RT compared to having everything run faster with less CU's.

The norm is to disable CU's for yield purposes, so if they haven't gone that route, will have a good reason. (guess obvious statement)
You will always disable CUs for yields, so having fewer them is still advantage.

Lockhart using Anaconda dies doesnt make sense either, since it would be roundabout 4TF, so even if Anaconda had 52CUs (or 48) with up to 20 defective ones, they would easily hit Lockhart target.

64CUs make 0 sense.
 
Such definitive statements in such a baseless discussion amuses me.
 
Klee is not a reliable source.

More telling is the goalpost shifting between camps when it comes to insiders.

Weeks before…
Klee: The PS5 has a slight edge over Xbox-Next
Xbox Camp: Klee is lying. Don’t believe him. He's no insider.
PlayStation Camp: Jerks-off with glee…

Now…
Klee: XBSX has 64CUs (12TF)
PlayStation Camp: Klee has no official info on the PS5. He has been wrong lately. He's lying.
Xbox Camp: Jerks-off with glee…
 
More telling is the goalpost shifting between camps when it comes to insiders.

Weeks before…
Klee: The PS5 has a slight edge over Xbox-Next
Xbox Camp: Klee is lying. Don’t believe him. He's no insider.
PlayStation Camp: Jerks off with glee…

Now…
Klee: XBSX has 64CUs (12TF)
PlayStation Camp: Klee has no official info on the PS5. He has been wrong lately. He's lying.
Xbox Camp: Jerks off with glee…
Yeah it's par for the course for the internets, they only use the insiders that fit with their wishful thinking, and there's enough choice to go around.... but the attitude of "everyone should believe me without question" and the "thank you for the kind words and believing in me!" and then whining if someone doesn't believe him. WTH, the story is that he's using his "friend" breaking a serious NDA and he takes credit for spreading it on the internet. It's the attitude that brings up big red flags.
 
Can anyone tell me why would MS go wide (64CU) and slow (1.4GHZ) if that clock is comfortably below RDNA sweet spot (which I assume will be even higher in RDNA2)?

Thats literally doing two things, and neither is good :

1. You leave performance on table. A lot of it. Especially if your competitor clocks it at almost 50% higher clocks

2. You get bigger chip with worse yields...and performance to boot.

Its lose lose situation, and any insider that stands behind that rumor loses credibility in my eyes.

Entire GPU design is shifting towards higher clocks, especially with process being so expensive, so I am absolutely baffled MS would ditch "sweet spot" strat they had for all 3 gens till now and go below it...for what?


Risk and Reward.
By going for a wide and slow approach they can basically guarantee a specific level of performance, and if it is 64CU wide, that will be a pretty high performing part, even @ 1.4Ghz.
I think that is a reasonable goal to reach given the constraints of: power budget + good dies per wafer.
So in one way it's a safe bet, BUT...
There is enormous potential for an upclock prior to release. target 64CU @ 1.4, and then 1 month prior to release they will have FAR more information about the actual performance of retail boxes,
and if there is any possibility of an upclock they can do it easily. I'm sure MS would be quite happy to dump and extra 50 - 150Mhz on the GPU clock if possible, in the last few months.

However with the opposite approach ie. thin and fast, your GPU MUST clock at high Mhz to get the perf you need, or your done for.
And 50Mhz on a 52CU 2Ghz clock GPU part aint gonna move the needle much - especially relative to a 64CU part @ 1.4 ghz, where even a meager 50Mhz will have significant improvements.

I actually think it's a smart move the only downside is that it makes for a more $$ console. But potentially less strict cooling requirements.
Also while we have some good data for the "sweet spot" of power/perf for a 40CU Navi/RDNA Gpu, we dont have that info for a 64CU part, it is likely in the same ballpark, but probably not exactly the same.

Additionally we don't know how well RDNA 2 RT resources scale? perhaps they scale much in line with CU count, than they do with clock speed, so IF thats the case a 64CU unit, may perform much better at RT workloads, than a faster 52 CU unit...
 
Risk and Reward.
By going for a wide and slow approach they can basically guarantee a specific level of performance, and if it is 64CU wide, that will be a pretty high performing part, even @ 1.4Ghz.
I think that is a reasonable goal to reach given the constraints of: power budget + good dies per wafer.
So in one way it's a safe bet, BUT...
There is enormous potential for an upclock prior to release. target 64CU @ 1.4, and then 1 month prior to release they will have FAR more information about the actual performance of retail boxes,
If we assume Klee's comment of 64CUs is true, then his other info is worthy of noticing (PS5 is 10% more powerful than xsx).

Sony must have some extremely good cooling solution (for example, V shaped case) so they can clock higher. I am very curious to see the retail version.
 
Can anyone tell me why would MS go wide (64CU) and slow (1.4GHZ) if that clock is comfortably below RDNA sweet spot (which I assume will be even higher in RDNA2)?

In general, it's easier to go wide and slow versus narrow and fast. On the face of it, the biggest trade off is in transistors used. But it's not always that simple either as sometimes you require a lot of transistors just to clock everything higher AND feed everything at that speeds AND have provisions to avoid stalls. That is, of course, an overly simplistic view of it.

The Pentium 4's Netburst architecture is one such example where pushing for high clock speed isn't necessarily easy or a good idea. There were a lot of compromises made in order to architect the chips for high clock speeds (like long instruction pipelines).

With absolutely no details about what MS or Sony are doing other than some questionable "sources," there's not much to say.

Regards,
SB
 
You will always disable CUs for yields, so having fewer them is still advantage.

Lockhart using Anaconda dies doesnt make sense either, since it would be roundabout 4TF, so even if Anaconda had 52CUs (or 48) with up to 20 defective ones, they would easily hit Lockhart target.

64CUs make 0 sense.
Even if it was 64CU does that mean 0 disabled? Thought 64 wasn't hard and fast rule.

Also why would you always have disabled CU's when defective ones can still be used by disabling those CU's and using them in the Lockhart?
The biggest issue with it is Lockhart die size/cost due to the fact that probably will end up using dies that are good and could be used on XSX
 
Status
Not open for further replies.
Back
Top