To be fair, this is the "make shit up" thread.
Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]
- Thread starter BRiT
- Start date
- Status
AbsoluteBeginner
Regular
That I agree with, but I also think we can go with what we know - technically. That is, Gonzalo is most probably PS5 SOC with GPU part running at 1.8GHz boost. This would, along with what we know about desktop NAVI 5700XT part, mean that its most likely not over 40CU part. If we get chip resembling Scarlett production SOC, and it is clocked at 1.8-1.9GHz I would expect shown die size to be fake (or shit ton of RT units).
In any case, Gonzalo leak most likely tells us that :
a) This is PS5 SOC
b) Its latest revision is clocked at 1.8GHz boost
c) Its therefore likely not a chip with more then 40CUs active
AbsoluteBeginner
Regular
40CUs @ 1.8GHz would yield better performance then 12TF Vega GPU. 36CU @ 1.8GHz would yield better results then Vega 56 at 10.6TF.
So IMO, if we get RT hardware alongside this, it would be anything but disappointment. People are clinging on CU number, but CUs are wider in NAVI and therefore not comparable at all. Its like putting RTX2700 inside console and being disappointed.
Speaking of, I do wonder about the implications of the midgen butterfly effect (
) in the context of Navi. The Navi shader engine (NSE?) looks roughly akin to what would have been either of the Sea Island base console GPU configurations (i.e. 2 primitive/setup units, 2 sets of CUs, up to 8 RBEs etc). At least based on Scorpio, the number of CUs attached to the setup engine (distributor?) seems less of a problem for compatibility - It Just Works. Then 2NSEs would mimic the midgen twins.*That said, it'd be a fun coincidence if 5 DCUs (dual compute units) in Navi 10 is akin to the 10 CUs per shader engine in Liverpool/Orbis & Neo - not that it should matter... I guess, "why not" if they were trying to build a mid-high range desktop part and needed to settle on something and they were simultaneously building something else andfaisd;flnkae rfinqalfasdf who knows ¯\_(ツ)_/¯ (It shouldn't matter - Pitcairn 2.0 :V ).
-----
*(Only a little annoyed at the re-organized definition of Shader Engine, but maybe there's a non-marketing reason for it????????)
Last edited:
What matters most is the instructions it supports and then the throughput for the instructions.
Why would BC require a multiple of ps4's number of CUs?
Why not disable whatever number of additional ones in the lowest compat mode, if all busses are at least as fast?
Patents are showing a massive amount of work to make any arch behave identically for BC purposes. Including narrowing busses and caches for identical behaviours. This wouldn't be required with the ps4pro method.
Why not disable whatever number of additional ones in the lowest compat mode, if all busses are at least as fast?
Patents are showing a massive amount of work to make any arch behave identically for BC purposes. Including narrowing busses and caches for identical behaviours. This wouldn't be required with the ps4pro method.
hm? # of CUs within a grouping shouldn't matter. Who are you talking to?
Edit: It's pretty clear things can be done as evidenced on PC (drivers). Not sure why patents are required. Binary compatibility in the absence of system level drivers? Which patent is specifically discussing the GPU side
Last edited:
HBRU
Regular
My Pro-2 dream coming true... ... I think the 1,822 Ghz clock boost can be true but only on part of the chip... So I belive more in something like 18x3=56@1500 with only 36 CUs that can reach 1,822 Ghz..
AbsoluteBeginner
Regular
Interesting few tweets from Komachi. He thinks two IDs for Gonzalo means it will be HBCC on that one (as all Vega cards with HBCC have 2 IDs). He also speculates in his ID table that Navi12 or Navi20 are Arden related, we shall see.
From currently available info, I think both consoles are under 10TF, with Gonzalo being PS5 chip with 36CU part @1800MHz.
From currently available info, I think both consoles are under 10TF, with Gonzalo being PS5 chip with 36CU part @1800MHz.
Wondering if the development kit rumors are due to a misunderstanding of the design philosophies. I could see MS including dedicated RTRT hardware while Sony allocated more of their budget for more general compute which could be used for RT or additional FPS with PSVR. Comments without context could very easily lead to misunderstanding about relative power of each platform.
Please post links or tweets next time, especially dealing with speculation or rumors.
Update:
Last edited:
HBRU
Regular
The 36@1822 mhz kits may just mean -in the beginning- they are just targeting to upgrade ps4-pro@30 fps titles going to be produced (and alredy on the market) at 60 fps in some sort of automatic turbo mode... I belive the final specs will be 54 CU @1500 mhz (10,32 TF)... with the capability of reaching 1822 mhz for just 36 CUs... (8,4 TF)
The 54@1500 mode (ps5 mode).... will also unlock the specific upgrades of the Navi GPU...
They may work now maybe with just normal ps4-pro chips under liquid Nitrogen cooling....
The 54@1500 mode (ps5 mode).... will also unlock the specific upgrades of the Navi GPU...
They may work now maybe with just normal ps4-pro chips under liquid Nitrogen cooling....
Last edited:
My speculation is that these "LITE" Navi byproducts aren't Sony/MS related GPUs at all. But rather cut-down Navi counterparts possibly relating to the AMD partnership with Samsung (cellphone, mobile devices, car entertainment video/audio systems, etc.) or cheap Chinese AMD APU solutions (i.e., steam boxes, etc.) related to the booming PC gaming market in China.
No. No it wouldn't.
- Status
Similar threads
- Locked
