Advertisement
Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- Blueisviolet aka Mistercteam
- PS5 data
- ==========
- march 27 2020
- ps5 firestrike archive http://archive.vn/BY7qC#
- original link https://www.3dmark.com/fs/19011253
- twit https://twitter.com/blueisviolet/status/1243221588679389184 (image etc)
- flute repost
- https://twitter.com/blueisviolet/status/1243209600825876481
- Jan 23 2020
- older but close
- almost 1 year ago (May 21st 2019)
- https://pastebin.com/YQuqh7Mv
- suprisingly
- listed 56CU of XSX
- also 32CU of X2 lite
- (even for me MS wont release this X2 lite)
- also 36-40CU of PS5
- all in May 2020,before all digging,before RDNA2 and RT thing become more clear after E3 2019
- Jan 21
- Verona truth
- https://pastebin.com/bgmfUzfp
- https://pastebin.com/ti53eXFE
- other
- https://pastebin.com/vdBcGgrX
- Jan 10 2020
- Jan 12 2020
- added link to arden_ariel_github backup
- (found online)
- Jan 19 2020
- added this slide
- https://i.imgur.com/NeOouYC.png Full Chip resule=36CU, also Ariel still need fixing etc
- also indicate oberon achieved mostly 72-80% theoritical target.
- https://i.imgur.com/YiFTUQx.png 21 Jun 2019, Ariel native,= 18 WGP=36 CU
- https://i.imgur.com/Bammjgg.png ariel goes to E0, probably as still need fixing
- Platform/board
- Flute
- Gonzalos
- SOC devkit Oberon
- SOC retail Ariel
- Link
- ================
- arden ariel github test (backup): (found online)
- https://mega.nz/#!JMtSnapD!M6yQ-CNQqzBKI0tY7UMmFLAHqBZJhshYalg4UdFXNi8
- Resume from below data
- ==========================
- 13F9 = Latest Oberon/Ariel = 00840F40
- FLute CPU semms only as fast as Ryzen 1700x but better scaling
- from Latest Apisak probably Stepping E0 is for better clock etc
- Komachi listed latest ARL A0 and E0, where E0 still related to 00840F40
- Rogames =
- based on github Ariedl 36CU 448 GB/sec,
- also no VRS/RT per github link
- for me RT/VRS is super important why Ariel benchmark not listed VRS/RT?
- answer:
- if forum said because it will be tested later, then not making sense, as Arden is tested
- considering Ariel gpu already several iteration not make sense B0 rev they still not
- testing VRS etc
- definitive answer:
- because they dont have it, as simple as that and PS5 will rely on Prorender method (Shader)
- or HW GPU to do RT, still accelerated vs GCN for sure
- github Leak
- ===============
- Oberon/Ariel A0-B0 , also full chip 36 CU, https://i.imgur.com/Lv0e6Xh.png
- Ariel A0 clearly clocked <2Ghz https://i.imgur.com/Il6mwn0.png
- my own dig based on vulkan etc
- ===============================
- https://twitter.com/blueisviolet/status/1163883749651562497
- Navi10 aka RDNA1=max 40 CU
- (Navi10/rdna1 shared some GCN charateristic)
- max 2SE,/SE=max 2SH,/SH=max 10CU
- PS5=Navi10based=40CU (max)
- and if XSX has 56 CU clearly=RDNA2 !!=Navi21!!
- some other data:
- =========================
- https://hexus.net/tech/news/cpu/133001-amd-flute-soc-zen-2-cpu-navi-10-lite-gpu-spotted/
- AMD Flute/100-000000004-15_32/12/18_13F9
- QS/8/16/1.6GHz/3.2GHz/Navi 10 Lite/13F9
- ?2Ghz
- AMD Gonzalo/ZG16702AE8JB2_ 32/10/18_13F8
- QS/8/16/1.6GHz/3.2GHz/Navi 10 Lite/13F8
- 1.8GHz
- AMD Gonzalo/2G16002CE8JA2_ 32/10/10_13E9
- ES/8/16/1GHz/3.2GHz/Navi 10 Lite/13E9/
- 1GHz
- this is the important thing about flute=latest data
- =====================================================
- https://twitter.com/MehNitesh2/status/1153305440299737089
- https://twitter.com/peresuslog/status/1153390297013002245
- with L3 4MB per CCX, as we Know CCX normally 16MB
- Z2 chiplet is 7nm 80mm2= 8core,L3=32MB
- https://en.wikipedia.org/wiki/Zen_2
- now some backup slide as flute is deleted from userbenchmark
- showed cut down L3 4MB https://i.imgur.com/20IhaOl.png
- showed Z2 PS5 perf, https://i.imgur.com/6b0TVUS.png
- 1 core = 94.x
- Q core = 376
- M core = 1071
- as Comparison Ryzen 1700 (Zen1)
- https://cpu.userbenchmark.com/SpeedTest/266290/AMD-Ryzen-7-1700-Eight-Core-Processor
- 1 Core = 118
- Q core = 435
- this Gaf user doing good research
- and they dont want the thread surfaced !!
- https://www.neogaf.com/threads/simulating-gonzalo-rumoured-nextgen-ps5-leak.1493305/
- Some data
- ============
- Komachi
- https://twitter.com/KOMACHI_ENSAKA/status/1215394283752062976
- 00840F00 = ARL A0
- 00840F40 = ARL E0
- 00840= Oberon !! still 13F8/F9
- https://twitter.com/komachi_ensaka/status/1153157871963521030?lang=en
- 13F9/Flute/1.6 Ghz base clock/4MB-CCX L3 ? 20mm2
- "13F9 is in the range of Ariel's PCI-ID which I guess is PS5 (13E0~13FF). In other words, yeah. I think this is a custom chip for consoles. However, it is a mystery that numbers like OPN are attached."
- User Benchmark
- https://gpu.userbenchmark.com/SpeedTest/856185/13F9-OBR-A0
- 13F9-OBR
- APISAK
- ===========================================================
- https://twitter.com/TUM_APISAK/status/1213637902082564097
- per Jan
- ES= ESample stil accurate 13F9/F8 Oberon
- Gonzalo Benchmark June 2019
- https://twitter.com/TUM_APISAK/status/1143418629666430976
- 20K ~ equal to 7-8TF 36CU
- April
- https://twitter.com/TUM_APISAK/status/1115882516240633860
- ZG16702AE8JB2_32/10/18_13F8
- Before April
- https://twitter.com/TUM_APISAK/status/1138332197004251136
- Gonzalos 13E9, 13E8, Navi10 lite
- roGames data
- later
- =====
- ariel 36CU/18WGP (Full CU)
- Oberon A0/B0 2ghz , 448 GB/sec
- https://i.imgur.com/HZAWhKn.png?1
- Ariel/oberon no RT/VRS
- https://i.imgur.com/0SCkwSa.png
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement