The point is power and area savings, there's really no difference in performance (running out of memory it's going to get slow regardless)
AMD Navi Product Reviews and Previews: (5500, 5600 XT, 5700, 5700 XT)
DavidGraham
Veteran
RX 5600 XT full specs:
36CU (same as RX 5700)
game clock 1460MHz (vs 1625MHz for RX 5700)
12bps memory on a 192 bit but bus (vs 14bps on 256 bit bus for RX 5700)
single 8 pin power connector
https://videocardz.com/newz/asrock-leaks-radeon-rx-5600-xt-challanger-d-oc-specifications
36CU (same as RX 5700)
game clock 1460MHz (vs 1625MHz for RX 5700)
12bps memory on a 192 bit but bus (vs 14bps on 256 bit bus for RX 5700)
single 8 pin power connector
https://videocardz.com/newz/asrock-leaks-radeon-rx-5600-xt-challanger-d-oc-specifications
Fair bet with those specs that it’s based on Navi 10.
Obviously it is, now the big question is "wtf is Navi 12 then". Entry-level GPU? Big 1st gen Navi? Planned but scrapped design?
DmitryKo
Veteran
So Arcturus is not on the Navi architecture, as I hoped it to be based on a few leaked bits, but is actually a high-end Vega 30 (or dual Vega 20?), with 128 CUs (8192 shader units) and HBM memory -probably a 7nm+ mid-2020 refresh. Am I reading this right?
Last edited:
It's apparently "pure compute chip" in the sense that "it doesn't have 3D engine" (graphics command processor? displaycontroller?)
Frenetic Pony
Veteran
Yah, AMD said they'd be splitting compute/gaming before right? Redesigning Vega to eliminate all the raster stuff should allow more room for shoving compute engines in, and Vega was more efficient at compute per mm than Navi anyway.
I wouldn't really expect "new RDNA" to hit till the end of the year, as a very large portion of foundry time is going to go towards PS5 and Xbox blech production, probably starting in the next few months. Though I wouldn't doubt AMD hopes to get a mainstream "6700" series or whatever out before years end next year.
D
Deleted member 13524
Guest
A waste of bandwidth, unless Navi 12 is either:
1 - a high end mobile part
2 - clocking well over 2GHz
3 - has a higher number of ALUs per CU
4 - Is aimed at a niche market like Arcturus with very specific performance targets
5 - a combination of the above.
Rootax
Veteran
D
Deleted member 13524
Guest
Option 1 then.
OTOH, higher number of ALUs per CU could be on both Navi 12 and Navi 14.
A 40 CU Navi 12 with 96sp per CU would have 3840sp total (justifying the use of dual stack HBM), and a 2304sp Navi 14 would have 24 CU.
Last edited by a moderator:
Except that we already know Navi 14 is 12 DCU (24 CU) / 1536 sp part
D
Deleted member 13524
Guest
We do?
Well, we have 3 clearly separate Navi 14 configurations which have 64 SP per CU, as well as a chip whichs size scales logically for 24 CU part compared to 40 CU Navi 10.
Unless you're suggesting they're firmware/software limited to work as 64 SP per CU parts with their 96 SPs taking physically same space as 64 Navi10 SPs, then yes.
edit: SKUs > configurations, of course there's more SKUs but IIRC there's 3 different configs they share
Last edited:
I honestly can't see Intel doing another G-model with AMD graphics in the midst of their push to discrete graphics on their own
D
Deleted member 13524
Guest
Oh I thought the RX5500 series were Navi 11, not Navi 14. Then this RX5600 is most probably just a cut down and lower binned Navi 10.
Could Navi 12 be an APU or similar design to the Kabylake G? Maybe AMD will make a console like chip for PC all in one solution or maybe someone like Apple or Microsoft will be bringing out a mass market mobile workstation. It would make sense in form factor constrained spaced to use HBM even if it is more expensive.
