AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread

I think it would be a nice sidekick for Kabini/Beema but not much else. Even low-end Kaveris have 128bit memory, and only the 1 module versions have less than 6 CUs. Though arguably AMD just loves the crossfired configurations (so a 64bit 5CU chip with probably higher clock than what some 128bit 6CU Kaveri has might make some sense), even though IMHO these configurations have near zero merit.
For the record though I don't think it's a rename of HD 8570M, since I suspect with these specs it's using Sun die now and not Mars/Oland (Sun is natively 64bit, 5CUs, no display controller/uvd etc.). AFAIK all the existing mobile chips were Mars.

Is Sun the mobile Hainan?
 
I actually always confuse the "standard" and mobile parts anyway (reason mostly because I see them as entirely the same, nothing which would warrant different codenames). Also with Hainan I actually have no idea what it's useful for, since it doesn't get its own display outputs you can't build a desktop graphic card with it anyway (well theoretically you could leverage the cpus igp display outputs just like with notebooks, but even then that thing is too low end to be really useful in such a role). I'd be interested to hear though if any such products exist or how they'd look like if they are planned...
 
Are GCN cores in all "new" chips equal?
e.g. if we have 1 "bonaire" (260) core and comapre to 1 tahiti/hawaii core, will the theoretical performance equal, given same speed memory access?
 
Are GCN cores in all "new" chips equal?
e.g. if we have 1 "bonaire" (260) core and comapre to 1 tahiti/hawaii core, will the theoretical performance equal, given same speed memory access?
Tahiti is going to be different as it's gcn 1.0 not 1.1 thus at least lacks some instructions, other than that I'm not sure if the "cores" themselves would perform differently in some scenarios, obviously the peak flop number is the same per CU. Also the "native" DP rate is different between bonaire and tahiti (I still haven't seen any confirmation of hawaii's DP rate).
There's obviously differences between these chips outside the CUs (between GCN 1.0 and 1.1 as well as obviously the number of units e.g. shader engines, rops etc.) so I'm not sure the question makes much sense.
 
My question is about theoretical math speed of cores, eg if there is any comparison of achievable speed in units.
Say maybe GCN 1.0 can achieve/sustain 90% of max and GCN1.1 can go as high has 95%

I believe there was some kind of program able to test speed of math instructions, but can't remember the name and obviously don't see any review doing such comparison.
Am I right to assume that changes between 1.0 and 1.1 are not big and for calculations number of cores&clock are way more important than supposed enhancements?
 
My question is about theoretical math speed of cores, eg if there is any comparison of achievable speed in units.
Say maybe GCN 1.0 can achieve/sustain 90% of max and GCN1.1 can go as high has 95%

I believe there was some kind of program able to test speed of math instructions, but can't remember the name and obviously don't see any review doing such comparison.
Am I right to assume that changes between 1.0 and 1.1 are not big and for calculations number of cores&clock are way more important than supposed enhancements?
I think that's quite right. Things like the more compute queues might help a lot in some cases, but I would believe that (unless you could leverage the new instructions) there's not really anything which makes the CUs more efficient in general. I am definitely not sure about this as there could be architectural changes behind the scenes.

Hawaii is 1/8 on Radeon-models, and according to Tom's would be 1/2 on FirePros
Well unless Tom's specifically quotes someone from AMD I don't consider that as confirmation :). Of course there have to be some changes for different non-native rates too (if the 1/8 really isn't native) but these should be tiny.
 
Hawaii is 1/8 on Radeon-models, and according to Tom's would be 1/2 on FirePros

1/2??

that's sound strange... 24x2=48, you still miss 4 bit for the mantissa in double, and I guess it would just be a waste to add it to all chips just for... firepro's market?


I doubt it...
 
Some volcanic rumbling in the last drivers? ;)

AMD6900.1 = "AMD Radeon Series"
AMD6900.2 = "AMD Radeon Series"
AMD6900.3 = "AMD Radeon Series"
AMD6900.4 = "AMD Radeon Series"
AMD6900.5 = "AMD Radeon Series"
AMD6900.6 = "AMD Radeon Series"
AMD6900.7 = "AMD Radeon Series"
AMD6900.8 = "AMD Radeon Series"
AMD6900.9 = "AMD Radeon Series"
AMD6900.10 = "AMD Radeon Series"
AMD6900.11 = "AMD Radeon Series"
AMD6900.12 = "AMD Radeon Series"
AMD6900.13 = "AMD Radeon Series"
AMD6900.14 = "AMD Radeon Series"
http://forums.guru3d.com/showpost.php?p=4751651&postcount=3

AMD6920.1 = "TONGA (6920)"
http://forums.guru3d.com/showpost.php?p=4618625&postcount=1
 
Dual Hawaii is/was Vesuvius (maybe canned):
AMD67A2.1 = "HAWAII GL Gemini (67A2)"
AMD67A1.1 = "HAWAII GL40 (67A1)"
AMD67BE.1 = "HAWAII LE (67BE)"
AMD67B1.1 = "HAWAII PRO (67B1)"
AMD67B0.1 = "HAWAII XT (67B0)"
AMD67A0.1 = "HAWAII XTGL (67A0)"

AMD67B9.1 = "VESUVIUS (67B9)"
http://forums.guru3d.com/showthread.php?t=379451

Other VI GPUs (maybe some 20nm or already canned):
Added AMD Volcanic Islands/Crystal: Maui, Iceland/Topaz, Tonga/Amethyst.
http://www.hwinfo.com/txt/History3264.txt

This AMD6900 devices could be Tonga testing boards.
 
Last edited by a moderator:
I think remember the name " vesuvius " have been set for the dual Hawaii. I dont know if it is the official codename from AMD or something like dual gpu's from Asus or other name ( they are listed under driver )
 
With Maxwell getting to be released soon, is there anything at all know about upcoming silicon at AMD?

TBH it feels like the whole Volcanic Islands is kinda "backup plan" with 20 nm coming available later than anticipated, so now they play old chips and 1 new high end chip 'till 20 nm is available
 
Back
Top