AMD: Speculation, Rumors, and Discussion (Archive)
- Thread starter iMacmatician
- Start date
- Status
I thought the narrative from AMD in the past was Polaris 10 was desktop with Polaris 11 being mobile?
If the leak is suggesting a Polaris 10 mobile with 5.5Tflops, then what was the GPU compared to NVIDIA 950 back at CES????
That would make Polaris 10 covering at least two positions early on in the desktop plus an early laptop GPU more powerful than at least one of the Polaris 10 desktop GPUs and also with more compute performance than a 390 (great if was true for laptop gamers).
But sounds more like a desktop part to me, which is what TPU infers as well.
Possibly their lowest Polaris 10 desktop (mainstream version and not the Performance segment one) shipping first *shrug*.
Cheers
If the leak is suggesting a Polaris 10 mobile with 5.5Tflops, then what was the GPU compared to NVIDIA 950 back at CES????
That would make Polaris 10 covering at least two positions early on in the desktop plus an early laptop GPU more powerful than at least one of the Polaris 10 desktop GPUs and also with more compute performance than a 390 (great if was true for laptop gamers).
But sounds more like a desktop part to me, which is what TPU infers as well.
http://www.techpowerup.com/222450/more-polaris10-and-polaris11-specifications-revealed
Possibly their lowest Polaris 10 desktop (mainstream version and not the Performance segment one) shipping first *shrug*.
Cheers
D
Deleted member 13524
Guest
Looks like next Thursday could be the formal unveil:
https://twitter.com/AMD/status/731158531664171008
2 weeks before Computex.
It may just be a presentation about the architecture so they still may not show end products, but it'll be something nonetheless.
EDIT:
Looks like the GTX 1080's NDA lift is May 17, and this is the day after that.
If this is indeed a product reveal they seem to be showing some confidence with what they're showing.
https://twitter.com/AMD/status/731158531664171008
2 weeks before Computex.
It may just be a presentation about the architecture so they still may not show end products, but it'll be something nonetheless.
EDIT:
Looks like the GTX 1080's NDA lift is May 17, and this is the day after that.
If this is indeed a product reveal they seem to be showing some confidence with what they're showing.
Last edited by a moderator:
Not sure it would be truly an unveil because no mention of Raja or Lisa on the agenda.
Maybe a more detailed webinar following on from that CES event (where they showed the low power Polaris).
Cheers
I wouldn't read too much into that, confidence or forced hand it can be either of the two.
Polaris 11 targeting the notebook segment, Polaris 10 targeting mainstream desktop and high-end gaming notebook segments.
http://api.viglink.com/api/click?format=go&jsonp=vglnk_146317669812012&key=7777bc3c17029328d03146e0ed767841&libId=io69i1f601000kb5000DAl6b4rtsw&loc=http://www.overclock.net/t/1598134/vc-amd-officially-confirms-polaris-10-and-polaris-11-market-positioning&v=1&out=http://phx.corporate-ir.net/External.File?t=1&item=VHlwZT0yfFBhcmVudElEPTUyMjMwODZ8Q2hpbGRJRD02MzAyNTE=
Last edited:
Thanks,
I guess we will have to see who interpreted it right between TPU and Videocardz from their source.
Will be good if that is the mobile part though.
Thanks again.
Last edited:
Last edited:
renderstate
Newcomer
???
Those specs are from Tonga and Bonaire, right?
Yeah. It is not hard to make such a screenshot... Every browser has an in-built inspector nowadays.
My main prediction for the architecture reveal: On Polaris, they have concentrated on compute efficiency and reduction of stalls over raising the theoretical peak performance. Concrete steps taken include doubling the VGPR files and reducing latencies. Polaris will perform significantly better at the same clock and same amount of CUs as previous GCN, but CU counts can't rise as much as the shrink would hint simply because the new CUs take substantially more transistors.
No sources for this, I just think it would make sense.
No sources for this, I just think it would make sense.
Are there compute efficiency issues with GCN?
Infinisearch
Veteran
Aren't perf/watt problems compute effieciency issues?
No, not at all.
Perf/W is about doing things in a power inefficient way. Compute efficiency is about using the available compute resources as efficiently as possible.
Fermi had terrible perf/W yet had great compute efficiency.
Love_In_Rio
Veteran
R9 M485X with 32 CUs,1,25 Ghz and 256 bits bus.
R9 M470X with 14 CUs,1,5 Ghz and 128 bits bus.
The M of Mobile right?.
If they can reach 1,5 GHz with a mobile part maybe AMD has been trolling everybody all this time with those 800Mhz-1050Mhz leaks...
With those clocks in a 40 CUs GPU they would surpass Fury performance easily...And we would be talking about the best architecture ever.
If such a jump is true no wonder console vendors chose to iterate instead of shrinking.
Last edited:
These are memory clocks. And they are probably just rebrands of Tonga/Bonaire. There was probably a mix-up with these rebrands and the actual Polaris data. Expect 36/16CUs @ ~1.2GHz.
Last edited:
- Status
