I don't see anything in that document that suggests AMD used those service. They just use the release of Fiji as a hook to launch their own alternative.
AMD: Speculation, Rumors, and Discussion (Archive)
- Thread starter iMacmatician
- Start date
- Status
Memory types have been labeled based on core speed and data clocks, and I remember some early stories on GDDR5 that were thrown by WCK operating at 2x CK.
MT/s or GT/s would focus on the rate of data being sent over the interface, rather than the specific clock transitions that could be SDR, DDR, DDR but 2x the rate, QDR, ODR, etc.
Ok does anyone know if AMD developed their own HBM Controller?
NW Logic's controller goes back before Fiji-Fury release.
But yeah it is going to be a pig to fully prove with evidence available on the internet.
Cheers
Agreed, just that my point is its already messy because a few sites are getting this wrong with regards to reporting clocks, and compounding this further with an effective clock to show a difference.
Cheers
What does it matter if they did or not?
Regardless which, it's not going to be publically information either way.
I suppose it could make a difference if that specifically marketed HBM controller with HBM1 and HBM2 support were the one in Fiji. That might indicate some flexibility for that chip. Even if AMD didn't design its own, there is no public information either way on its feature set versus the demonstration product or specifications.
D
Deleted member 13524
Guest
From the link posted above:
Alessio1989
Regular
R9 390 (non-X) 8GB are starting falling down under 300€, which is a good price in my country.
Supposed new specs for Polaris 10/11
http://videocardz.com/59903/possible-polaris-10-and-polaris-11-specifications-emerge
Polaris 10: 32CU @ ~1350Mhz for 5.5 Tflops.
Polaris 11: 14 CU @ ~1395 for 2.5 Tflops
I think this is more believable due to the higher clock thats to the new process.
http://videocardz.com/59903/possible-polaris-10-and-polaris-11-specifications-emerge
Polaris 10: 32CU @ ~1350Mhz for 5.5 Tflops.
Polaris 11: 14 CU @ ~1395 for 2.5 Tflops
I think this is more believable due to the higher clock thats to the new process.
Love_In_Rio
Veteran
By PS4K docs GPU has 36 CUs at 911Mhz.Doesnt fit with this news.
PS: it is added in the news it could be mobile versions.
upnorthsox
Veteran
It doesn't have to, ps4k is a custom gpu anyhow and adding another 8 CUs is probably the easiest part.
Unless PS4 isn't Polaris base.
Polaris 10 GPU could have 8 CU's disabled, while the PS4K would only have 4 disabled, but would sacrifice clock instead.
D
Deleted member 13524
Guest
If true, mobile Polaris 11 doing 2.5 TFLOPs is quite a feat.
Then again, 150W for mobile Polaris 10 seems like too much, and wouldn't fit in anything but a high-end desktop replacement.
Well it could have 4 disabled CUs for the GDDR5 mobile version, and the whole 36 CUs for the desktop version with GDDR5X.
150 w for a mobile isn't bad if they are trying to remove the gtx 980m out of OEM space which uses around the same wattage but this will surely provide much better performance.
Love_In_Rio
Veteran
yeah, but the 5,5 Tflops figure doesnt have sense.Maybe they are mixing leaks and is right that the 32CUs GPU is mobile and the 5,5 Tflops and 150 watts part is in relation to the 40 CUs desktop part.
Rurouni
Veteran
I actually wouldn't mind that 32CU version for desktop if it is going to be priced as a 380x replacement. I was going to buy 380x but decided to wait for Polaris. But if Polaris ended up only either as a mobile or a higher end desktop parts, then unfortunately I might have to stick to my initial plan on buying 380x. Should be cheaper by the time Polaris launched.
gamervivek
Regular
New AIDA64 adds Vega 10 aka Greenland.
gamervivek
Regular
1.4Ghz for mobile parts sounds too good to be true, desktop parts would be clocking 150-200Mhz above that from what little I've seen of mobile chips vs. their desktop counterparts and it's pretty unlikely that AMD ends up that close to nvidia's new series.
- Status
