Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
AMD Confirms RX 480 At $199 USD, Other APU & Polaris Announcements
- Thread starter lowendlinux
- Start date
- Sort by reaction score
Did you find any more powerful replacement fans with a connector?
The RX460 hit 81C in one loop of Heaven with the PC.
I tried several custom coolers but it's too much effort and cost to be worth it. Ended up cutting the front section of the inner enclosure to fit the whole GPU with its stock cooler. Doesn't look half bad that way.
Last edited:
Vega is on the rise:
http://www.fudzilla.com/news/graphics/41671-navi-is-7nm-late-2018-early-2019
http://www.fudzilla.com/news/graphics/41669-vega-20-gpu-comes-in-late-2017
And Iris Pro on a fall:
http://www.fudzilla.com/news/41667-intel-might-be-killing-off-its-iris-pro
After such an effort to pump up iGPUs with some muscle now just give up? And not compete with the dGPU players?
This might put rMBP in a tough spot, only lower performance parts will not help, specially if they up the resolution in the nrMBP.
http://www.fudzilla.com/news/graphics/41671-navi-is-7nm-late-2018-early-2019
http://www.fudzilla.com/news/graphics/41669-vega-20-gpu-comes-in-late-2017
And Iris Pro on a fall:
http://www.fudzilla.com/news/41667-intel-might-be-killing-off-its-iris-pro
After such an effort to pump up iGPUs with some muscle now just give up? And not compete with the dGPU players?
This might put rMBP in a tough spot, only lower performance parts will not help, specially if they up the resolution in the nrMBP.
It seems like Intel is acknowledging it hasn't been delivering its quad core iris graphics notebook parts. As far as I know broadwell with iris never shipped and skylake iris is available in only a handful of nucs.
This looks like a window amd could exploit if zen can deliver on efficiency and they can make a competitive apu. Otherwise apple could just go back to making all 15" MacBook pros with discrete graphics and the normal crappy integrated graphics.
This looks like a window amd could exploit if zen can deliver on efficiency and they can make a competitive apu. Otherwise apple could just go back to making all 15" MacBook pros with discrete graphics and the normal crappy integrated graphics.
Forza Horizon 3 PC benchmarks comparing GTX 1060 to RX 480, and testing the performance of FH3 in 4K on GTX 1080.
GTX 1060 is pretty much on par with RX 480(RX 480 has slight edge), on averages, but with increased resolution the minimums drop. Possibly because 192 bit memory bus.
Overall very, very good looking game, and very, very good performance on those GPUs.
Thanfully is available on Windows 10.
It looks like GPUs have changed their advancements from "tick and tock" advancements, to tick-tock ones.
Nvidia:
Kepler - Die Shrink and new architecture at the same time.
Maxwell - New architecture on the same node as Kepler.
Pascal consumer GPUs - the same architecture as Maxwell, on new node.
Pascal GP100 - new architecture on new node.
Volta - (appears to be) reused GP100 architecture on the same node as Pascal consumer GPUs.
AMD: Tahiti - brand new architecture, made on new node.
Tonga - new architecture, on the same node
Fiji - same architecture as Tonga, with tweaks on the same node.
Polaris - variation of the Tonga/Fiji architecture made on new process.
Vega - new architecture on the same process as Polaris.
Vega 7 nm - same architecture, ported to 7 nm process.
Navi - next generation architecture on 7 nm process.
So It appears that we are at max 2 nodes from the end of silicon. And 7 nm will not be here till 2019-2020.
Nvidia:
Kepler - Die Shrink and new architecture at the same time.
Maxwell - New architecture on the same node as Kepler.
Pascal consumer GPUs - the same architecture as Maxwell, on new node.
Pascal GP100 - new architecture on new node.
Volta - (appears to be) reused GP100 architecture on the same node as Pascal consumer GPUs.
AMD: Tahiti - brand new architecture, made on new node.
Tonga - new architecture, on the same node
Fiji - same architecture as Tonga, with tweaks on the same node.
Polaris - variation of the Tonga/Fiji architecture made on new process.
Vega - new architecture on the same process as Polaris.
Vega 7 nm - same architecture, ported to 7 nm process.
Navi - next generation architecture on 7 nm process.
So It appears that we are at max 2 nodes from the end of silicon. And 7 nm will not be here till 2019-2020.
FWIW.. Beta 12.1 introduces more OpenGL surface rendering issues for the rx480. 
With that said, I've never seen my cMP boot faster... with or without an EFI card. Granted this is an unscientific observation and more of a holy sh** moment where one says.. Did that just happen? Worth reporting in any case.
home
'09 cMP flashed to '10 efi, 32GB, rx480, booting from a JPV-SSUAX x4 1TB SSD.
work
'10 cMP, 970 GTX, booting from JPV-SSUBX 512 PICe SSD.
With that said, I've never seen my cMP boot faster... with or without an EFI card. Granted this is an unscientific observation and more of a holy sh** moment where one says.. Did that just happen? Worth reporting in any case.
home
'09 cMP flashed to '10 efi, 32GB, rx480, booting from a JPV-SSUAX x4 1TB SSD.
work
'10 cMP, 970 GTX, booting from JPV-SSUBX 512 PICe SSD.
I need to see to believe, that we will see at that timespan consumer products made on those processes.
They are not EUV and I think they would be equivalent to Intel's 10nm in practice.
http://semiaccurate.com/2016/09/26/globalfoundries-7nm-process-isnt-even-close-name/
Like I have said, I need to see to believe
I do not expect any consumer parts made on 7 nm processes before 2019.
But it seems there would be no real 7nm until 2022.http://semiaccurate.com/2016/09/26/globalfoundries-7nm-process-isnt-even-close-name/
Like I have said, I need to see to believe
I do not expect any consumer parts made on 7 nm processes before 2019.
https://www.linkedin.com/jobs/view/155161110
Same job offering that is floating for at least 12 months.
Important part: "be one of the first 10 applicants for this job" (roughly translated by me from Polish)
The following was on front page of macrumors a while back.
https://www.macrumors.com/2016/09/22/intel-mobile-roadmap-coffee-lake/
GT4e is disappearing (from H line up), but GT3e is still around (in the U line up for Coffee Lake). if Intel wants to compete with the mobile dGPUs then it makes a bit more sense to actually make a dGPU. I think Intel has gotten to the point where they don't want to allocate any more percentage of the die ( i.e., some sized transistor budget) to the GPU. That giving some of that back to x86 cores (e.g., going from 4 -> 6 cores) is a better trade off for designs that can scale up to larger system TDP ranges.
"Iris Pro" is probably being killed off as a brand. HD xyz where 'y' ( 5 or 8 ) may indicate "Iris Pro" like eDRAM and larger than average EUs. Already done with Skylake https://en.wikipedia.org/wiki/Intel_HD_and_Iris_Graphics#Skylake
If Intel stepped up to mobile dGPU with 100% of die then could still be cheaper with a Intel dGPU package with GPU and eDRAM built in and still leverage the the main RAM for bulk Graphics RAM storage.
Could be cheaper and more 2D space efficient than dGPU with VRAM/HBM.
[doublepost=1474966041][/doublepost]
what new Mac product doesn't Apple label as "revolutionary" ( "insanely great" or whatever the hyperbole adjective du jour is)? So same stuff, different day.
http://www.anandtech.com/show/10710/amd-announces-embedded-radeon-e9260-e9550
Can someone explain to me, how come Ellesmere XT can have 95W TDP with, nominal 1.266 GHz boost clocks?
Can someone explain to me, how come Ellesmere XT can have 95W TDP with, nominal 1.266 GHz boost clocks?
Probably a similar scenario to the Fury Nano, where it has official boost clocks but the chip thermal throttles much more quickly and those boost clocks are not sustained for very long or at all.
Reference model of RX480 Throttles quite a lot, if the voltages are standard. Put them lower and you get lower power consumption and higher performance because of no throttling. However, the power consumption is not 60W lower(155 vs 95W), but only up to 30W lower. So where did they got another 30W of lower power consumption?
What is however funny, even at the base clocks it makes this GPU the most efficient GPU in the world right now with 54 GFLOPs/watt.
Nano on the other hand is power gated. It will not exceed certain level of power consumption for longer periods of time. Average core clock is 880 MHz at 185W of power draw for whole GPU board.
So maybe the board design makes the 30W difference? Because embedded model is MXM.