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
That's a possibility for those who'd want to tear their iMacs apart to swap GPU.
I'm talking more about which GPU Apple may use in the refreshed iMac. 2007-present iMacs use MXM graphics cards.
Last edited:
Now this would make a more welcome sight if it was x2 in the MacPro7,1. MXM form factor, low TDP, could be good.
Not as good as having regular PCIe components, but at least some effort towards standardisation would be nice;
Simpler for Apple to develop for.
Simpler for us to upgrade or replace.
I have not seen this before. 896 GCN core part has 50W TDP and 1.4 GHz turbo clock O_O.
Nothing, what we have seen in desktop market would indicate that the GPUs are so efficient. What has changed? MXM design and efficient boards? Low voltages, not overdone, like they are on desktop? Better ASIC's than they are on desktop?
Nothing, what we have seen in desktop market would indicate that the GPUs are so efficient. What has changed? MXM design and efficient boards? Low voltages, not overdone, like they are on desktop? Better ASIC's than they are on desktop?
https://forum.beyond3d.com/threads/...mors-and-discussion.59649/page-4#post-1945370
Quite interesting discussion about the architecture and possibilities of what is Graphics IPv9.
Quite interesting discussion about the architecture and possibilities of what is Graphics IPv9.
Now, these could be in the nMP:
http://www.tomshardware.com/news/amd-polaris-embedded-e9550-e9260,32758.html
The e9550 of course.
Sorry, beat me to it.
[doublepost=1475005355][/doublepost]Still on Polaris, but a different one
http://www.tomshardware.com/news/samsung-ssd-pop-m-2-pro,32754.html
these would be sweet on the nMP
http://www.tomshardware.com/news/amd-polaris-embedded-e9550-e9260,32758.html
The e9550 of course.
Sorry, beat me to it.
[doublepost=1475005355][/doublepost]Still on Polaris, but a different one
http://www.tomshardware.com/news/samsung-ssd-pop-m-2-pro,32754.html
these would be sweet on the nMP
Volta being announced at GTC 2017? Makes sense.
First implementation seems to be in Xavier Tegra SOC soon:
http://wccftech.com/nvidia-xavier-soc-tegra-volta-gpu-announced/
8 ARM cores and 512 Volta CUDA cores.
It was nice of NVidia to name the chip after me (I'm also Xavier) Or not...
And:
http://wccftech.com/nvidia-pascal-volta-gpu-leaked-2017-2018/
First implementation seems to be in Xavier Tegra SOC soon:
http://wccftech.com/nvidia-xavier-soc-tegra-volta-gpu-announced/
8 ARM cores and 512 Volta CUDA cores.
It was nice of NVidia to name the chip after me (I'm also Xavier)
Or not...And:
http://wccftech.com/nvidia-pascal-volta-gpu-leaked-2017-2018/
Last edited:
Not necessary, my HIS IceQ R9 280 has a big blower, but it's a very quite card. You can google the review of this card, it's actually quieter than quite a few open fan type R9 280.
Probably downclocked to the point where it doesn't have any impact on thermals. Apple's always favored longevity / reliability over performance when it comes to GPUs.
They are not down clocked. What is more, Polaris 11 GPU, the smaller one has 1.4 GHz core clock - 200 MHz higher than desktop version, in 40W smaller thermal envelope.
The bigger one has pretty standard, reference, desktop clocks in 60W smaller thermal envelope.
Only thing that can come to mind is that they have finest silicon ASICS reserved for this, alongside with not overdone voltages on the GPU cores.
Small Vega - 3072 GCN core design with 6(!) Shader engines, 8 CU's in each shader engine and 48 ROPs, total.
Big Vega - 4096, 8 Shader engines, 8 CU's in each shader engine, and 64 ROPs, total.
Current design of Fiji has 4 shader engines, so that would tell you quite a bit about how the GPU might perform. Heavily increased power gating technology efficiency. Its easier to power down unused whole shader engine that is made from 8 CU's, than from 16 CU design. Thats why Fiji was so unbalanced in terms of performance. This will be possible only with next generation schedulers on the GPU.
Do not ask where did I got this.
Big Vega - 4096, 8 Shader engines, 8 CU's in each shader engine, and 64 ROPs, total.
Current design of Fiji has 4 shader engines, so that would tell you quite a bit about how the GPU might perform. Heavily increased power gating technology efficiency. Its easier to power down unused whole shader engine that is made from 8 CU's, than from 16 CU design. Thats why Fiji was so unbalanced in terms of performance. This will be possible only with next generation schedulers on the GPU.
Do not ask where did I got this.
Lowering the clock is not the only way to throttle down a CPU/GPU, and actually isn't the most desirable, a more effective way to save WATTS is introducing hold states or "freezing" the card every N clock cycles as the thermals require, so those mxm Polaris 10 may perform at full "clock" briefly and at downgraded "clock" when thermal require.
Wait states is not the same as throttle the clock but is 8n fact the same thing, the only restriction is it can not be used to cool an heavy overclocked system since the damage due silicon fatigue isn't mitigated this way .
And this technique is very good to trick benchmarks...
Wait states is not the same as throttle the clock but is 8n fact the same thing, the only restriction is it can not be used to cool an heavy overclocked system since the damage due silicon fatigue isn't mitigated this way .
And this technique is very good to trick benchmarks...
Maybe this will be in fact in nMP late 2016:
http://wccftech.com/amd-vega-10-gpu-hbm2-q4-2016-professional/
Apple might already have ES and they could get first silicon from AMD.
This would go well with the "professional" market.
http://wccftech.com/amd-vega-10-gpu-hbm2-q4-2016-professional/
Apple might already have ES and they could get first silicon from AMD.
This would go well with the "professional" market.
Aaaaand... 21st October is the day of release of Battlefield 1, there were rumors that AMD will tangle this event with their presentation of upcoming GPUs.
What is more. Nvidia on October 24th released GTX 1050 Ti. AMD still hasn't released full Polaris 11 desktop GPU
. You can guess where I am going with this . In this industry there is no coincidence.What is funnier, there is a rumor that the Radeon embedded GPUs are using TSMC process instead of GloFo, thats why they are so efficient. But that remains to be confirmed.
So what is the state of AMD RX 470's in Mac Pro 5,1's at this time? I know no boot screen, but will a system with macOS Sierra and this card be able to work at least? And accelerate OpenCL based programs like Final Cut, Resolve, Maya, etc.?
Want to know why they are so much more efficient than desktop versions?
Design is the same of the ASIC, they have exactly the same number of memory as desktop versions, clocks are the same, with exclusion of small Polaris, which has 200 MHz higher clock. What is differing them?
Porting the design from 14 nm to 16 nm would cost AMD 80kk$. The thing is: GloFo and Samsung have synched their process some time ago. Embedded GPUs are using the 14 nm Samsung process.
This is confirmation of it: it is from december of last year! http://www.reuters.com/article/us-samsung-elec-amd-chips-idUSKBN0U500P20151222
It is beyond belief what has gone in the background, but is now logical.
It also explains why 1024 GCN core Polaris 11 can have 35W TDP, and why 896 GCN core GPU with 1.4 GHz can have 50W TDP.
AMD used Polaris, and the fact that Samsung and GloFo synched their process to get rid of Wafer Supply Agreement terms!
And the best part is that reason for synching the process was Apple, and their computers. I suspect that the GPUs we will have in Macs will be built on Samsung process, not in the GloFo fabs.
http://semiaccurate.com/2013/08/26/a-third-player-emerges-apples-foundry-plans/
http://semiaccurate.com/2013/12/17/apple-samsung-intel-foundry-plans/
http://semiaccurate.com/2014/04/17/semiaccurate-right-apples-foundry-plans/
http://semiaccurate.com/2013/07/12/apple-has-their-own-fab/
http://semiaccurate.com/2014/04/17/global-foundries-samsung-sync-14nm-processes/
semiaccurate said: