[CALUG] Looking for advice
b.j.smith at ieee.org
Wed Feb 13 19:41:58 EST 2019
Open Source Rocks <opn.src.rocks at gmail.com wrote:
> Fantastic discussion - i learnt a few new things. I have always played it
> safe since I switched over to linux back in 2001 for my main desktop.
> Always went with Intel CPUs. I am not gamer, so did not need Nvidia or any
> external video cards.
I long tended to prefer AMD for double-precision FPU, coming from an
engineering background and early CFD clusters in the '90s.**
The funny thing is that AMD's relationship with TSMC and other fabs means
it's going to beat Intel in feature size this year, even going to 7nm while
Intel is still trying to get to 10nm. Add in the Zen2 improvements, and
it's going to be extremely difficult for Intel to complete.
E.g., with the Zen ('17) and Zen+ ('18) at 14nm versus Intel at the same
fabrication, even in single threaded applications, Intel had - - at most -
- an 8% lead. In heavily threaded applications, it was no contest, AMD Zen
smacked Intel's latest i-Core, forcing Intel to quickly' re-brand their
next generation, and even pull in their Xeon high core count (at a loss of
speed) as the i9 consumer/workstation product.
It will be interesting to see where Zen2 ends up, and how the 7nm yields
are coming out of TSMC, but right now it's not looking good for Intel this
year in the consumer/workstation space. And then we have a whole, new slew
of 32 and 64-core ARM processors, with massive cache sizes, on the Server
end, challenging Xeon too - - matching some of the top performance, at far
less power requirements.
Now a days, my son plays on the computer and I have Nvdia card. In the last
> 7 to 8 years, I have not had an issue with Nvidia drivers. The OS prompts
> me if I want to use Nvidia drivers or Open source drivers and I go on.
AMD's own proprietary builds are really going away. Their AMDVLK code is
open source, and the Mesa RadeonSI and RADV developments are extremely
competitive... Even against nVidia's proprietary drivers.
That's why I'm seriously considering an all-AMD solution on my next
self-assemble SFF 'cube,' let alone a notebook. And AMD's SoC really
brings pricing of the board/system down in integration too, something Intel
is only getting around to doing.
For those that don't game, I'd love to see some inexpensive Epyc BGA
Nano-ITX and Pico-ITX options in a tiny little box, like they did with the
prior Jaguar/Kabini units 2011-2013 (that are also at the cores in the XBox
and PS4 as well).
I have noticed on one of my laptops (toshiba - 4k touch screen) it has some
> kind of a hybrid nvidia/intel graphics card or chip (intel 530, nvdia
> gm107m/gtx 950m). The OS (deepin linux) prompted me multiple options (a)
> intel default (poor compatibility), (b) open source driver (poor
> performance), (c) Bumblebee solution (save power to reduce power
> consumption) & (d) NV-Prime Solution (for laptops with hybrid graphics).
Optimus is a real PITA, with either driver set. I disable it on my Dell
Precision m4000 series notebooks, and just use nVidia all-the-time. I'm
more hopeful with the AMD equivalent in an all AMD stack when I go that
I chose (c) and it has been working well, at least with the steam games my
> son and I play.
A note of OS - I have gone down a different path from Ubuntu or Fedora or
> even Linux Mint - I switched over to Deepin Linux. It has worked with all
> kinds of hardware I have thrown at it - right from Core 2 Duo to i7 4th
> gen. Sadly though, they no longer support 32-bit hardware, so my core 2 duo
> is stuck on an older version of Deepin Linux.
It's just difficult to support i686 at this point. Lots of extra work.
With uEFI (always 64-bit in a PC, sans early Apple's) taking over, it'll be
even less and less. Intel has basically pushed all mainboards to drop all
CSB (16-bit BIOS Service compability) starting next year.
Good discussion, thanks for sharing your thoughts. I will try to make it in
**P. S. Don't get me started, especially since the whole idSoftware/Doom
episode made people [incorrectly] believe Intel had a better FPU, no, it
didn't, the Pentium just had a very defective ALU, long story.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the CALUG