FOAK: Questions for memory/BIOS experts (long and technical...)
- From: "Dr Ivan D. Reid" <Ivan.Reid@xxxxxxxxxxxx>
- Date: Thu, 9 Jul 2009 19:40:29 +0000 (UTC)
I bought an Acer Aspire Revo "nettop" last week; a 1.6 GHz Atom
processor with an Nvidia ION peripheral chip (including a CUDA-capable
GPU), 1 GB of RAM, 160 GB hard disk, in a case smaller than some of the
Harry Potter novels. Not bad for £150 (mini keyboard and dinky Logitech
mouse included).
I couldn't get XP to load, tho' I had Fedora 11 running for a
night. In the end I put on Windows 7 RC 64-bit (unlike the one in my
laptot, this Atom supports 64-bit), whose minimum requirement (10 GB)
meant I had to wipe the disk and start anew.
I set the video graphics to 256 MB, leaving 768 MB for the OS,
marginal but usable. With that much VRAM it can run seti@home CUDA
software, so it's been crunching two CPU hyperthreads and a GPU process.
Overall the GPU (I believe it has 16 cores) is giving throughput of the
order of 20% greater than the combined hyperthreads.
I want to test it on some CUDA software I'm benchmarking,
reconstructing digital holograms. To do that I need more VRAM -- a
4096x4096 complex FFT requires more than 512 MB so I can't do an image
that big on my home machine (a 2Kx2K sub-image runs in 512 MB).
Acer's documentation says that the Revo can run up to 4 GB of RAM,
and up to 896 MB of VRAM. Since the one comes from the other this means
that at 896 MB VRAM you'll only have 3200 MB of OS RAM.
So I bought 4 GB (2x2GB PC2-6400 SODIMMS) and swapped them in for
the one 1 GB SODIMM delivered as stock. When I booted up I noticed that
the BIOS reported only 3072 MB of RAM. In Setup I was now given the
choice of 512 MB of VRAM; this pushed the reported RAM down to 2816 MB --
and it's stayed there despite changing VRAM allocation all the way down to
16 MB. (The seti application confirms that it's seeing 512 MB of VRAM.)
Thinking this might be a BIOS limitation, I put in a help request
to Acer. They came back saying that I can add hardware to the machine
without voiding my warranty but they will give no support for any added
hardware.
So I'm on my own with this, but it suggests that maybe it's not a
BIOS limitation (I have the "Linux" version of the BIOS). Unfortunately I
have no modern experience with memory chips, but from what I did years ago
I find it hard to believe a device fault could cut 2x2 GB dual-channel
down to 3.25 GB -- or even 3.0 -- as the addressing doesn't work like
that. Does anyone know if this is likely? Is it normal for a BIOS to
limit the amount of RAM it reports (I thought 32-bit Windows had a 3 or
3.5 GB limit; MSDN suggests it's 4 GB)?
Tomorrow I'll try each memory card separately to verify that they
each give 2 GB single-channel. If not, RMA time. If so, try again together
(the second, previously unused, socket was a bit stiff, but I thought I
got the memory well seated).
Memtest on the FC11 live SDHC card, to be complete, only tests
2816 MB of RAM (with VRAM set to 512). CPU-Z reports 4 GB but I suspect
it just goes on the data it gets from the chips, not an actual memory test.
Constructive comments welcomed.
(Had a thought while writing this; would it be worth clearing the
CMOS RAM? Perhaps something is being memorised that I'd rather be reset.)
--
Ivan Reid, School of Engineering & Design, _____________ CMS Collaboration,
Brunel University. Ivan.Reid@[brunel.ac.uk|cern.ch] Room 40-1-B12, CERN
KotPT -- "for stupidity above and beyond the call of duty".
.
- Follow-Ups:
- Re: FOAK: Questions for memory/BIOS experts (long and technical...)
- From: Andy Burns
- Re: FOAK: Questions for memory/BIOS experts (long and technical...)
- Prev by Date: Re: Groneworthy business name
- Next by Date: Re: FOAK: Questions for memory/BIOS experts (long and technical...)
- Previous by thread: I have nothing to say to anyone on USENET
- Next by thread: Re: FOAK: Questions for memory/BIOS experts (long and technical...)
- Index(es):
Relevant Pages
|