Re: PC requirements

Posted by Craig Brideau on
URL: http://confocal-microscopy-list.275.s1.nabble.com/PC-requirements-tp7587774p7587782.html

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Should have clarified: PCIe is only part of the attachment process. The
motherboard must also support the NVMe standard via PCIe to achieve maximum
performance. Most motherboards that support the latest NVMe protocol have a
dedicated M.2 slot (or two) specifically for mounting M.2 NVMe compatible
drives. (Some also are backwards compatible with SATA, this is slot
dependent!) You *can* attach such drives via PCIe adapter cards but this
depends on if the motherboard supports NVMe through the standard PCIe
slots. The drives will still work in a PCIe slot, but you really only get
the full potential out of them with the NVMe protocol implemented on the
motherboard.

Craig

On Mon, Jan 15, 2018 at 11:04 AM, Michael Giacomelli <[hidden email]> wrote:

> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> Post images on http://www.imgur.com and include the link in your posting.
> *****
>
> Hi Craig,
>
> You can buy M.2. NVME to PCIe adapters for a few dollars on Amazon.
> They're pure passive (they just route the PCIe pins to the M.2. connector)
> and work well for adding M.2. to motherboards that don't support it.  You
> lose an x4 PCIe slot, but usually there is at least one extra on a full
> size motherboard.
>
> Otherwise we usually just use Intel's single socket HEDT platform
> (currently X299-based) which supports up to 18 cores and 8 DIMMs (128GB of
> RAM).  These typically come with 2x M.2. slots for NVME storage.  Dual
> socket would allow more than 20 cores, but the value is lower due to the
> cost of the Xeon platform.
>
> Mike
>
> On Mon, Jan 15, 2018 at 12:02 PM, Craig Brideau <[hidden email]>
> wrote:
>
> > *****
> > To join, leave or search the confocal microscopy listserv, go to:
> > http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> > Post images on http://www.imgur.com and include the link in your
> posting.
> > *****
> >
> > Comment on the M.2 drives: They are significantly faster than SATA SSDs,
> > but for full speed the motherboard must support NVMe which is a dedicated
> > protocol for M.2 drives featuring a dedicated slot on the motherboard.
> > Annoyingly, most dual-CPU-socket 8-DIMM slot motherboards don't feature
> > this, so it is not an available option if you also want >64GB of RAM. If
> > anyone happens to come across an NVMe board that also has 8-DIMM slots
> > please let me know!
> > I use an M.2 NVMe for a system at home and it boots absurdly fast with
> only
> > an older i5 core.
> >
> > Craig
> >
> > On Mon, Jan 15, 2018 at 8:06 AM, Menelaos Symeonides <[hidden email]>
> > wrote:
> >
> > > *****
> > > To join, leave or search the confocal microscopy listserv, go to:
> > > http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> > > Post images on http://www.imgur.com and include the link in your
> > posting.
> > > *****
> > >
> > > Hi Peter,
> > >
> > > If you have ever built your own PC, or know someone who has, you can
> > > benefit greatly from building it custom. Off-the-shelf systems can be
> > > upmarked considerably and often use outdated components or skimp on
> > > important things like a good power supply. Of course, the benefit of
> that
> > > would be presumably a warranty on the whole system, in case you are not
> > > able to diagnose things yourself, but if you build a custom system,
> each
> > > component will come with its own warranty. It would just be up to you
> to
> > > determine which component is faulty and claim the warranty on it
> > yourself.
> > >
> > > I recently built such a computer for light sheet data acquisition and
> > > processing/analysis. Here is the configuration:
> > >
> > > Motherboard: Asus Z10PE-D16 WS
> > > CPU: Intel Xeon E5-2650 V4 (x2)
> > > CPU cooler: Noctua NH-U12DXi4 (x2)
> > > Memory: Samsung M393A2G40EB1-CRC 16GB DDR4 2400MHz ECC-Reg (x8 = 128 GB
> > > total)
> > > OS drive: Samsung 960 EVO 500GB M.2-2280 SSD (on a PCIE M.2 adapter
> > > because the motherboard runs its onboard M.2 at SATA speed)
> > > Data drive: Samsung 850 EVO 1TB SATA (6 drives in RAID10 configuration)
> > > Video card: GeForce GTX 1080 Ti 11GB
> > > Case: Phanteks Enthoo Primo
> > > Power supply: EVGA SuperNOVA P2 1200W 80+ Platinum
> > >
> > > You will obviously also need monitors, keyboard, mouse, Windows
> license,
> > > and some kind of data server - you should assume that you will not be
> > > storing your data on this computer if you are generating 1TB datasets.
> I
> > > got a Synology DS1817+ 8-bay NAS with a 10GbE adapter and loaded it
> with
> > 8
> > > Western Digital Gold 10TB drives (WD101KRYZ), and run them in SHR2
> > > (basically RAID6, for double-parity, i.e. you can lose two drives and
> > still
> > > be able to recover). I also got a 10GbE adapter for the workstation PC,
> > > this allows utilization of the maximum transfer rates the RAID array is
> > > capable of, which is 3-4 times faster than standard Gigabit Ethernet.
> > This
> > > data server is costly but, for light sheet data, basically necessary if
> > > your institute does not provide substantial data storage.
> > >
> > > Another thing to note is that component prices can fluctuate quite a
> bit.
> > > I bought the video card in the summer for under $740. Right now they
> are
> > > out of stock everywhere (1080 Ti of any brand), and when they are in
> > stock,
> > > they go for anywhere up to $2,000 (you have bitcoin mining to thank for
> > > that). You can get a Titan Xp for less than that now so the
> > > price/perfomance advantage of the 1080 Ti is no longer there, but look
> > for
> > > the Titans to also disappear off the market really quickly as mining
> > > expands higher up the range.
> > >
> > > All told, I spent around $9,000 for the workstation and $4,500 for the
> > > data server. Comparable off-the-shelf workstations cost around $12,000
> -
> > > $15,000 (and possibly even more now with the GPU market up in the air
> and
> > > RAM prices currently rising), and off-the-shelf data servers of similar
> > > capacity would cost much more than that.
> > >
> > > Assembly and setup of the PC was not trivial, so you will want someone
> > > with experience to help you. Workstation (Xeon) motherboards can be
> very
> > > finicky, and the installation manuals for some components can be pretty
> > > incomprehensible. Also be aware that if you need multiple PCIE slots
> > > available for data acquisition cards etc., the GPU only takes one but
> > > actually occludes at least one more slot (my card took up three slots
> > > total) so your choice of case/card should be made carefully. I ended up
> > > cutting a hole in the back panel of the case and mounting the GPU
> > > vertically with a 2-slot PCI bracket and PCIE riser cable, as I needed
> > > every last PCIE slot on the motherboard. Some cases come with vertical
> > GPU
> > > mounting positions, but some of those will block the regular PCIE
> slots,
> > or
> > > will be incompatible with SSI-EEB format motherboards (which is what
> most
> > > dual CPU boards will be).
> > >
> > > Good luck!
> > >
> > >
> > > Mel
> > >
> > >
> > > On 1/15/2018 7:22 AM, Owens, Peter wrote:
> > >
> > >> *****
> > >> To join, leave or search the confocal microscopy listserv, go to:
> > >> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> > >> Post images on http://www.imgur.com and include the link in your
> > posting.
> > >> *****
> > >>
> > >> Dear listers,
> > >>
> > >> I am looking into purchasing a high end image processing PC , that
> will
> > >> be capable of processing large multidimensional data sets up to 1 TB
> in
> > >> size.
> > >> Does anyone have any recommendations on a PC configuration that would
> be
> > >> suitable?
> > >> Do people build custom PCs or buy off the shelf?
> > >> Are high spec gaming PCs up to this task?
> > >>
> > >> thanks for any advice on this .
> > >>
> > >> all the best
> > >>
> > >> Peter
> > >>
> > >>
> > >> Peter Owens
> > >> Centre for Microscopy and Imaging,
> > >> National University of Ireland Galway.
> > >> P: +35391494036 m: +353863326749
> > >> W: www.imaging.nuigalway.ie e: [hidden email]
> > >>
> > >>
> > >>
> > >>
> > >> --
> > > Menelaos Symeonides
> > > Post-Doctoral Associate, Thali Lab
> > > Department of Microbiology and Molecular Genetics
> > > University of Vermont
> > > 318 Stafford Hall
> > > 95 Carrigan Dr
> > > Burlington, VT 05405
> > > [hidden email]
> > > Phone: 802-656-1161
> > >
> >
>