Linux-Fan wrote: > I am constantly needing more computation power, RAM and HDD storage such > that I have finally decided to buy a server for my next "workstation". The > reasoning is that my experience with "real" servers is that they are most > reliable, very helpful in indicating errors (dedicated LEDs next to the > PCIe slots for instance) and modern servers' noise seems to be acceptable > for my working envirnoment (?) >
they are loud - for me it is unacceptable to have it in some living space > I currently use a Fujitsu RX 1330 M1 (1U server, very silent) and it > clearly is not "enough" in terms of RAM and HDD capacity. A little more > graphics processing power than a low-profile GPU would be nice, too :) > Servers usually do not have a good GPUs - they are mostly not used. Why don't you split your setup into powerful workstation and your current server? > Rack-Mountability is a must, although I am open to putting another tower > in there, sideways, should that be advantageous. > > In terms of "performance" specifications, I am thinking of the following: > > * 1x16-core CPU (e.g. AMD EPYC 7302) > > * 64 GiB RAM (e.g. 2x32 GiB or 4x16 GiB) > > I plan to extend this to 128 GiB as soon as the need arises. > As I am exceeding the 4T mark, I am increasingly considering the > use of ZFS. > > Currently, I have the maximum of 32 GiB installed in the > RX 1330 M1 and while it is often enough, there are times where I > am using 40 GiB SSD swap to overcome the limits. > > * 2x2T HDD for slow storage (local Debian Mirror, working data), > 2x4T SSD for fast storage (VMs, OS) > I will do software-RAID1 (ZFS or mdadm is still undecided). > I possible, I would like to use the power of the modern NVMe PCIe > U.2 (U.3?) SSDs, because they really seem to be much faster and that > may speed-up the parallel use of VMs and be more future-proof. > > * 1-2x 10G N-BaseT Ethernet for connecting to other machines to share > virtual machine storage (I am doing this already and it works...) > > * a 150W GPU if possible (75W full-sized card would be OK, too). > > Typical workloads: > data compression (Debian live build, xz), > virtual machines (software installation, updates) > you could off load your virtual machines to dedicated hardware. Depends on the number and configuration you indeed may need a server for that - better look at the requirements of the VM software manufacturer (VMWare or Oracle or whatever else) > Rarely: > GPGPU (e.g. nVidia CUDA, but some experimentation with OpenCL, too) > single-core load coupled with very high RAM use (cbmc) > > Some time ago, there was this thread > https://lists.debian.org/debian-user/2020/06/msg01117.html > It already gave me some ideas... > > I am considering one the following models which are AMD EPYC based (I > think AMDs provide good performance for my types of use). > > * HPE DL385 G10 Plus > * Dell PowerEdge R7515 > > I have an old HP DL380 G4 in the rack and while it is incredibly loud, it > is also very reliable. Of course, it is rarely online for its excessive > loudness and power draw, but I derive that HPE is going be reliable? > Before the Fujitsu, I used a HP Z400 workstation and before that a HP > Compaq d530 CMT and all of these still "function", despite being too slow > for today's loads. > Servers are loud - same for DL380 G10, same for the PowerEdge > I am also taking into consideration these, although they are Intel-based > and I find it a lot harder to obtain information on prices, compatibility > etc. for these manufacturers: > > * Fujitsu PRIMERGY RX2540 M5 > * Oracle X8-2L > (seems to be too loud for my taste. especially compared to the others?) > > I have already learned from my local vendor that HPE does not support the > use of non-HPE HDDs in the server which means I would need to buy all my > drives directly from HPE (of course this will be very expensive). > Additionally, none of the server manufacturers list Debian compatibility, > thus my questions are as follows: > > * Does anybody run Debian stable (10) on any of these servers? > Does it work well? > > * Is there any experience with "unsupported" HDD configurations i.e. > disks not bought from the server manufacturer? > HPE does not support means not that you can not do it - rather you will not have warranty/support in case of troubles. If you can live with it? > I would think that during the warranty period (3y) I best stay with > the manufacturer-provided HDDs but after that, it would be nice to be > able to add some more "cheap" storage... > They use Seagate or Samsung disks - I am not sure ATM. Disks are disks you can put whatever you want there. The disks they sell seem to be manufactured specifically for HPE and likely are tested/certified by HPE, although here one disk failed after 1,5y of operations in a conditioned server room where the machine did not have much disk load. I mena it is good to stay for the 3y with their hardware. > * Of course, if there are any other comments, I am happy to hear them, > too. > > I am looking into all options although a fully self-built system is > probably too much. I once tried to (only) get a decent PC case and failed > at it... I can only imagine it being worse for rackmount PC cases and > creating a complete system composed of individual parts? Not at all - you can buy case and all the parts you need to put inside. A friend did it for me 8y ago - really excellent machine. 12 HDD bays 800Watt PU. The only problem was the hight 2RU and the hight of the CPU cooler fan (flat one). I simply cut the cover and build a normal hight CPU fan after the first one died. If you have the rack in or near a room you live, I would split the setup in several servers - perhaps also buy 2 switches and fiber at least to the storage. But remember that noise gets multiplied (not added), so it could be that few machines are as load as one server. It might be better to find an isolated place for the rack. These are just my thoughts Unbelievable what people do at home :) good luck