Time to update my 7-year old PC as it, and Centos 5.6, are really old :-).My budget is several hundred $, and I'm planning to get a white-box PC and install Centos 7 on it. The machine will be accessed 100% remotely (through VNC) by various other local and remote PCs. I'm doing development with mid-size arria-10, and my builds and sims can run an hour or more on my current machine, so speed and good I/O are paramount. It looks like I can afford an Intel QuadCore i5-7500 (about 2x Passmark of my current machine), along with 32GB of memory (DDR4?) I get that 64GB would be better, but I think 32GB will be ok for now, and (politically) it's better to stay in budget now and replace the memory later (it will be cheaper then as well). Any suggestions around best hardware, MOBO, and best O/S would be appreciated though. Is Centos a bad choice? Would I get a big win with an equally-priced i7 processor? Thanks in advance /j
Hello Jeffw_00In my case (Dell Precision Xeon, 16GB, SSD), processor power is the bottleneck, not the size of the RAM or the speed of the SSD. I choose i7 over i5 instead of 32 gb over 16gb. But i compile Cyclone IV/V designs that run for about 15 minutes. My best friend has an i5 (HP), it takes about 30% more time to compile the same design. Best regards, Johi.
its got Recommended Physical RAM for each device, so for Arria 10 18-48 GB. You might not need to be so worried about getting 64gb or you might be able to squeese an extra 16 out of your budget etc.
I would recommend the CPU with the highest possible turbo frequency, like i7-7700K or i7-8700KI've run some tests (mostly with Vivado, but also with Quartus). One machine with Xeon firstname.lastname@example.orgGHz (Tubro 4GHz), 32GByte RAM, NVME SSD and another Threadripper 1950X@3.4GHz (Turbo 4GHz), 64GByte RAM. One of examples AWS F1 projects takes 2h 25min on Xeon and 2h 15min on Threadripper. Anothers takes 51min on Xeon and 58 min on Threadripper. 10% of time all of the cores/threads were occupied (8 Threads on Xeon 1275v5, 32 Threads on Threadripper), the other 90% of time only one core was running at 100% (Turbo).
--- Quote Start --- 10% of time all of the cores/threads were occupied (8 Threads on Xeon 1275v5, 32 Threads on Threadripper), the other 90% of time only one core was running at 100% (Turbo). --- Quote End --- What wersion of Quartus are you using? Is it Lite, Standart or PRO?
--- Quote Start --- What wersion of Quartus are you using? Is it Lite, Standart or PRO? --- Quote End --- For this test I used Xilinx Vivado 2017.1, not Quartus. But Quartus (Pro/Standard) with other tests behaves very likeliy like Xilinx. As I see, synthesis and report/timing analysis can use multicore. While fitting/placement mostly one thread/core is used. Its sad, but its true. My rule of thumb for FPGA development is "think, simulate, think, pause, think, simulate, think while synthesis" ;-)
Rough rules of Thumb in order of things that cause the biggest spead-up + slowdown1. RAM - dont have enough and your compile will slow to a crawl, if it even works at all. Ensure you have enough that virtual ram is not needed. 2. Processors - Faster processors are better, more cores are only good to a certain level. Most of the compile will probably be done on only 1 core, so the faster the better. Ensure you have actual cores, not virtual or hyperthreaded. 3. SSD might speed up a little, though this is unlikely to make a difference. Having done compiles on the same PC on a local disk and network disk, it didnt really make a difference, unless the network was heavily loaded.
Also interesting point to note (and this was a while ago) : Mentor admitted that the 32 bit version of Modelsim ran faster than the 64 bit version, but this may have been fixed now.