pabloAlonso :
Hi guys,
I have been working in music production and software development on my Lenovo L450 laptop for two years (I'm not interested in gaming or other purposes). Since I'm beginning in machine learning and some decent hardware is needed for this task I'm starting to design a build with a budget of 1200€. I've naively chose some components, mostly inspired on the recommendations from this page and some audio and ML forums, but the whole thing is more expensive than expected.
Here is the list of components with their price in the Spanish shops where I am planning to buy (I couldn't find significant lower prices over the main European shops) :
Components I wouldn't like to change:
CPU: Intel Core i7 6700K 339€
GPU: EVGA GeForce GTX 1060 SC 6GB 279€ (recommended for beginners in training ML models)
Components I don't mind changing:
BOX: Zalman z9 NEO 65€ (I would like something with good noise isolation)
SSD: Toshiba OCZ TR150 480GB 119€
HDD: WD Blue 1TB 50€
Motherboard: ASUS Z170-K 135€
PSU: EVGA 650 GQ 650W 85€
RAM: 2x Avexir Budget DDR4 2400 PC4-19200 16GB 176€
Monitor: LG 25UM58-P 25 161€
Approx Total: 1410€
Maybe I'm expending too much money in components that doesn't improve very much the overall performance so I would appreciate very much any advise on cheaper components that doesn't compromise the behavior or places where I can find them for a lower price. Thank you!
pabloAlonso,
Optimizing a system for these uses requires associating the characteristics of the software with the hardware. These hardware list at the moment is more gaming-oriented, with high clock speeds and very high single-thread performance, whereas your uses may benefit from more CPU cores and GPU computing.
Music Production: My question is as to whether the performance emphasis is going in the right direction. For mucsic productiion, if you're using a higher end production suite ca[able of a high levels of multi-tracking with a lot of plug ins Sonar Producer fro example, the core count is more imprtant thatn clock speed, as is a lot of RAM memory and fast disk.
Development: For development, running multiple VM.s, likewise, the ability to dedicate a core to each simultaneous VM is an advantage.
Machine Learning: Machine learning today is heavily GPU-compute oriented and the system may greatly benefit from GPU- compute modules. In this use, the compute cycles per second is important, but depends on the way the programs /algorithms are written- whether it is multi-threaded. For analysis and simulation, having similar requirements to run multi-threaded custom algorithms (Matlab and Wolfram) I use a Quadro K2200(4GB) + Tesla M2090 (6GB GPU coprocessor and that combination costing $350 + $86 (the M2090 cost $2,500 new), performs similarly to a Quadro M5000- a $1,900 GPU. The key is having plenty of CUDA cores so a GTX will work also.
In general, I'd suggest, at least a 6-core LGA2011-3 CPU, possibly a Xeon E5-1650 v4, on an X99 motherboard, an M.2 NVMe SSD followed by a Samsung 850 Evo SSD for active projects, a GTX 1060 ,1070 or Quadro M2000, and a full tower case rated for quiet running.
I tried searching a couple of online stores in Barcelona and Madrid but have not found one with a very good selection of components.
Without knowing the nature of the programs used and project size, it's not entirely useful to suggest an exact system, but I'd mention my approach in the last six years has been to buy used LGA2011 workstations- which can use 8-core CPU's and upgrade according to the use. Workstations are designed to have high performance and reliability exactly for this work and they are also made to be very quiet. Not everyone accepts the idea of a used system, but since 2009 I've had Dell PRecision 390, T3500, T5400, T5500 and HP 420, HP z420 NO.2, and HP z620 with 100% rebiliabity. And performance is performance is performance, the z620 has a Passmark CPU rating of 22625 whereas the top CPU score for i7-7700K is 12507. Yes, that has a single-thread mark of 2640 and if all your programs are single-threaded, that will be better, but certainly music production doesn't need it-and I think only 4-cores will be a disadvantage.
I you'd like to list the software you're using, whether used hardware is possible- this might be the CPU only, and the online shops at which the parts will be purchased, specific recommendations are possible.
An example of upgrading a used workstation:
Purchased for $270 US:
HP z620 (Original) Xeon E5-1620 4-core @ 3.6 /3.8GHz) / 8GB (1X 8GB DDR3-1333) / AMD Firepro V5900 (2GB) / Seagate Barracuda 750GB + Samsung 500GB + WD 500GB
[ Passmark System Rating= 2408 / CPU= 8361 / 2D= 846 / 3D = 1613 / Mem =1584 / Disk = 574 ] 7.13.16
Upgraded, cost about $1,200US:
Analysis / Simulation / Rendering:
HP z620 (2012) (Rev 3) 2X Xeon E5-2690 (8-core @ 2.9 / 3.8GHz) / 64GB DDR3-1600 ECC reg) / Quadro K2200 (4GB) + Tesla M2090 (6GB) / HP Z Turbo Drive (256GB) + Samsung 850 Evo 250GB + Seagate Constellation ES.3 (1TB) / Creative Sound Blaster X-Fi Titanium PCIe sound card / 800W / Windows 7 Professional 64-bit > Logitech z313 2.1 speakers > HP 2711x (27" 1980 X 1080)
[ Passmark System Rating= 5675 / CPU= 22625 / 2D= 815 / 3D = 3580 / Mem = 2522 / Disk = 12640 ] 9.25.16
[ Cinebench R15: OpenGL= 119.23 fps / CPU = 2209 cb / Single core 130 cb / MP Ratio 16.84x] 10.31.16
The cost to performance ration is impossible to better with this method as the CPU's are depreciated, most cost may be redirected to GPU's. the MTBF (average time before failure) of the Xeon e5-2690 is 170,000 hours, or 19.4 years' continuous running. As that CPU was released in 2012, it has an expected use of more than 15 more years.
Cheers,
BambiBoom