51–60 of 71 posts
Btw. I love my desktop PC and use it more than my Laptop. But I love it, because it has a passive cooling system (0 db) on an i7 CPU ;-)
Earlier quoted context omitted.
> I suspect your main motivator is to want to build a PC. Chances are that the extra computational power or cores won't actually make your work faster for day to day use. Another potential factor - If you play games or develop 3d applications, high end GPUs are much more affordable in desktop form factors than in laptop ones and generally have more horsepower than their laptop equivalents. If you do a lot of video en…
This may not be relevant to the OP, but I'm interested in machine learning, and having a graphics card with a big GPU is important. Serious machine learning computers have sky high prices, but for learning, you can get a graphics card with 640 or more gpu cores for $250 or less. I don't know of any laptops in the requested price range that have that kind of gpu power. Of course, a big graphics card can also be used f…
https://en.wikipedia.org/wiki/General-purpose_computing_on_g...
If you're interested in any of these and don't need the mobility for another reason, a desktop may be a very good consideration indeed.
https://pcpartpicker.com/user/nalply/saved/pXrD3C
This PC part list starts with about $2900 and only one GPU. Additional GPUs will cost about $1200 (unsure), so if it is maxed out with four GPUs we talk about $6500 and 1250 Watt.
I will wait till I understand deep learning well enough before investing this much money and power.
Why?
Machine learning - specifically, I think we're likely to see more and more applications using GPUs for ML-based tasks. For instance, maybe Photoshop implements that relatively new Google super-resolution ML-based technique as a filter or something in their software. While it could run ok on a multi-core CPU - it would probably run much better and faster on a GPU.
I can see this becoming more of the norm on a lot of software as machine learning continues going "mainstream". Sooner than we think, we might all find ourselves purchasing high-end graphics cards for things other than graphics. At least until specialized ML optimized CPUs or other compute boards become the norm (right now, you can get GPU boards - mainly from NVidia - that incorporate a GPU specifically for compute acceleration needs - with no video output; you find them mainly used in workstation-class machines, and of course supercomputers).
For deep learning: https://pcpartpicker.com/user/nalply/saved/pXrD3C This PC part list starts with about $2900 and only one GPU. Additional GPUs will cost about $1200 (unsure), so if it is maxed out with four GPUs we talk about $6500 and 1250 Watt. I will wait till I understand deep learning well enough before investing this much money and power.
For deep learning: https://pcpartpicker.com/user/nalply/saved/pXrD3C This PC part list starts with about $2900 and only one GPU. Additional GPUs will cost about $1200 (unsure), so if it is maxed out with four GPUs we talk about $6500 and 1250 Watt. I will wait till I understand deep learning well enough before investing this much money and power.
I'd say not only wait until you understand it, but to wait until it's your day-to-day job or business interest (or main hobby). Otherwise you'll be throwing money out the door (though you'll have one heck of reason to brag!)...
If you are willing to get used/prev-gen, then ebay is a good option, and you can probably get something In my opinion ECC isn't something I want to be without. I run lots of VMs. Crashes will quickly eat up any potential savings from scrimping on that.
Earlier quoted context omitted.
> I suspect your main motivator is to want to build a PC. Chances are that the extra computational power or cores won't actually make your work faster for day to day use. Another potential factor - If you play games or develop 3d applications, high end GPUs are much more affordable in desktop form factors than in laptop ones and generally have more horsepower than their laptop equivalents. If you do a lot of video en…
This may not be relevant to the OP, but I'm interested in machine learning, and having a graphics card with a big GPU is important. Serious machine learning computers have sky high prices, but for learning, you can get a graphics card with 640 or more gpu cores for $250 or less. I don't know of any laptops in the requested price range that have that kind of gpu power. Of course, a big graphics card can also be used f…
That said, I'm currently working on the behavioral cloning project (teach a virtual car to drive using what we've learned so far), and given the amount of data we'll gather (10s of thousands of frames we have to process) to train our models - I tend to wonder if I will need to upgrade.
Fortunately, that won't be too difficult for me - I have both a 960 and 970 waiting in the wings; they were cards originally destined for a mini-itx gaming machine build that I've never gotten around to (hmm - maybe I should just build it for my next workstation instead).
Also - 960s and 970s are dropping in price like a rock, since the hardcore gamers are moving to the 10xx line; while they may not have as many cores or as much speed, they are still very capable cards for ML purposes.
I've also been considering getting a multi-pcie slot server motherboard (maybe with dual xeons or such - these mobo combos are actually pretty cheap on ebay), and dropping in one of those cards, then later adding on more GPU cards - strictly for an ML box...
In my opinion there is only one answer to this question for anyone doing development: Buy the fastest non-extreme consumer intel processor, right now thats 6700k, add at least 16gb of ram, 1 mid-size ssd, 2x 4tb drives in raid. If you want to do any kind of GPU work add the fastest or second fastest consumer nvidia card, like the 1080 or 1070. If you want to game get a second ssd and install windows on it. It goes wi…
> the fastest non-extreme consumer intel processor > the fastest or second fastest consumer nvidia card > represents the sweet spot in price/performance, saving money on your axe is a false economy. Care to explain your reasoning? Going with high-end components seems to be the opposite of the sweet spot. A $200 i5-7600 and $250 GTX 1060 are going to be indistinguishable in non-benchmark settings for most workloads wh…
There are a number of tasks we do that are compute bottlenecked, how about compressing an archive, or compiling, or transcoding?
Even things like apt-get install is faster with a faster processor, and vim plugins can be ridiculous with their lag. It all adds up to hours and days and eventually weeks of your life over time.
I don't think the tradeoff of saving $100 or $200 on a chip that will last three years and will be worth more on resale is worth it.
In my opinion there is only one answer to this question for anyone doing development: Buy the fastest non-extreme consumer intel processor, right now thats 6700k, add at least 16gb of ram, 1 mid-size ssd, 2x 4tb drives in raid. If you want to do any kind of GPU work add the fastest or second fastest consumer nvidia card, like the 1080 or 1070. If you want to game get a second ssd and install windows on it. It goes wi…