Sean Meighan
Hardware => General hardware => Topic started by: JonB256 on May 12, 2017, 08:06:40 AM
-
I've supported AMD for years (it's an underdog thing) and the new Ryzen cpus and motherboards look and are testing great.
Any experience with xLights yet?
I don't game and xlights is my most CPU intensive program.
Sent from my SM-G900V using Tapatalk
-
I've read otherwise, they are underperforming.
Sent from my Pixel using Tapatalk
-
When it comes to differences between Intel and AMD cpu's, or AMD and nVidia gpu's a lot can be said for what the code being run is optimized for. In Xlights case I don't think you will see much of a difference in performance from Ryzen compared to other chips in it's class considering I don't think there are any special optimizations in xlights for it. Knowing a little bit about computer hardware and programming after 30+ years in the industry, I would bet you will see much larger performance gains in xlights by using it on an ssd drive which is set as your primary partition (ie where your operating system is installed also).
The other thing that can slow xlights or other programs down is not enough ram. If your computer is constantly accessing the virtual memory because of a lack of ram, that could really make your system crawl. While I don't think xlights is a real memory hog, if you have a lot of programs loading up in the background that could also slow you down quite a bit.
Sent from my SM-G920V using Tapatalk
-
I've usually always built my PCs since the 90's. I don't even remember the issue but I did an AMD processor once and had issues and it just left a bad taste so I don't get suckered in to the lower prices. Also when people were having video issues in xLights a year or so ago it was primarily AMD video cards having problems. So I guess using Intel is like using a Falcon....sure others may save you a nickel but cost you a dime.
-
Admittedly, I'm leaning toward an i7 and good nvidia card. I've had good luck with my last 5 or 6 AMD based boards but it may be time to swap.
Sent from my SM-G900V using Tapatalk
-
I guess everybody's experience is different I've primarily built AMD machines because I don't need top notch performance for what I do and couldn't justify the extra cost. Recently I've built a budget AMD gaming machine for my son, initially he had some trouble but that was solved with an SSD.
-
JohnB let me save you a couple hundred dollars on your purchase, if you don't game or use some other high end graphics program, don't get a pc with an nVidia graphics card. It'd be complete overkill for running xlights. The onboard graphics on an Intel processor will be plenty adequate for running xlights and other similar applications. You would do better to invest that money you would spend on a discrete graphics card in an ssd, stronger cpu, and or more ram.
I like Gilrock have been building my own PC and PC's for others since the early 90's. And I too like Gil have gotten a bad taste in my mouth from obscure AMD issues. Most everything is written and optimized for Intel processors. So you will have less problems with an Intel chip.
I have a high end Asus laptop with Windows 10 and a nVidia GTX970 gpu and on my setup I can choose between Intel HD4000 onboard graphics and the nVidia GTX970 discrete graphics card, and I noticed zero performance increase in XLights when using the vastly more powerful nVidia card. If anything I would say XLights ran worse on the nVidia card since it appeared to crash more often. So I switched back to the Onboard Intel HD4000 graphics when using XLights.
Sent from my SM-G920V using Tapatalk
-
Building on the thread...
what would work better?
More cores and slower proc speed...
Fewer cores and faster proc speed...
I also build my own computers because I like the faster performance for the dollar compared to what I can buy off the shelf.
With processors that have 10 cores or more... I'm wondering if its simply a matter of multiplying clock speed X core when drawing comparisons, or if xLights will take particular advantage of higher core count.
btw... Came across this site recently. Its cool because it takes performance measurements from your system (processor, memory, disk, graphics, ...) and compares your results against what other users are getting with the same equipment -- this can tell you if your system is optimized or can be improved. it also shows how your hardware choices compare to other options that are out there.
www.userbenchmark.com
-
Let me first say that I am not very knowledgeable about the xlights code yet, only just started poking around in it.
But software must be written specifically to take advantage of multicore processors. And even when it is, having many cores compared to just a few doesn't necessarily make your software faster. Some work loads just don't lend themselves very well to running on many cores.
An example of a work load that benefits from multicore processor is a chess engine using a parallel search algorithm to traverse the game tree.
The reason I say that is because chess is a game of perfect information (ie you know the current state of the game at all times), so the work load can be easily distributed by all the available moves from the root node (ie current position). Since the order you search the moves from the root doesn't matter. They all have to be searched in order to find the best move.
However, you cannot use parallel processing to search the depth of the game tree and expect the same speed gains since you cannot easily predict what order you need to search the nodes in and the order of the nodes you search at depth is critical to the speed of the search (See alpha-beta pruning to understand why).
But one thing for sure, rarely if ever does multi-core processing result in a linear increase (ie 2 cores are twice as fast, 4 cores are 4x's faster).
So thinking logically about xlights, I don't think there are a lot of ways to improve it's performance using multicore processing techniques. Since I doubt xlights does much to tax the limits of a single core. Maybe background rendering performance could be improved a little, or doing a full render of a sequence could be improved since the order that the effects get rendered doesn't matter. But considering that there is a pretty big bottle neck that occurs during rendering since the rendering has to be saved to an fseq file (and secondary storage is pretty slow compared to the rest of the pc), you probably couldn't increase it by all that much.
So if xlights is programmed to take advantage of true multicore processing, it probably doesn't increase it's performance all that significantly. And certainly using a 10 core compared to a 4 core would probably add no tangible benefits. I can imagine few software packages that would benefit from a 10 core processor.
So I would pay special attention to the single core performance of a processor when you are building.
When you see 8 or more cores on a processor, it is usually more for power management reasons. So they are set up in a big/little configuration where half the cores are slow power efficient cores, and run when you aren't doing anything very demanding, but then when you do something really demanding the big powerful not so power efficient cores take over to complete your demanding task, then they turn control back to the little cores. Cell phones, tablets, and laptops benefit from this sort of arrangement.
Of course I am only talking about the consumer side of computing, business servers are a whole other type of beast where high core counts are mandatory since they are serving the needs of numerous users simultaneously, and it is easy to break down those work loads.
Hope this makes sense.
-
xLights is definitely designed to take advantage of multiple cores for our rendering engine. It spawns many threads that will render in parallel. The amount of gain depends on your sequence and layout because if we need to render a group before we render a model in that group then those two operations cannot be run in parallel. But if you have a bunch of models that don't conflict or depend on each other we can render them all at once using multiple cores.
-
So thinking logically about xlights, I don't think there are a lot of ways to improve it's performance using multicore processing techniques. Since I doubt xlights does much to tax the limits of a single core. Maybe background rendering performance could be improved a little, or doing a full render of a sequence could be improved since the order that the effects get rendered doesn't matter. But considering that there is a pretty big bottle neck that occurs during rendering since the rendering has to be saved to an fseq file (and secondary storage is pretty slow compared to the rest of the pc), you probably couldn't increase it by all that much.
There are so many assumptions that are incorrect here. We use multiple cores in many instances per my last reply and the saving of the fseq is the fastest part of the operation. We render to memory and then perform a write to disk to create the fseq file which happens in under a second normally.
-
Gil I was hoping you would chime in, it's always a pleasure to hear from the xlights dev team.
Obviously pure speculation on my part about what areas parallel processing is used in xlights. Can't say I am too surprised that the fseq is rendered in memory, and saved add a whole. I just thought I had saw during a full render that there was more than one rendering... And writing.... Output message displayed at the bottom of xlights. But now that I go back and look, there isn't.
Sent from my SM-G920V using Tapatalk
-
I have finally assembled the new system. Went with a Ryzen 7 1700 and an MSI B350 motherboard.
This board supports the M.2 style SSD, so I put a 500GB SSD as my boot drive. Two new 16GB DDR4 modules were installed.
Did a fresh install of Windows 10 Pro 64bit. That has cost me some time (a few hours, anyway)
Plus, I was initially getting a LOT of "clock_watchdog_timeout" reboots (blue screen). There seemed to be some kind of timing problem between the SSD and the CPUs and the SATA drives.
Turns out that the initial Win 10 install from DVD needed many, many updates. Eventually, one of them brought stability to the system.
Now, the good news - two sequences in xLights that used to take 72 seconds to render and save now take 9.5 seconds. Since the Ryzen 7 has 8 cores (shows as 16, counting hyperthreading), I was curious about how it split up the load. In the Resource Monitor, CPU tab, it shows the first 3 CPUs doing the most, then tapering off. A few of the hyperthread CPU graphs stayed as "parked" with no visible activity.
The motherboard has integrated graphics capability if you use a integrated GPU/CPU. The Ryzen 7 does not, so I continue to use a Zotac GT 730.
-
Excellent news. Glad to hear things have improved for you. Rendering in about an eighth of the time is pretty good increase. Saves you over a minute each time!
Sent from my SM-G920V using Tapatalk