text
stringlengths
9
408
[1605.56 --> 1608.76] So I used the Plex sync for offline
[1608.76 --> 1612.22] playback feature and I chose a high bit
[1612.22 --> 1612.50] rate.
[1612.62 --> 1615.78] So about a 38 megabyte a second.
[1616.44 --> 1617.54] Was it megabit?
[1617.62 --> 1618.52] I always get confused.
[1618.96 --> 1622.62] Video file encoded with MPEG-4 and H.264
[1622.62 --> 1626.52] with a DTS master audio soundtrack.
[1626.92 --> 1631.04] And I used the four megabytes a second 720p
[1631.04 --> 1634.00] sync for offline playback option within Plex.
[1634.00 --> 1636.36] And I did that on an iPad.
[1636.90 --> 1637.90] You know, that was the client.
[1638.02 --> 1639.48] I don't think that bit really mattered too
[1639.48 --> 1641.56] much, but I did a few different tests.
[1641.68 --> 1642.86] So I did a software render.
[1643.04 --> 1647.76] So this is using the Intel i5-8500 CPU.
[1648.24 --> 1650.20] I picked that up for about a hundred dollars
[1650.20 --> 1650.88] used on eBay.
[1651.04 --> 1652.32] So it's a pretty good value.
[1652.68 --> 1653.44] Four gigahertz.
[1653.54 --> 1656.44] I think 4.1 gigahertz, six core CPU.
[1657.20 --> 1661.60] With CPU rendering only, I saw a 1.1 times
[1661.60 --> 1662.32] speed.
[1662.32 --> 1665.96] The best I saw running it natively in QuickSync
[1665.96 --> 1668.00] on the host was 10.2.
[1668.52 --> 1671.22] And then on a sliding scale between those two
[1671.22 --> 1676.86] numbers of 1 times and 10 times, with the GVTG stuff set
[1676.86 --> 1682.78] into two slice mode, I saw only about a 1.8, 1.9 times.
[1683.32 --> 1687.68] So it was faster than CPU software encoding, you know, twice as fast
[1687.68 --> 1695.28] actually, but it was 80, what, something, 82% slower than running it on the
[1695.28 --> 1695.96] bare metal host.
[1695.96 --> 1702.70] So the other problem that I ran into, besides, you know, leaving 80 plus percent performance
[1702.70 --> 1705.44] on the table, was stability.
[1705.86 --> 1708.76] Unfortunately, that wasn't a great story either.
[1709.18 --> 1715.52] The problems I ran into were, so I was running Proxmox as the base OS because the Proxmox
[1715.52 --> 1720.86] wiki actually has a really great entry about enabling GVTG and QuickSync and pass through
[1720.86 --> 1721.64] and all that kind of stuff.
[1721.64 --> 1723.58] And it was really painless to get going.
[1723.68 --> 1726.52] It only took me an hour or so to figure it all out and get it going.
[1727.38 --> 1729.50] But the stability was just not there.
[1730.02 --> 1731.92] My evidence is only anecdotal, though.
[1732.34 --> 1736.12] Because of the instability, I couldn't really get the log files that I needed
[1736.12 --> 1741.68] because the system either had to be hard reset or it was just hanging and processes were just
[1741.68 --> 1743.48] hanging or I was getting kernel panics.
[1743.48 --> 1745.24] It was just a mess.
[1745.80 --> 1748.32] And, you know, at the end of the day...
[1748.32 --> 1751.92] Just not something you could just let run and just not have to think about it.
[1752.06 --> 1753.70] No, unfortunately, it wasn't.
[1753.80 --> 1759.76] And, you know, things worked fine until I powered on the Windows VM that was running Blue Iris.
[1760.00 --> 1763.54] And that system has six 4K cameras going into it.
[1763.58 --> 1765.38] So it's got a decent amount of load.
[1765.38 --> 1772.26] Now, I know for a fact that the i5 CPU can handle that load because it's been running in my HP 290 Slim
[1772.26 --> 1776.72] for six months just fine, handling everything perfectly.
[1777.24 --> 1784.86] But for some reason, when it's in the GVTG mode, it just, I guess, because the performance is so poor
[1784.86 --> 1791.04] with the emulation, whatever they're doing to slice the GPU up, however they're doing it in the Intel drivers,
[1791.04 --> 1799.72] it just meant that as soon as I powered up Blue Iris, within 30 minutes, the whole Proxmox system was just unhappy.
[1800.06 --> 1804.30] You know, like the web UI wouldn't load sometimes or you actually go and reboot the system
[1804.30 --> 1810.78] and you'll see SystemD printing out waiting on Kimu guest to shut down for like 30 minutes.
[1811.32 --> 1815.62] So it wasn't just guest stability issues, but the entire host.
[1815.82 --> 1816.14] Uh-huh.
[1817.06 --> 1817.90] Ooh, yeah.
[1818.18 --> 1819.56] Well, that's just a deal breaker.
[1819.56 --> 1822.18] Oh, man, Alex.
[1822.36 --> 1824.98] At the end of the day, you want it to be on and functional.
[1825.24 --> 1831.26] And yes, this is a hobby, and I do enjoy messing about with servers, but there comes a point.
[1831.34 --> 1832.34] It's not a full-time job.
[1832.50 --> 1833.24] No, exactly.
[1833.58 --> 1837.70] There comes a point where you're like, this S just needs to work now.
[1838.04 --> 1838.26] Yeah.
[1838.64 --> 1839.98] And it just didn't, unfortunately.
[1840.26 --> 1844.58] I think you crossed the threshold of devoting more time to this than most folks quite a while ago.
[1844.58 --> 1848.36] I mean, even taking a pass at the different encoding options, that's good insight.
[1848.36 --> 1854.76] And it really shows you that you're, I mean, yeah, it's almost twice as fast if you use GVT.
[1854.76 --> 1860.42] When you consider the stability issues, I would rather just use CPU encoding.
[1860.42 --> 1869.42] CPU encoding with QuickSync, because that's only pulling down about 10 watts when it's doing a full 1080p stream, and it's running at 10x real time.
[1869.92 --> 1874.70] So it actually, you know, in terms of performance per watt is the best thing out there.
[1874.94 --> 1879.18] I actually also, just for giggles, use my 1080 Ti to run one of these transcodes.
[1879.76 --> 1881.62] So QuickSync is running at 10x.
[1882.02 --> 1888.02] A 1080 Ti, which is pulling down 18 times the amount of power, is running at 17 times.
[1888.20 --> 1889.98] So 10x versus 17x.
[1890.08 --> 1890.50] Wow.
[1890.70 --> 1897.44] It's a good way to see the performance discrepancy between Intel's GPU and the NVIDIA GPU right there, isn't it?
[1897.46 --> 1897.96] You can really.
[1898.28 --> 1900.82] And then, of course, CPU with its measly 1.1x.
[1901.24 --> 1901.40] Yeah.
[1902.02 --> 1905.70] Performance per watt of the NVIDIA card was just hilariously bad.
[1906.06 --> 1906.58] Well, sure.
[1906.98 --> 1907.30] Sure.
[1907.30 --> 1914.94] So what I've ended up doing, actually, and this is to go back to Ryan's question now, is I've ended up splitting things back out into physical boxes.
[1915.56 --> 1920.04] The Blue Iris box is an HP290 Slim that I already had.
[1920.44 --> 1927.58] I've put the i5 back into there, and I bought another i5 to put into my server, so I have two now, which is a shame, but hey-ho.
[1928.06 --> 1932.36] I was doing some testing on the HP290 for average power usage.
[1932.36 --> 1938.12] When I say idle, what I mean by that is Blue Iris is running with its normal sort of load.
[1938.30 --> 1945.72] So an average sort of power usage I see on that box is anywhere from sort of 10 to 25 watts, depending on what it's doing.
[1945.72 --> 1951.08] Which, for the performance I'm getting, you know, that kind of power usage is fantastic, really.
[1951.82 --> 1960.20] And then my main server is pulling it anywhere with, you know, it's got, I think, 12 hard drives in it, anywhere from 40 to 80 watts at idle.
[1960.20 --> 1967.36] So, you know, it's all pretty good, pretty low, you know, to have all that performance for under 100 watts at idle is, I'm pretty happy with that.
[1967.90 --> 1969.64] Yeah, thanks for sharing the details with us.
[1969.68 --> 1970.86] You mentioned the blog in there.
[1970.96 --> 1973.38] Well, Sam actually writes in with our next question.