Overclocking the $30,000 Computer! - 7 Gamers 1 CPU Part 3

Linus Tech Tips ·Linus Tech Tips ·2016-05-06 · 2,862 words · ~14 min read
Floatplane YouTube

Transcript

JSON SRT VTT 239
0:01 okay so we built seven Gamers one CPU a
0:04 $30,000 computer running unraid as an operating system and using it and some
0:09 really cool open- Source virtualization technology to run seven discret gaming
0:16 virtual machines at once all fully capable of like high performance gaming
0:21 so that was cool $30,000 computer two thumbs up very awesome 10 out of 10 IGN
0:26 Etc then what we did was we showed it off at CES okay so we got that out of
0:31 the way the world was able to see it in person then after that we brought it
0:34 back to the office converted our entire conference room here into a land Center
0:39 and had an inter office gaming competition complete with tan being
0:44 terrible at video games yes the very worst one here except he did kill me
0:47 once and you can check that video out here so you might be asking yourself why
0:52 is lion is still milking this damn thing what could possibly be left to do and
0:57 the answer is of course we're going to overclock
1:09 it Intel brings ddr4 to the mainstream
1:13 with their core i7 6700 K and core i5
1:16 6600k processors check out the link in the video description to learn more so
1:21 the first step anytime you do any Performance Tuning is to get a baseline
1:25 measurement because if you don't then you have no idea if the dials you're
1:29 turning are doing anything at all so I'm
1:32 going to introduce you guys to the benchmarks that we're going to be running cinebench R15 is going to tell
1:37 us what we need to know about CPU performance we are going to be using
1:40 Crystal dis Mark to tell us what we need to know about our SSD array performance
1:45 IID to 64 cash and memory Benchmark is going to be interesting especially once
1:50 we run it on all machines at once more on that later and we're going to be
1:53 using fire strike extreme to see how our
1:56 gpus are doing some of the stuff that's open is just tempor monitoring and stuff
2:01 like that and finally we're going to be having a look using our handy dandy wall
2:05 power meter dudad here at what kind of power consumption we see when we're
2:09 hitting the whole system at once but let's start with a single VM Baseline
2:15 with all the other ones sitting idle and I'll show you guys on screen hopefully
2:19 right now how I plan to go through this entire test suite and show you guys the
2:25 scores so here we go we're getting a cin bench score of around 600 we're getting
2:30 Crystal dis Mark scores in the uh area of sort of you know two uh gigabytes per
2:36 second reads and around 400 megabytes per second rights to our array our
2:41 memory scores we're seeing about 5 GB
2:44 per second so on average reads and wrs to our quad Channel ddr4 memory and our
2:50 3D Mark fir strike extreme score is 6148
2:53 so the next test and this one I will
2:56 need some helping hands for is going to be to hit all 7 VMS at once for our
3:02 Baseline score to see what happens when they have to actually share resources
3:07 which especially for things like memory it's not very pleasant sometimes okay so
3:12 I've assembled a crack Squad of helpers whose intelligence and physical
3:17 coordination uh lead me to believe that it is probable nay likely are both kind
3:23 of the same thing that they will be able to perform the skills that I set out for
3:27 them today task actually not skills don't worry too much about that basically what you guys have to do is
3:32 all press a button at the same time oh that's that's not not going to work okay
3:36 Colton you're fired again thank you okay
3:40 so two of you will have to press two
3:43 buttons at once at the same time Johnny
3:46 you got this okay Johnny you're on machines one and two yvon you're on
3:50 machine three Colton you're here Nick you're here and I'll go over there and I
3:53 will tell you which button to press and you will all press it at the same time
3:57 you're going to click the button that says Run next to CPU in the cinebench
4:03 R15 program has everyone got your cursor positioned yes okay so I'm going to say
4:08 don't do it yet I guess teran's not here I don't have to give instructions like
4:11 that okay it 3 2 1
4:15 go okay so you can see here with our seven instances of cinebench running we
4:20 are actually pinning the entire CPU here
4:23 at 100% usage that was exactly our intent and will tell us without any
4:29 extra a turbo boost for only activating a couple of cores at a time how all of
4:33 our VMS can run in a worst case scenario where they are sharing as much resources
4:38 as possible okay so we're scoring around 500 points which is about 100 lower than
4:43 we did when we were just running it on a single VM fascinating okay everyone
4:50 ready 3 2 1 go okay so this is
4:55 interesting on the read side of things really there wasn't much an effect from
5:00 the numbers I've looked at so far but on the right side of things some of our VMS
5:04 were so starved for disc time that they
5:07 actually timed out on their sequential right Benchmark very very interesting
5:12 this isn't something we're going to retest after we overclock because we can't overclock storage but uh I just
5:17 wanted to have this as just a for my information sort of number Ida 64 cash
5:22 and memory Benchmark it's a gray window this one I'm really interested in
5:27 because desktops really don't care much
5:31 about their memory bandwidth these days I mean dual Channel quad Channel DDR3
5:36 ddr4 it's all fast enough but if you're trying to hit
5:40 something with s VMS at once I'm
5:43 expecting the performance to be affected somewhat it's this is really interesting
5:48 so you can actually see the systems are running the benchmarks at different
5:51 speeds NX VM over here so this is player
5:54 number five is actually still running
5:58 the memory benchmarks or at least at least it was before I uh and mine is
6:02 almost done running all of the level one and level two cash benchmarks so it does
6:06 seem like their ability to even run these tests is affected by the fact that
6:10 we're hitting them all at once here we go last one position your cursor over
6:15 Run 3 2 1 fire strike
6:19 extreme thank you for that clicking like random times like okay I guess that's it
6:24 3 two one f with extreme okay you guys are dismissed for
6:28 now thank you I'll call you back once I overclock the system so check this out
6:32 this is what our uh memory and cach
6:35 benchmarking looked like this is what our benchmarking looks
6:41 like in fir strike extreme so with seven
6:45 instances of it running we're actually getting up near 100% utilization closer
6:50 to like 90 here
7:00 okay very interesting so 3D Mark scores not affected much by our CPU and RAM
7:06 sharing because 3D Mark is mostly dependent on the graphics card and each
7:10 VM has its own R9 Nano so all that's
7:13 left now to do is to record our score so we can create some average score graphs
7:18 for you guys and establish our Baseline so that we can find out if our
7:22 overclocking helps now most Server Motherboard makers
7:27 do not have any overclocking features but this is a Seuss and this is
7:32 technically a WS or workstation board so
7:36 in the AI tweaker menu we've got their
7:39 AI overclock tuner which has two settings neither of which I have played
7:43 with at all manual and OC
7:47 tune so I'm just going to go let's OC
7:52 tune it level three and find out if the system works
7:57 there's a very good chance this not going to work I have not tested did
8:00 this at
8:04 all so this is actually a really bizarre Quirk of this system it starts up
8:13 here oh oh if it turns on at all it
8:18 starts up here oh wow yeah that's not
8:24 posting let's clear simos and then
8:28 uh we will go from there
8:32 okay something to see if that does
8:37 anything oh that was it I need this bar
8:40 to get past like
8:44 here oh that wasn't it like I I
8:52 oh I know what it is because we did a
8:57 seimos clear that bio setting that I need the 4G
9:03 decoding setting is not enabled so we
9:07 are not able to boot with all of these PCI
9:10 devices so maybe by
9:16 unplugging all the video cards I will be able to have them not
9:22 get picked up and I will be able to get into the
9:26 BIOS this is how you remove the graphics
9:30 cards from Seven Gamers one
9:36 CPU there we go okay they're out so
9:42 now can try again okay so that was
9:48 it the problem with having a system that doesn't Post in its default
9:54 configuration okay so let's try manual this time um
9:59 my understanding is that on these types
10:02 of boards you can expect somewhere in
10:05 the neighborhood of like5 MHz so just
10:09 for LS let's try 105 and see what
10:13 happens I'm guessing not a whole lot and
10:17 that's what I
10:21 got okay then okay so after four
10:24 attempts we're back in the BIOS our setting did stick but obviously didn't
10:29 work so let's try a more conservative 103 which would be a 3% overclock okay
10:34 so because we have no video card we have to use VNC to get into our VM and find
10:40 out if everything is working as intended
10:43 well that is not reporting correctly at
10:48 all so that's a thing so we're going to have
10:53 to run a benchmark and find out okay I
10:56 tell you VMS really screw up hard
10:59 Hardware reporting both CPU
11:03 Z and Ida 64 are reporting numbers that
11:07 would indicate that our CPU is not overclocked but cinebench is reporting
11:12 that it is overclocked and even though these scores are not directly comparable
11:16 because we don't even have any of the other VMS booted we're getting 900 in
11:20 cinebench which would indicate that everything's working properly at the
11:25 very least so now we can proceed to the next step and put our graphics cards
11:29 back back in okay all our video cards are showing up which basically means we
11:33 can start the array and fire up our
11:38 VMS did it just do that or did you do that hey all
11:44 right say that again so now it is time to overclock our video cards now the R9
11:49 Nano isn't uh really known to be much of
11:52 an overclocker uh but we're going to go
11:56 ahead and increase the power limit as high as it goes and we're going to
12:00 increase our core clock to like I don't know 1050 and see what happens hey
12:06 Graphics 1050 MHz just like that I guess
12:11 water cooling helps a little bit with the whole overclocking the R9 Nano
12:16 situation very nice what the hell
12:20 afterburner is detecting all seven gpus
12:24 that shouldn't be possible cuz we've passed them through
12:28 to separate VMS it's also detecting wow
12:32 afterburner doesn't know what's going on so it's got them all being used while
12:36 we're running The Benchmark so it knows there's seven but it doesn't know that
12:41 it's only using one of them and it's just cloning this core clock graph which
12:47 actually has the card not reaching uh
12:51 not reaching 1,000 mahz anyway
12:54 so I guess all that's left now then is to run some benchmarks and find out if
12:59 we got any benefit for our trouble everyone ready overclocked cin bench
13:06 Benchmark what damn it Nick if ID wanted someone to just start The Benchmark
13:10 early I'd have asked Taran to help please work please
13:15 work please work no no what no just kidding player
13:22 whatever it is so it's 1 2 3 4 5 6 7 and
13:27 then you put OC score in a bench but my
13:30 keyboard says six don't worry about that
13:34 okay actually that does seem to be
13:37 faster good yeah 40 points better on
13:41 mine 30 points everyone ready ready 3 two 1 run
13:49 it please don't
13:55 crash oh M oh
14:03 why it's okay give it
14:07 time power draw 1,600 Watts so did
14:12 everything but Nicks launch yeah okay so
14:15 after much struggling we managed to achieve whopping overclocks of anywhere
14:21 from zero to 70
14:26 mahz on our graphics cards you building
14:30 fire strike extreme scores in the very
14:34 very close to what we were already getting range leading us to believe that this
14:39 whole thing was a terrible idea but I'm really glad that you guys got to come
14:44 along for the adventure with us because um that is what we do on this
14:51 channel that is what we do this is a
14:54 sponsor spot I have been looking forward to for quite some time we recoverd
15:00 dat.com what they do is pretty obvious
15:04 they recover data and why we're affiliated with them will be pretty
15:08 obvious if you checked out the data recovery Vlog which if you haven't
15:13 already is a pretty uh pretty exciting
15:17 piece of content as far as they go on this channel anyway they gave me some
15:22 bullet points to talk about and I will do that but then I'll get into why
15:26 they're so special to me so first of all they're one of the only dat recovery
15:29 companies to invest in R&D so if the technology to recover your data doesn't
15:34 exist they create it they're fast they're techsavvy they have friendly
15:38 customer service available 247 and that was indeed my experience their data
15:43 recovery guarantee policy means that if they don't recover any data there is no
15:48 recovery fee they have worldwide emergency data response options
15:52 including onsite remote and in lab that's 247 365 they've got
15:57 state-of-the-art equipment utilizing today's most advanced technology with
16:00 proprietary in-house developed tools they've got iso4 class 10 and iso5 class
16:05 100 clean rooms to enable them to provide a Optimum solution for a
16:10 recovery environment and basically what it boils down to for me is they got my
16:16 data back it was an extremely
16:19 complicated raid array not a recommended raid array so I had three raid fives
16:25 that were Hardware raid fives on separate LSI controllers and then I had
16:29 those striped together in raid zero for an effective raid 50 so three raid five
16:34 stripes in Windows I thought it should be pretty safe because I can lose up to
16:40 one SSD per controller before I should run into any issues but what I didn't
16:45 account for is losing a controller
16:48 outright so one of the controllers failed causing the array to go into an
16:53 unrecoverable State and we recoverd dat.com had me plug in all those 24
16:58 drives to a host bus adapter and they
17:01 were able to remotely using their own Linux tool reconstruct the whole thing
17:06 allow me to copy it off and basically
17:09 save my turkey which uh thanks to those
17:13 guys and if you guys have a data recovery emergency then check them out
17:17 the link is at the video description I heartily endorse we recoverd dat.com so
17:23 thanks for watching guys if you disliked this video hit the dislike button but if
17:26 you liked it and you thought it was crazy and a an extreme hit the like
17:30 button get subscribed and maybe even consider supporting us by buying a cool
17:33 shirt like this one changing your Amazon bookmark one with our affiliate code instructions for up there or just
17:38 contributing directly through our all new Linus Tech tips Forum freshly
17:42 updated for 2016 which you can check out
17:45 at the link in the video description and now that you're done watching this and
17:48 you're wondering hey what should I watch next maybe check out the video where I
17:51 show you how to quadruple your network speed for 100 bucks that is assuming
17:55 that you have a couple of really good computers already and you have like networking equipment in you have eBay
17:59 and etc etc etc terms and conditions they
18:03 apply links up there