WEBVTT 00:00:00.000 --> 00:00:05.759 this is the fastest seat the fastest CP 00:00:03.300 --> 00:00:10.500 fastest CPU on the planet in the world in the world and why do I feel like this 00:00:08.280 --> 00:00:16.400 has happened before it's because AMD keeps doing this they're all new epic 00:00:13.500 --> 00:00:23.580 Genoa CPUs are available with up to 96. Zen 4 cores and a single socket they 00:00:20.400 --> 00:00:28.560 support up to six terabytes of ddr5 00:00:23.580 --> 00:00:31.320 memory and feature 128 Lanes of PCI 00:00:28.560 --> 00:00:35.340 Express Gen 5 connectivity there is quite literally nothing else in the 00:00:33.120 --> 00:00:40.440 world like it even the size it's absolutely monstrous and this video is 00:00:38.340 --> 00:00:47.100 sponsored by super micro Who provided not one but two servers rocking a total 00:00:44.160 --> 00:00:51.840 of three of these CPUs does the math even work out to on that that's like 00:00:49.260 --> 00:00:55.399 almost 300 cores we've got to break some World Records today ladies and gentlemen 00:01:02.180 --> 00:01:08.700 the first thing that's new about epic Genoa aside from everything is the 00:01:06.180 --> 00:01:11.939 socket like its predecessors it uses an LGA style socket meaning that the pins 00:01:10.500 --> 00:01:16.820 are built into the motherboard rather than the bottom of the CPU but unlike 00:01:14.159 --> 00:01:23.400 its predecessors it has a whopping 600 and 96 pins necessitating a massive 00:01:21.060 --> 00:01:27.840 landing pad on the motherboard and a seven screw hold down mechanism in order 00:01:25.740 --> 00:01:31.979 to make sure that even contact is made with all those little pins but they 00:01:30.479 --> 00:01:36.720 absolutely need them and the biggest reason they need them is that to keep 00:01:33.540 --> 00:01:39.799 those 96 cores fed with data and working 00:01:36.720 --> 00:01:43.079 efficiently aside from having 00:01:39.799 --> 00:01:45.780 384 megabytes of onboard cache epic 00:01:43.079 --> 00:01:52.680 Genoa features a not two not four not six but 12 channel ddr5 memory 00:01:49.740 --> 00:01:58.860 controller yeah it only runs at 4 800 Mega transfers per second but it's 12 00:01:55.259 --> 00:02:01.500 channels that is an effective 12 Xing of 00:01:58.860 --> 00:02:06.240 the bandwidth of a single module hey we never did the size comparison with an 00:02:03.240 --> 00:02:08.220 Intel Desktop CPU check this out look at 00:02:06.240 --> 00:02:11.220 it like a little baby we seem to have run into a problem already I cleaned all 00:02:09.720 --> 00:02:15.239 the thermal compound off of this when we disassembled the server earlier and I 00:02:13.739 --> 00:02:22.379 don't know if we have enough paste in this building feeling like at like X for extreme 00:02:19.980 --> 00:02:26.580 we just used half a tube of thermal paste you ain't building more than one 00:02:24.840 --> 00:02:31.860 server with one of these if you got a dual CPU system on that note super micro 00:02:29.700 --> 00:02:36.180 who sponsored this video has a whole lineup of Epic Genoa systems available 00:02:34.319 --> 00:02:40.200 right at launch everything from their Cloud DC Workhorse affordable systems 00:02:38.819 --> 00:02:44.459 like the one I'm working on right now all the way up to their hyper and GPU 00:02:42.660 --> 00:02:49.140 oriented servers which are for absolutely zero compromises performance 00:02:47.160 --> 00:02:54.599 I mean can you really say it's a compromise and performance if you have 00:02:50.459 --> 00:02:56.519 up to 192 CPU cores and up to 160 PCIe 00:02:54.599 --> 00:03:01.560 lanes and a single system you cannot even though AMD has switched 00:02:59.160 --> 00:03:05.760 to a heatsink actuated socket so that is to say screwing in the heatsink is what 00:03:03.599 --> 00:03:10.860 applies enough pressure to the CPU to make contact with the pins I Gotta Give 00:03:07.860 --> 00:03:12.659 Them credit for still having that tray 00:03:10.860 --> 00:03:17.640 and that holder to keep it in place because men when you install some of the 00:03:15.840 --> 00:03:24.540 competition CPUs where they don't have that retention it is scary 00:03:20.819 --> 00:03:26.280 oh wait okay this is bad CPU just lifted 00:03:24.540 --> 00:03:30.900 up quick note if it was important to have AMD's previous high performance 00:03:28.260 --> 00:03:34.980 sockets torqued correctly it is triply important to have these sockets torqued 00:03:32.940 --> 00:03:38.580 correctly because even one screw slightly over or under tightened is 00:03:37.019 --> 00:03:41.580 going to have the CPU sitting like this in the socket that's going to be a real 00:03:40.019 --> 00:03:46.500 bad time you can have everything from memory channels not detected to flaky 00:03:43.860 --> 00:03:50.760 PCI Express Behavior to A system that doesn't boot at all one of the drawbacks 00:03:48.900 --> 00:03:53.879 of having a 12 channel memory controller is that if you want to get the maximum 00:03:52.440 --> 00:03:59.580 performance out of your system you better be ready to install a lot of memory sticks which isn't to say that 00:03:57.239 --> 00:04:02.819 you have to use 12 though if you want to do less than that I would strongly 00:04:01.080 --> 00:04:07.560 recommend checking AMD's documentation because it is not as simple as just 00:04:05.220 --> 00:04:12.540 popping sticks in wherever you want oh this is the first time I've been hands 00:04:09.120 --> 00:04:15.599 on with a ddr5 server module now regular 00:04:12.540 --> 00:04:18.720 ddr5 has a form of error checking and 00:04:15.599 --> 00:04:22.440 correction or ECC but it's only handled 00:04:18.720 --> 00:04:25.320 at the Die Level whereas a full ECC 00:04:22.440 --> 00:04:33.180 module is going to have to have an extra die for every eight which actually is 00:04:29.940 --> 00:04:35.460 not how this is configured these have 10 00:04:33.180 --> 00:04:42.540 packages per side so do they actually have 00:04:38.240 --> 00:04:44.220 two ECC packages per eight after a bit 00:04:42.540 --> 00:04:49.080 of off-camera research it seems to be related to the way that individual ddr5 00:04:46.620 --> 00:04:53.580 memory dies operate more like two previous generation dies meaning that 00:04:51.180 --> 00:05:00.120 instead of needing one chip per group of eight you need one chip per group of 00:04:56.280 --> 00:05:01.800 four so these are fully proper ECC dimms 00:05:00.120 --> 00:05:06.419 that will protect data even when it's in flight there's some other differences 00:05:03.300 --> 00:05:09.000 too just like consumer ddr5 modules the 00:05:06.419 --> 00:05:13.139 power management IC or pmic is built into the module rather than into the 00:05:10.860 --> 00:05:16.919 motherboard but unlike consumer modules the notch is in a different place 00:05:14.820 --> 00:05:22.199 meaning that you won't be able to have a motherboard that will support both 00:05:18.860 --> 00:05:25.139 unbuffered and ECC memory at least as 00:05:22.199 --> 00:05:28.919 far as I can tell this is all very very new at the moment before we go too much 00:05:27.419 --> 00:05:32.940 further and start covering things up now is the perfect opportunity to talk not 00:05:30.720 --> 00:05:37.380 just about the performance but also the flexibility a platform like epic Genoa 00:05:35.580 --> 00:05:43.199 provides I mentioned already that each of these CPUs supports up to 128 Lanes 00:05:40.380 --> 00:05:47.600 of PCI Express gen 5. that means even on a basic server from Super micros Cloud 00:05:45.300 --> 00:05:54.000 DC lineup you are looking at an utterly unbelievable amount of connectivity this 00:05:50.639 --> 00:05:55.919 one alone we've got dual ocp 3.0 cards 00:05:54.000 --> 00:06:02.580 which are typically used for networking we've got these cxl slots okay each of 00:05:58.979 --> 00:06:04.500 these running at pciegen 5x16 and cxl is 00:06:02.580 --> 00:06:08.820 so cool it uses the same physical connections as PCI Express but a new 00:06:06.960 --> 00:06:13.860 protocol that allows you to do wild things like put a a separate memory 00:06:11.940 --> 00:06:19.740 controller into a slot in your computer and add a whole bunch more RAM to it 00:06:17.220 --> 00:06:23.639 like what and even if you fill all of this stuff up in a server like this one 00:06:21.840 --> 00:06:28.259 that's using three and a half inch Bays you are going to run out of slots before 00:06:25.979 --> 00:06:32.479 you're going to run out of PCIe lanes as you cram the front of it full of ssds 00:06:32.900 --> 00:06:39.180 now unfortunately PCI Express Gen 5 ssds 00:06:36.960 --> 00:06:44.340 haven't hit the market yet but when they do things are going to get really cool 00:06:41.280 --> 00:06:46.740 because instead of needing four lanes to 00:06:44.340 --> 00:06:50.819 get the most out of these ssds in non-performance critical applications 00:06:48.539 --> 00:06:55.319 like say it's more about capacity you're going to be able to take just two Gen 5 00:06:53.400 --> 00:07:00.720 lanes and get the same performance you would have gotten from four Gen 4 Lanes 00:06:58.139 --> 00:07:06.360 just one generation ago and what was really mind-blowing for me thinking 00:07:02.340 --> 00:07:09.900 about how fast PCI Express Gen 5 is a 00:07:06.360 --> 00:07:15.380 single PCIe Gen 5 Lane is the same speed 00:07:09.900 --> 00:07:15.380 as a 16x slot of PCIe gen 1. 00:07:16.199 --> 00:07:22.319 I think size doesn't matter that much Yvonne you're watching this right look 00:07:20.340 --> 00:07:27.060 they can't even put full-size slots on the board they just got these 8X are 00:07:24.720 --> 00:07:30.660 those 8X I think they're 8X and this has applications outside of just storage as 00:07:29.039 --> 00:07:35.940 well say for example we were going to take this PCIe gen 5x16 slot right here 00:07:33.780 --> 00:07:43.380 and plug a you know daughter board or mezzanine card into it we could plug in 00:07:38.639 --> 00:07:46.139 literally four PCIe gen 4x8 cards and 00:07:43.380 --> 00:07:49.259 have all of them run at full freaking speed 00:07:47.400 --> 00:07:53.520 that's wild well you need a pretty complex PCIe switch to go back to Gen 4 00:07:51.840 --> 00:07:57.120 and have that full bandwidth but but that's not the point the point is that 00:07:55.199 --> 00:08:01.919 outside of storage and networking almost nothing needs Gen 5 bandwidth anyway and 00:08:00.060 --> 00:08:06.240 you can see super micro knows that so this is just a simple daughter board 00:08:03.360 --> 00:08:13.259 that is absolutely covered in Gen 5 Lanes we go from 16 to 00:08:09.800 --> 00:08:15.660 16 to 80 lanes and they figure this will 00:08:13.259 --> 00:08:18.060 probably be fine and they're probably right 00:08:16.560 --> 00:08:22.020 not that we're doing anything like that with it we're just using one of these 00:08:19.740 --> 00:08:28.680 slots for an RJ45 network card because we don't have the PC build corner wired 00:08:25.020 --> 00:08:30.479 up with qsfp which is with the network 00:08:28.680 --> 00:08:34.560 card that super micro sent over with this thing uses okay there's one more 00:08:33.060 --> 00:08:41.940 thing about this socket that I didn't tell you guys yet unlike AMD's previous 00:08:37.320 --> 00:08:44.760 server socket sp5 is rated for up to 400 00:08:41.940 --> 00:08:49.800 watts which is like what I thought Zen 4 was more efficient 00:08:46.380 --> 00:08:54.240 well yeah it is but we went from up to 00:08:49.800 --> 00:08:56.040 64 cores to up to 96 cores some 00:08:54.240 --> 00:09:00.720 concessions had to be made and that's why can I remind you again this is a 00:08:58.440 --> 00:09:06.300 cloud DC like Workhorse class server super micro is shipping 1200 watt 00:09:04.019 --> 00:09:11.459 redundant power supplies in a server that's not even designed for GPU use 00:09:08.240 --> 00:09:13.560 spicy look how long this thing is not 00:09:11.459 --> 00:09:16.080 that it's about length Yvonne you're watching this right it's 00:09:14.820 --> 00:09:21.480 not about length but is it longer than a 40 90. oh 00:09:19.320 --> 00:09:24.320 40 90 for scale okay 00:09:22.680 --> 00:09:30.720 let's fire this thing up we have a liftoff now for best possible 00:09:28.980 --> 00:09:35.459 cooling we're going to want to put the top cover on this because otherwise 00:09:32.399 --> 00:09:38.519 these fans instead of drawing air across 00:09:35.459 --> 00:09:41.160 these NVMe drives and blowing it through 00:09:38.519 --> 00:09:46.380 the CPU heatsink they're just drawing air from here and blowing it out here 00:09:42.839 --> 00:09:48.959 also that's really annoying 00:09:46.380 --> 00:09:53.880 that's a little better three hours later no but in all seriousness at the moment 00:09:51.779 --> 00:09:59.519 epic Genoa takes about 10 minutes to boot up but AMD says they 00:09:56.640 --> 00:10:03.060 are working on that so good 00:10:00.600 --> 00:10:07.620 gotta get that memory training going hey hey 00:10:05.040 --> 00:10:11.459 we had time to go get monitors plug them all in get 00:10:09.420 --> 00:10:15.120 all the network and we're only halfway there oh good 00:10:13.320 --> 00:10:20.100 since we're still waiting we might as well take this opportunity to talk about 00:10:16.920 --> 00:10:22.560 one of Super micro's Hyper servers the 00:10:20.100 --> 00:10:28.140 main difference between Cloud DC and Hyper platform servers is that instead 00:10:24.959 --> 00:10:31.860 of one CPU with up to 96 cores you've 00:10:28.140 --> 00:10:33.680 got up to two CPUs with up to 96 cores 00:10:31.860 --> 00:10:41.519 that's right 192 cores that's 00:10:36.980 --> 00:10:44.760 384 threads and up to 12 terabytes of 00:10:41.519 --> 00:10:47.519 ddr5 memory you've also got okay you 00:10:44.760 --> 00:10:54.300 lose some of the PCIe lanes in CPU to CPU connectivity so 160 Lanes of PCIe 00:10:51.959 --> 00:10:59.339 Gen 5 and that means that whether you want to connect 10 flip and gpus to one 00:10:56.760 --> 00:11:04.260 of these things or many many additional nodes of storage devices 00:11:02.040 --> 00:11:07.740 they have got the horses one cool thing is that if you're more 00:11:05.820 --> 00:11:12.240 into compute than expansion you can configure the CPUs to allocate more of 00:11:09.899 --> 00:11:17.040 their PCIe Lanes to socket to socket communication leaving you only 128 Lanes 00:11:15.300 --> 00:11:21.839 but more bandwidth between your sockets 00:11:19.500 --> 00:11:23.820 indeed huh we're gonna try and put this in here 00:11:23.040 --> 00:11:30.959 um hmm hey did it finally boot yeah we're in 00:11:27.959 --> 00:11:33.420 Windows now that was literally over 15 00:11:30.959 --> 00:11:36.660 minutes we need something to put that GPU on because it's sitting a little 00:11:34.800 --> 00:11:39.920 funky what could go wrong huh 00:11:38.399 --> 00:11:45.720 perfect holy crap 00:11:43.019 --> 00:11:50.820 this is one CPU 96 cores I mean I only have 1.85 00:11:49.560 --> 00:11:55.680 gigahertz right now but they'll get faster what we're expecting is somewhere 00:11:53.220 --> 00:11:59.459 in the range of around 2.4 gigahertz when all cores are under load which 00:11:57.420 --> 00:12:04.019 means you actually kind of need to think carefully about your application when 00:12:01.440 --> 00:12:08.760 picking an epic Genoa processor if you want higher clock speeds you may not 00:12:06.480 --> 00:12:12.899 necessarily be better off with the the higher tier higher core chip one cool 00:12:11.459 --> 00:12:17.100 thing by the way about Super micro's Hyper lineup that we're not showing off 00:12:14.820 --> 00:12:22.260 very well is that compared to Cloud DC they are totally tool-less it's pretty 00:12:19.920 --> 00:12:26.940 nice when you got to work on a hundred of them so how will I use my Screwdriver 00:12:24.480 --> 00:12:30.779 from lttstore.com we ready 00:12:28.920 --> 00:12:36.720 boom well that's a disappointment JK JK JK I 00:12:34.980 --> 00:12:40.980 ran it in single core mode I actually just wanted to see what it would clock 00:12:38.040 --> 00:12:44.600 to and it looks like task manager is not going to give us a useful account of 00:12:42.720 --> 00:12:49.800 that so on here 00:12:46.500 --> 00:12:52.620 you don't have HW if I put it I I 00:12:49.800 --> 00:12:56.519 focused on the other server you focused because I had two of them oh he focused 00:12:54.480 --> 00:13:01.019 he says you should be thankful I got it I should be thankful Hardware info has 00:12:59.459 --> 00:13:05.279 either crashed or hung both times I've tried to launch it I'm not blaming them 00:13:02.940 --> 00:13:09.300 this is an extremely new platform I'm just saying we're not going to know 00:13:07.500 --> 00:13:15.079 exactly what frequency that chorus running at 00:13:10.980 --> 00:13:15.079 sorry here we go 00:13:15.240 --> 00:13:23.399 uh oh my goodness 00:13:18.959 --> 00:13:27.120 whoa whoa whoa whoa whoa whoa buddy 00:13:23.399 --> 00:13:29.160 to put it's done that's a 32 core 00:13:27.120 --> 00:13:34.740 threadripper with that tiny little blue bar 75 00:13:31.200 --> 00:13:36.899 000 and this is this is one chip 00:13:34.740 --> 00:13:42.540 now that's not that impressive compared to a threadripper which obviously will 00:13:39.180 --> 00:13:45.779 Turbo quite a bit higher but it's just 00:13:42.540 --> 00:13:48.600 one it's just one we can have 00:13:45.779 --> 00:13:51.899 two think this one's done booting up now is it is it in have we hacked the 00:13:50.519 --> 00:13:57.480 Mainframe are we getting the Windows spinny unfortunately our RTX 49d 00:13:54.600 --> 00:14:01.860 experiment did not work it's not showing up we did manage to get a Quadro GPU 00:14:00.060 --> 00:14:05.880 working earlier but we're not going to break any world records with that so 00:14:03.600 --> 00:14:13.079 we're going to focus on CPU performance for the time being starting of course 00:14:08.639 --> 00:14:15.500 with cinebench but now with two 96 core 00:14:13.079 --> 00:14:21.540 CPUs 384 threads 00:14:18.660 --> 00:14:25.740 and two sockets they completely break the UI 00:14:23.100 --> 00:14:30.660 look at this you have to scroll it's full screen and you have to scroll look 00:14:27.899 --> 00:14:34.500 at the L3 cache oh shoot I had level three Cash on the Mind earlier when I 00:14:32.700 --> 00:14:40.740 said these things had three quarters of a gig of cash no no they've got almost 00:14:36.779 --> 00:14:43.860 an entire Gigabyte of cash the 768 Megs 00:14:40.740 --> 00:14:47.459 is just level three this thing has more 00:14:43.860 --> 00:14:49.920 cash than my first gaming system had RAM 00:14:47.459 --> 00:14:56.459 we've got a problem this is only running 256 of our 00:14:54.199 --> 00:15:01.800 368 threads I didn't think about that but cinebench 00:14:59.279 --> 00:15:07.560 actually has a thread count limit we only got 82 000 points 00:15:04.740 --> 00:15:12.839 now hold on just a gosh darn minute here is there a quick and easy way for us to 00:15:09.839 --> 00:15:14.339 turn off smt oh no oh my God we have to 00:15:12.839 --> 00:15:18.120 reboot we're not rebooting we're gonna find other benchmarks to run time to 00:15:16.680 --> 00:15:22.500 Benchmark pie for those unfamiliar why cruncher 00:15:20.100 --> 00:15:26.639 calculates Pi up to what is this two quadrillion decimal places we have 00:15:24.720 --> 00:15:30.899 absolutely no idea what we're doing which is why it'll be so embarrassing 00:15:28.980 --> 00:15:35.160 when we're at the top of the league enter your choice Benchmark pie one 00:15:33.600 --> 00:15:39.540 multi-threaded okay so we need two and a half billion 00:15:37.680 --> 00:15:45.660 seven or nine no we'll do seven we're going for seven let's go one moment 00:15:42.360 --> 00:15:47.160 please wow it's done and the fans barely 00:15:45.660 --> 00:15:53.399 even kicked in holy crap 00:15:49.639 --> 00:15:55.199 15.783 seconds holy crap this is not 00:15:53.399 --> 00:15:59.699 just faster this is like 00:15:56.579 --> 00:16:02.880 half the time and this record was set 00:15:59.699 --> 00:16:05.100 back in December of 2021 okay how do we 00:16:02.880 --> 00:16:12.240 submit a score I have no idea we're logged into hardware bot Linus Tech 6969 00:16:08.880 --> 00:16:14.579 and wait what invalid data file but 00:16:12.240 --> 00:16:19.320 unable to encrypt the what what I'm able to encrypt what it's a text file what 00:16:17.399 --> 00:16:24.180 does it say in the instructions blah blah blah blah blah blah well I have no 00:16:21.779 --> 00:16:29.279 idea this is not encrypted you guys saw it it's fast this video is wrong past 00:16:27.240 --> 00:16:33.180 Linus we moved the server over to the server room and we are remote into it 00:16:31.199 --> 00:16:37.620 right now using super micro's integrated ipmi KVM and all we're gonna do is Click 00:16:36.240 --> 00:16:43.800 right here and ready ready wait for it 00:16:40.079 --> 00:16:46.980 bloody hell where is it yeah there it is 00:16:43.800 --> 00:16:48.959 go Team Canada not only do we have the 00:16:46.980 --> 00:16:54.899 record in this No Holds Barred competition but we are holding it by 00:16:51.300 --> 00:16:57.300 nearly half the time but one record is 00:16:54.899 --> 00:17:01.199 not enough for us that's why we installed Ubuntu and fired up the 00:16:59.639 --> 00:17:03.919 faronix test Suite where we will be benchmarking I don't know why don't we 00:17:02.940 --> 00:17:11.900 start with x264 ah boss first four okay 00:17:08.660 --> 00:17:15.740 winner dinner 00:17:11.900 --> 00:17:19.980 made of chicken 00:17:15.740 --> 00:17:23.880 uh can chickens here's what we're going 00:17:19.980 --> 00:17:26.640 for the number to beat is 75 frames per 00:17:23.880 --> 00:17:30.179 second average for this test we compare results on open benchmarking.org which 00:17:28.679 --> 00:17:35.160 isn't a competitive site like hardware bot but it does show us the top result 00:17:32.520 --> 00:17:41.280 which was a dual 32 core epic system that managed 75 frames per second plus 00:17:37.919 --> 00:17:43.860 or minus two where are we at here 108 00:17:41.280 --> 00:17:50.760 frames per second see you later record goodbye 00:17:45.200 --> 00:17:54.000 box plot of samples old line new line 00:17:50.760 --> 00:17:56.880 yes I certainly would like to upload 00:17:54.000 --> 00:18:00.720 this now it's time to take a run at John the Ripper which is a password cracking 00:17:59.340 --> 00:18:04.260 benchmark recently saved to enter a name of the 00:18:02.760 --> 00:18:09.360 results file okay dinner fish winner delicious but 00:18:08.160 --> 00:18:17.160 fishy estimated time to completion four 00:18:11.640 --> 00:18:19.100 minutes no unacceptable wait am I 00:18:17.160 --> 00:18:23.400 reading this right 26 million I would like to upload it 00:18:22.679 --> 00:18:28.380 though see you later records previous winner 00:18:26.940 --> 00:18:34.160 fish dinner time to run blender let's go classroom 00:18:34.440 --> 00:18:42.960 let's see if we can cause some trouble on their site here we go estimated time to completion 00:18:39.419 --> 00:18:44.940 two minutes I doubt it 19 seconds is the 00:18:42.960 --> 00:18:51.960 time to beat for classroom oh crap my file name was too long 00:18:48.840 --> 00:18:53.100 um whoops oh crap I definitely broke 00:18:51.960 --> 00:18:59.940 some things um wait 21.9 seconds uncool 00:18:58.320 --> 00:19:03.840 uncool it's not that this system isn't ripping 00:19:01.799 --> 00:19:12.020 fast it's just that clearly someone with another dual 96 core test system got 00:19:06.780 --> 00:19:12.020 here first well it's a long day 00:19:12.299 --> 00:19:14.900 breaking record 00:19:16.620 --> 00:19:25.260 four seems seems pretty good in conclusion epic Genoa really fast super 00:19:23.039 --> 00:19:29.100 micro really cool hardware for epic Genoa check them out with the link down 00:19:27.000 --> 00:19:32.640 below if you guys enjoyed this video I think I probably recorded one of these 00:19:30.539 --> 00:19:36.780 before so let's just cut that back in if you guys enjoyed this video you might 00:19:34.020 --> 00:19:41.220 also enjoy the uh maybe more a structured platform overview that we did 00:19:38.460 --> 00:19:45.740 when AMD announced epic Genoa and you can check that out here here somewhere