1 00:00:00,359 --> 00:00:10,040 there we go 10 gig networking on an M1 2 00:00:03,800 --> 00:00:12,120 Mac Mini that's just wow or well maybe 3 00:00:10,040 --> 00:00:15,599 the thing is just because a port is present on your computer doesn't mean 4 00:00:13,719 --> 00:00:19,840 that it will operate at full speed and it certainly doesn't mean it will 5 00:00:16,960 --> 00:00:24,519 operate at full speed all the time USB for example is rated by the speed that 6 00:00:21,840 --> 00:00:29,679 is shared across multiple ports high-speed IO and SSD slots often share 7 00:00:27,760 --> 00:00:34,239 resources meaning that both of them will slow down if you use them simultaneously 8 00:00:32,160 --> 00:00:39,360 and in the early days of gigabit networking basically no motherboards 9 00:00:36,879 --> 00:00:44,000 could run those ports at full speed because they were connected to the 10 00:00:40,559 --> 00:00:47,239 lethargic PCI bus you know the old 11 00:00:44,000 --> 00:00:49,039 non-express PCI bus of course we only 12 00:00:47,239 --> 00:00:53,079 know these things because vendors like Intel publish block diagrams of their 13 00:00:51,079 --> 00:00:57,680 platforms to help us understand their capabilities apple on the other hand 14 00:00:55,719 --> 00:01:01,840 does No Such Thing meaning that if there are internal bottlenecks the only way to 15 00:01:00,320 --> 00:01:06,960 find out about them is to have your performance drop while you're in the middle of doing something that makes 16 00:01:05,159 --> 00:01:11,799 today's adventure interesting for two reasons one it'll tell us if Apple 17 00:01:09,840 --> 00:01:16,840 delayed this version of the M1 mini perhaps over performance concerns and 18 00:01:14,520 --> 00:01:22,119 two it might give us some fresh new insight into the black box that is 19 00:01:18,840 --> 00:01:22,119 Apple's M1 20 00:01:28,360 --> 00:01:35,680 so the M1 Max Mac Mini blew us away not just with its performance but also with 21 00:01:33,240 --> 00:01:41,280 its price it beat out every previous Intel Mac and single threaded benchmarks 22 00:01:37,880 --> 00:01:44,200 and ayzen 5 5600x one of the fastest 23 00:01:41,280 --> 00:01:48,640 singlethread desktop CPUs in the market all without breaking the bank it didn't 24 00:01:46,560 --> 00:01:54,119 have the cores to back that up in heavily multi-threaded tasks but so far 25 00:01:51,560 --> 00:01:57,799 that hasn't generated many if any user complaints thanks to m1's tight 26 00:01:55,680 --> 00:02:01,960 integration with macOS and of course the availability of the Mac Pro if you 27 00:01:59,399 --> 00:02:08,679 really need more horses what has generated complaints is the io with just 28 00:02:05,560 --> 00:02:12,160 two USBS two Thunderbolt ports and a 29 00:02:08,679 --> 00:02:14,080 single 1 GB Ethernet Jack m1's mobile 30 00:02:12,160 --> 00:02:19,519 pedigree seemed to be holding it back for people like me who love the CPU 31 00:02:16,800 --> 00:02:24,800 performance but need to work with heavy video footage or other shared resources 32 00:02:21,840 --> 00:02:29,959 over a network if one gig wasn't enough you had to use a bulky expensive dongle 33 00:02:27,640 --> 00:02:35,560 and use up one of your Thunderbolt ports or you had to buy into a deadend Intel 34 00:02:32,480 --> 00:02:39,000 platform which was so disappointing for 35 00:02:35,560 --> 00:02:40,959 me personally after seeing Apple push 10 36 00:02:39,000 --> 00:02:45,480 gig forward over the last few years first with the iMac Pro then with the 37 00:02:43,200 --> 00:02:50,800 Mac Mini but it turns out I just didn't have enough faith in my hand right now 38 00:02:47,760 --> 00:02:52,879 is a shiny new 10 gig version of the M1 39 00:02:50,800 --> 00:02:57,959 Mac Mini and we're going to be running I perf on it as a server to give us full 40 00:02:55,159 --> 00:03:02,879 control over what other ports or devices might be hogging PCI Express bandwidth 41 00:03:00,480 --> 00:03:07,799 causing potential bottlenecks so to start we're running an absolute best 42 00:03:05,360 --> 00:03:13,239 case scenario here with nothing plugged into it other than networking and power 43 00:03:10,640 --> 00:03:21,239 not even our display let's come over here and have a look that is dang that 44 00:03:17,879 --> 00:03:25,360 is pretty stable 9.25 45 00:03:21,239 --> 00:03:28,480 9.45 gigabits per second not too shabby 46 00:03:25,360 --> 00:03:30,560 cuz remember guys 10 gbit is the line 47 00:03:28,480 --> 00:03:34,959 speed and it's perfect ly normal for the actual data speed to be a bit lower 48 00:03:33,239 --> 00:03:40,439 what's not normal is the price for our 40 o water bottles on LTT Store.com 49 00:03:37,560 --> 00:03:43,920 they're the same as 21 Oz what a deal now let's up the ante and connect a pro 50 00:03:42,480 --> 00:03:49,760 display xdr but vus you might ask why would 51 00:03:47,799 --> 00:03:52,920 plugging in a monitor affect how much bandwidth is left over for something 52 00:03:51,720 --> 00:03:58,000 like high-speed networking I'm so glad you asked the M1 53 00:03:56,000 --> 00:04:02,840 Mac Mini is notable for being the first mini with DSC or display stream 54 00:04:00,480 --> 00:04:06,879 compression and what that does is it allows it to have enough leftover 55 00:04:04,480 --> 00:04:12,040 bandwidth to run the integrated USB 3 Hub on this monitor at full speed 56 00:04:09,799 --> 00:04:17,600 previous Intel-based Mac minis ended up running either not at the full 6K 57 00:04:14,480 --> 00:04:19,799 resolution or in an uncompressed tiled 58 00:04:17,600 --> 00:04:24,560 mode that some users have reported actually took this Hub and knocked it 59 00:04:21,799 --> 00:04:28,520 all the way back to USB 2 speeds that's because 60 00:04:25,720 --> 00:04:34,000 uncompressed this monitor will suck up about 34 gigabits per second of 61 00:04:31,440 --> 00:04:38,160 bandwidth to be clear I'm not actually expecting this to throttle our network 62 00:04:36,199 --> 00:04:42,320 speed it's just an excellent illustration of how a system has a 63 00:04:40,039 --> 00:04:46,600 finite amount of resources that needs to be carved up also now that we've got our 64 00:04:44,880 --> 00:04:51,680 display up you can see that the terminal on our mini matches our SSH session so 65 00:04:50,199 --> 00:04:56,520 no smoking mirrors here ladies and Gentlemen let's go ahead and again shall 66 00:04:56,919 --> 00:05:05,720 we and add as expected we are exactly 67 00:05:02,639 --> 00:05:07,120 the same I mean no computer manufacturer 68 00:05:05,720 --> 00:05:11,800 in their right mind would rate the speed of their ports assuming that no display 69 00:05:09,280 --> 00:05:16,919 is plugged in especially not a display they make 70 00:05:13,840 --> 00:05:19,639 right but we're going to kick things up 71 00:05:16,919 --> 00:05:24,199 a notch and to do so Anthony repurposed some of a shell script that he wrote 72 00:05:21,319 --> 00:05:28,680 previously that constantly writes data to an SSD and then spits out speed info 73 00:05:27,039 --> 00:05:35,600 at regular intervals for graphing purposes and and he targeted it at this 74 00:05:31,800 --> 00:05:38,160 high performance Thunderbolt SSD maybe 75 00:05:35,600 --> 00:05:43,840 with this thing running we'll finally start to see some bandwidth sharing in 76 00:05:40,319 --> 00:05:45,680 our I perf test all right Apple all 77 00:05:43,840 --> 00:05:50,759 right I'm impressed our SSD right now is pulling 78 00:05:48,560 --> 00:05:55,360 over 10 gbits per second in reads putting us darn close to the up to 40 79 00:05:54,000 --> 00:06:02,479 gbits per second number that's advertised for these ports but all that tells us so far is 80 00:06:00,680 --> 00:06:08,240 that the two Thunderbolt ports don't share bandwidth so let's continue 81 00:06:06,440 --> 00:06:13,880 remember the USB hub I mentioned on the back of the pro display xdr well we're 82 00:06:10,720 --> 00:06:16,319 going to load it up with USBC ssds we're 83 00:06:13,880 --> 00:06:20,160 going to run that SSD script on all three of these at the same time then 84 00:06:18,560 --> 00:06:29,199 we're going to run our Network test and see if we can make this poor Mac Mini 85 00:06:23,039 --> 00:06:31,800 cry and good news sort of our ssds did 86 00:06:29,199 --> 00:06:35,599 slow down that means we managed to find a 87 00:06:32,800 --> 00:06:39,919 bottleneck but it's in the USB hub of the pro display xdr which present us 88 00:06:37,479 --> 00:06:43,479 with a couple of problems number one that doesn't actually tell us anything 89 00:06:41,560 --> 00:06:50,280 about internal bottlenecks of the M1 Mac Mini and number two is we are fresh out 90 00:06:47,240 --> 00:06:52,280 of USBC ports to plug ssds 91 00:06:50,280 --> 00:07:02,000 into unless a Thunderbolt dock ah yes we need 92 00:06:57,160 --> 00:07:02,000 a Thunderbolt Dock and and more 93 00:07:02,440 --> 00:07:06,560 ssds we're going to plug all these in 94 00:07:05,440 --> 00:07:14,639 and then we're going to run the script on them and then see how you like that 95 00:07:09,720 --> 00:07:17,680 there we go five ssds plugged into the 96 00:07:14,639 --> 00:07:20,440 Hub and then for good measure an 97 00:07:17,680 --> 00:07:26,039 additional one another Thunderbolt 3 one plugged into the Thunderbolt daisy 98 00:07:23,039 --> 00:07:28,800 chain guys if that 10 gig Ethernet is 99 00:07:26,039 --> 00:07:35,160 stealing bandwidth from somewhere we are going to find it huh our pre-planning 100 00:07:32,720 --> 00:07:40,840 did not have this actually go down at all we figured it out one of the discs 101 00:07:38,599 --> 00:07:47,319 was not showing up properly and running the command to a missing Target was 102 00:07:44,159 --> 00:07:50,520 causing something um I don't know maybe 103 00:07:47,319 --> 00:07:52,759 some kind of CPU overhead anyway now 104 00:07:50,520 --> 00:07:57,720 that all the discs are showing up we are getting the results we expected which is 105 00:07:54,360 --> 00:08:00,440 anywhere from 9.3 to 9.42 gbit per 106 00:07:57,720 --> 00:08:06,360 second and this is even with all of the drives running the CPU usage while we're 107 00:08:03,440 --> 00:08:10,599 just doing this is around 18% and that's with the all the drives plugged in if I 108 00:08:07,919 --> 00:08:14,280 add that drive that's not plugged iner what's our CPU usage at now 109 00:08:16,720 --> 00:08:23,240 32 36 significantly higher 110 00:08:23,280 --> 00:08:30,800 interesting I didn't test this on the bench I didn't even expect this okay 111 00:08:28,080 --> 00:08:36,880 it's gone 40 112 00:08:32,599 --> 00:08:38,440 77 okay wow way to way to write a script 113 00:08:36,880 --> 00:08:43,719 that hurts computers when you do it wrong but while 114 00:08:41,240 --> 00:08:49,040 our side Adventure was interesting you know hey turns out runaway CPU usage is 115 00:08:46,640 --> 00:08:53,040 going to affect network transfer performance it wasn't really the point 116 00:08:50,839 --> 00:08:57,200 of the video today so let's refocus and talk about what we've learned so far 117 00:08:54,560 --> 00:09:03,040 there's a few things actually so one we now know that the 10 gig M1 is legit 118 00:08:59,440 --> 00:09:05,640 legit nonshared 10 gig no switch chips 119 00:09:03,040 --> 00:09:10,680 no controller hubs just a whole Gen 4 PCI Express Lane That's why nothing that 120 00:09:08,279 --> 00:09:14,600 we did was able to bottleneck it so clearly Apple's delay had nothing to do 121 00:09:12,640 --> 00:09:21,079 with performance and you can buy this puppy with confidence that's cool but 122 00:09:17,959 --> 00:09:23,000 we're not done diving deeper while our 123 00:09:21,079 --> 00:09:27,880 theoretical Max Speed when testing the drives individually works out to just 124 00:09:24,720 --> 00:09:30,560 over 58 gbits per second it capped out 125 00:09:27,880 --> 00:09:35,240 somewhere around 50 in the real world and that means that we did throttle 126 00:09:32,760 --> 00:09:39,279 somewhere the ssds connected to the pro display xdr were obviously going to 127 00:09:37,360 --> 00:09:45,399 throttle as they were being crowded by that chunky 13.7 GB per second display 128 00:09:42,120 --> 00:09:48,399 port signal but the Thunderbolt 3 hubs 129 00:09:45,399 --> 00:09:50,279 throttling is another story and this one 130 00:09:48,399 --> 00:09:56,000 comes primarily down to the fact that each Thunderbolt 3 Port appears to be 131 00:09:52,360 --> 00:09:59,839 capping out at around 20 to 22 gbits per 132 00:09:56,000 --> 00:10:02,240 second but hang on a second lonus said 133 00:09:59,839 --> 00:10:08,120 each of these ports could do up to 40 gbit per second and you said they didn't 134 00:10:05,399 --> 00:10:13,519 share bandwidth what gives okay so remember how our 10 gig 135 00:10:11,399 --> 00:10:17,959 Nick is expected to be less than 10 gbit per second in the real world well the 136 00:10:15,519 --> 00:10:24,160 same is actually true for Thunderbolt 3 USB 4 whatever the hell this thing is so 137 00:10:21,120 --> 00:10:26,640 around 7.6 gbit per second of that is 138 00:10:24,160 --> 00:10:30,760 set aside for DisplayPort which our Thunderbolt 3 doc has and then the rest 139 00:10:29,399 --> 00:10:37,120 of it gets eaten up by the signal encoding bringing us down to a theoretical maximum of only around 26 140 00:10:34,640 --> 00:10:39,959 gbits per second and that's close enough to our numbers that the difference 141 00:10:38,560 --> 00:10:44,480 probably comes down to the fact that we're going through busy hubs rather 142 00:10:42,120 --> 00:10:48,399 than attaching directly hence the wavy patterns on the graphs compared to the 143 00:10:46,120 --> 00:10:53,160 individual Drive results that means that if you've got applications that are 144 00:10:49,880 --> 00:10:55,360 sensitive to IO variants like audio or 145 00:10:53,160 --> 00:11:00,760 anything involving real-time data you would do best to directly attach or use 146 00:10:58,600 --> 00:11:05,720 only one high-speed device on a hub or dock at a time of course for most people 147 00:11:04,040 --> 00:11:12,399 this is academic and it won't make a difference in the real world well sort 148 00:11:08,480 --> 00:11:14,440 of because M1 is still IO limited and it 149 00:11:12,399 --> 00:11:18,040 has affected the port options on Apple's M1 products like for example the new 150 00:11:16,440 --> 00:11:24,680 iMac which we'll be reviewing so make sure you're subscribed the question just 151 00:11:21,000 --> 00:11:27,399 becomes how limited is it what it looks 152 00:11:24,680 --> 00:11:29,800 like is we've got two PCI Express Gen 4 Lanes going to Apple's Thunderbolt 153 00:11:28,920 --> 00:11:36,959 control which Steeps down to two Thunderbolt 3 channels then we've got one gen 4 Lane 154 00:11:34,200 --> 00:11:41,639 for the 10 gig Nick there's 10 gbits to go around for the type A ports so that's 155 00:11:39,680 --> 00:11:46,200 another Lane that is maybe shared with the Wi-Fi 6 adapter and then there's an 156 00:11:43,839 --> 00:11:50,959 HDMI port connected directly to the GPU via an internal DisplayPort to HDMI 157 00:11:48,480 --> 00:11:56,079 converter that's your IO four lanes accounted for we ran the io terminal 158 00:11:53,279 --> 00:12:00,000 command to try and find out for sure but while it does show device hierarchies it 159 00:11:58,200 --> 00:12:05,600 doesn't show how they're connected to the main system just that they are 160 00:12:02,320 --> 00:12:08,959 connected of course most PC systems have 161 00:12:05,600 --> 00:12:13,320 a lot more than four PCI Express Lanes 162 00:12:08,959 --> 00:12:15,880 surely M1 does as well then right well 163 00:12:13,320 --> 00:12:22,240 not necessarily you might think it would have another four lanes at least for its 164 00:12:18,000 --> 00:12:24,160 SSD but it doesn't remember how the t2 165 00:12:22,240 --> 00:12:28,600 chip that we loved so much was the storage controller for the final 166 00:12:25,519 --> 00:12:31,000 generation of Intel Max and then it just 167 00:12:28,600 --> 00:12:34,320 disappeared with the M1 well the reverse engineering effort for getting M1 168 00:12:32,480 --> 00:12:40,160 running on Linux has actually discovered that Apple wasn't kidding the M1 so has 169 00:12:37,600 --> 00:12:44,519 its security chip built right into it and it's controlling the SSD directly 170 00:12:42,639 --> 00:12:51,720 through some kind of memory map Arrangement not through PCI Express and 171 00:12:48,079 --> 00:12:54,160 you got to remember NVMe is a protocol 172 00:12:51,720 --> 00:12:59,560 so while we associate it with PCI Express it doesn't technically have to 173 00:12:56,800 --> 00:13:03,720 run over PCI Express all of which is to say then that the M1 Mac Mini is 174 00:13:01,760 --> 00:13:08,519 probably near the limits of what Apple can manage with the 175 00:13:05,160 --> 00:13:10,240 m1c but it's got far more IO than we 176 00:13:08,519 --> 00:13:13,600 initially gave it credit for cuz remember we haven't even accounted yet 177 00:13:12,000 --> 00:13:18,120 for whatever bus they're using for the internal GPU so the reason they held off 178 00:13:15,880 --> 00:13:23,920 on the 10 gig version was probably more to do with Logistics or strategy but 179 00:13:21,519 --> 00:13:29,480 whatever it was Apple has done a lot here with little and that's pretty 180 00:13:27,360 --> 00:13:33,680 admirable kind of like how Apple used to be back in the good old days and kind of 181 00:13:31,560 --> 00:13:37,519 like my Segways to sponsors like Microcenter get the best prices and best 182 00:13:35,920 --> 00:13:41,720 selection on computer hardware and everything else technology at any one of 183 00:13:39,399 --> 00:13:45,959 microcenter's 25 locations across the United States microcenter's custom PC 184 00:13:44,320 --> 00:13:50,000 Builder will help you spec out the best PC for your budget by making sure all 185 00:13:48,000 --> 00:13:53,399 your parts are compatible finding stock that's available at your nearest Micro 186 00:13:51,399 --> 00:13:56,959 Center location making it so you can just add it to your cart and arrange 187 00:13:55,120 --> 00:14:00,160 same day in store pickup for a fee you can check the box marked same day Pro 188 00:13:58,560 --> 00:14:03,920 assembly to have a Microcenter expert technician assemble the PC for you and 189 00:14:02,279 --> 00:14:07,759 if you want help deciding what parts to put in your new custom gaming PC just 190 00:14:05,920 --> 00:14:11,480 join the new online Microcenter Community it's a great place to talk 191 00:14:09,399 --> 00:14:14,519 Tech with other enthusiasts check out the link in the description for a free 192 00:14:12,959 --> 00:14:19,040 pair of wireless bluetooth headphones valid instore only no purchase necessary 193 00:14:17,399 --> 00:14:23,399 so thanks for watching guys check out our review of the M1 Mac Mini when it 194 00:14:20,959 --> 00:14:27,480 first launched to get just a little bit more of an appreciation for how much of 195 00:14:25,199 --> 00:14:31,040 a game changer this modest piece of silicon could be