1 00:00:00,800 --> 00:00:04,760 Zeon W, 2 00:00:05,120 --> 00:00:12,320 Zeon scalable. 3 00:00:08,240 --> 00:00:16,240 You know, they're both called Xeon, 4 00:00:12,320 --> 00:00:20,640 but these things are really different. 5 00:00:16,240 --> 00:00:24,720 One of them is basically a Core i9 with 6 00:00:20,640 --> 00:00:29,039 ECC memory support and the other one is 7 00:00:24,720 --> 00:00:32,079 a server CPU that I fanled all over 8 00:00:29,039 --> 00:00:34,480 because I love super high-end expensive 9 00:00:32,079 --> 00:00:43,360 tech toys. Now, in the past, you needed multiple 10 00:00:38,480 --> 00:00:45,520 CPUs in one multi-socketed motherboard 11 00:00:43,360 --> 00:00:52,480 in order to handle intensive multi-threaded workloads. But is that 12 00:00:48,960 --> 00:00:55,920 still the case today? Do you still need 13 00:00:52,480 --> 00:01:02,960 two of these given that a single Xeon 14 00:00:55,920 --> 00:01:06,400 Platinum 8180 is 28 cores and 56 threads 15 00:01:02,960 --> 00:01:09,840 on a single chip? Well, I don't know 16 00:01:06,400 --> 00:01:12,640 what is the purpose today of a dual 17 00:01:09,840 --> 00:01:18,320 socket machine like this one and how much have single high core count CPUs 18 00:01:16,000 --> 00:01:31,119 eroded the market that they used to enjoy? Let's find out, shall we? 19 00:01:31,119 --> 00:01:37,200 All right, there's a lot more room down here and we are going to need it for 20 00:01:35,280 --> 00:01:44,880 this honking, not to mention heavy test bench. On this 21 00:01:40,479 --> 00:01:48,880 test bench, you will find the ASUS C621E 22 00:01:44,880 --> 00:01:54,960 Sage. This is a dual socket motherboard 23 00:01:48,880 --> 00:01:59,439 rocking two LGA 3647 sockets for Intel's 24 00:01:54,960 --> 00:02:02,159 Xeon scalable lineup of CPUs. And setups 25 00:01:59,439 --> 00:02:09,200 like this have actually been around as far back as the 486 in 1989 26 00:02:06,880 --> 00:02:14,400 with the resulting secondhand hardware giving enthusiasts the ability to get 27 00:02:11,360 --> 00:02:16,640 multiple physical cores in their homes 28 00:02:14,400 --> 00:02:25,520 over the years with the peak being somewhere in the mid 2000s or so. But 29 00:02:21,680 --> 00:02:28,879 that was then and this is now. Now you 30 00:02:25,520 --> 00:02:32,400 can get multiple processing cores in a 31 00:02:28,879 --> 00:02:35,200 single chip. So 32 00:02:32,400 --> 00:02:41,760 to see how far things have come, what we're going to do is pit this machine 33 00:02:38,400 --> 00:02:44,239 against the fastest single CPU that 34 00:02:41,760 --> 00:02:48,879 we've tested to date. We're going to try to keep the number of variables to a 35 00:02:46,400 --> 00:02:54,160 minimum in order to gauge the impact that these extra CPU cores will have on 36 00:02:51,599 --> 00:02:59,680 our setup. Though, it should be noted that there aren't many options when it 37 00:02:56,000 --> 00:03:02,720 comes to aftermarket LGA3647 38 00:02:59,680 --> 00:03:05,360 coolers because most of the folks 39 00:03:02,720 --> 00:03:09,760 selling these kinds of systems would figure out their own solution. So, that 40 00:03:07,120 --> 00:03:13,599 means that our dual socket workstation will run a little bit toasty, but we 41 00:03:11,840 --> 00:03:18,080 didn't observe any thermal throttling, so it shouldn't affect our performance. 42 00:03:16,159 --> 00:03:22,640 Let's start off then with good oldfashioned 43 00:03:19,840 --> 00:03:30,360 Cinebench. I mean, we've we've seen this run before, but it's always fun to see 44 00:03:25,519 --> 00:03:30,360 it finish that quickly. 45 00:03:31,120 --> 00:03:38,239 So, in a surprise to no one, the dual 46 00:03:34,319 --> 00:03:41,760 socket machine is faster. But 47 00:03:38,239 --> 00:03:44,720 considering its 56 48 00:03:41,760 --> 00:03:51,599 processing cores, not all of our workloads scale in the way that we might 49 00:03:48,159 --> 00:03:53,599 expect. 7zip, for example, shows a 50 00:03:51,599 --> 00:04:00,959 smaller thanex expected gain over our Core i9 Extreme Edition, and YC Cruncher 51 00:03:57,200 --> 00:04:03,599 even finds itself losing ground. ASUS 52 00:04:00,959 --> 00:04:08,879 Realbench demonstrates this, though, with that said, the encoding benchmark 53 00:04:05,519 --> 00:04:11,439 ees out a lead over our Core i9 7980XE. 54 00:04:08,879 --> 00:04:17,359 and then Blender. Well, here we actually get a victory for our dual CPU system 55 00:04:14,239 --> 00:04:21,120 again, showing this platform's potential 56 00:04:17,359 --> 00:04:23,199 for expanding render farms. 57 00:04:21,120 --> 00:04:29,759 But what is really going on here? Well, 58 00:04:27,600 --> 00:04:37,440 something you guys have to realize is that there is more to a dual socket 59 00:04:33,199 --> 00:04:40,960 configuration than just more cores. Do 60 00:04:37,440 --> 00:04:44,560 you remember when AMD managed a 3% 61 00:04:40,960 --> 00:04:47,919 improvement in IPC with second gen Ryzen 62 00:04:44,560 --> 00:04:50,880 just by improving cache latency? So on 63 00:04:47,919 --> 00:04:59,280 this motherboard, we've got two separate CPUs with two separate sets of cache and 64 00:04:56,160 --> 00:05:02,160 memory. See, these six banks go to this 65 00:04:59,280 --> 00:05:07,919 one and these six banks are wired into this one. And that means a lot of 66 00:05:04,880 --> 00:05:11,280 latency for compute tasks that require 67 00:05:07,919 --> 00:05:13,039 the same data sets. This latency is a 68 00:05:11,280 --> 00:05:18,880 necessary evil in the design of multiprocessor systems because of the 69 00:05:15,520 --> 00:05:22,240 need for nonuniform memory access or 70 00:05:18,880 --> 00:05:24,000 NUMA for short that allows these two 71 00:05:22,240 --> 00:05:29,520 processors to efficiently share resources or as efficiently as they can. 72 00:05:27,440 --> 00:05:34,560 So the short version of this is that it works by transparently allocating 73 00:05:31,759 --> 00:05:38,880 devices and memory to each CPU which means they can more easily avoid 74 00:05:36,479 --> 00:05:43,600 interrupting each other while accessing those resources. This in turn reduces 75 00:05:41,600 --> 00:05:48,639 the amount of waiting around that they have to do for those resources to become 76 00:05:45,840 --> 00:05:54,639 available. So that's what we're seeing during our testing like in Yunchruncher 77 00:05:51,360 --> 00:05:57,919 for example where both CPUs are working 78 00:05:54,639 --> 00:05:59,520 on the same data but it's not really the 79 00:05:57,919 --> 00:06:05,759 intended use case for this kind of thing. What if we could use different 80 00:06:02,960 --> 00:06:10,800 data sets? Then we should be able to find this kind of setups true calling. 81 00:06:08,800 --> 00:06:17,280 And how better to do that than to effectively turn this system into two 82 00:06:14,319 --> 00:06:23,600 independent computing machines using virtualization. So let's fire up on 83 00:06:20,240 --> 00:06:26,160 which uses Red Hat KVM as a hypervisor 84 00:06:23,600 --> 00:06:33,280 to see what kind of results we get splitting these resources into multiple 85 00:06:29,759 --> 00:06:36,880 independent machines. Immediately, we 86 00:06:33,280 --> 00:06:40,080 see worse results from our VMs than our 87 00:06:36,880 --> 00:06:45,120 original 56 core testing. 88 00:06:40,080 --> 00:06:49,120 But look closer at how much lower it is. 89 00:06:45,120 --> 00:06:52,000 It's not a whole lot. In every test, 90 00:06:49,120 --> 00:06:57,759 it's basically the same story here. And we are still way out ahead of the Core 91 00:06:55,039 --> 00:07:04,479 i9 Extreme Edition, particularly when it comes to Blender. Now if we consider the 92 00:07:01,360 --> 00:07:07,520 fact that we are getting simultaneous 93 00:07:04,479 --> 00:07:10,240 work done that gives us a good look at 94 00:07:07,520 --> 00:07:16,240 what an optimized workload might look like. I mean or heck uh virtualization 95 00:07:13,039 --> 00:07:20,080 itself is a legitimate task too. I mean 96 00:07:16,240 --> 00:07:22,240 this thing could be so many gamers in 97 00:07:20,080 --> 00:07:26,960 one PC. But I digress. I mean, nobody is going 98 00:07:24,720 --> 00:07:31,360 to buy something like this for their personal rig anytime soon, given the 99 00:07:29,680 --> 00:07:39,280 $10,000 per CPU price tag, which puts it 100 00:07:35,280 --> 00:07:41,039 squarely in the territory of big the 101 00:07:39,280 --> 00:07:47,520 size of the check doesn't matter business. What they care about is 102 00:07:43,840 --> 00:07:50,080 density. The more processing power a 103 00:07:47,520 --> 00:07:55,840 single computer can manage, the more processing power that can be physically 104 00:07:52,880 --> 00:08:01,280 fit into a building. And this is perhaps most important for data centers and 105 00:07:58,800 --> 00:08:07,199 render farms in particular. The less those guys have to spend on setting up 106 00:08:03,919 --> 00:08:10,240 the electricity and cooling management 107 00:08:07,199 --> 00:08:14,479 for a data center versus the amount of 108 00:08:10,240 --> 00:08:16,160 performance they can get, the better. 109 00:08:14,479 --> 00:08:20,800 So, will multiple sockets make a comeback in 110 00:08:18,400 --> 00:08:26,319 the proumer space? Outside of, you know, oil and gas 111 00:08:23,440 --> 00:08:31,039 exploration where there are still workloads that can benefit from this 112 00:08:28,240 --> 00:08:36,959 kind of thing, the chances look pretty slim if you ask me. But I don't 113 00:08:35,039 --> 00:08:42,880 necessarily think that it's Intel's intent to sell these chips in the 114 00:08:39,279 --> 00:08:45,519 proumer space. And for that matter, even 115 00:08:42,880 --> 00:08:51,279 in the enterprise space, I don't think they move a ton of them. For me, I look 116 00:08:48,800 --> 00:08:57,200 at a product like this as more of like a a future crafting exercise where it is 117 00:08:54,480 --> 00:09:02,880 available today, but it's more of a representation of what actually might be 118 00:08:59,839 --> 00:09:05,440 attainable a generation or two from now, 119 00:09:02,880 --> 00:09:10,800 just like the 22 core processors that we were playing around with a couple of 120 00:09:07,040 --> 00:09:13,519 years ago. Nowadays, those are much more 121 00:09:10,800 --> 00:09:18,640 affordable and businesses are using them to power your cloud computing services. 122 00:09:17,120 --> 00:09:22,640 So, thanks for watching, guys. If this video sucked, you know what to do. But 123 00:09:20,640 --> 00:09:26,000 if it was awesome, get subscribed, hit that like button, or check out the link 124 00:09:24,320 --> 00:09:29,120 to where to buy the stuff we featured in the video description. Also, linked down 125 00:09:27,839 --> 00:09:33,760 there is our merch store, which has cool shirts like this one, and our community 126 00:09:31,040 --> 00:09:41,360 forum, which you should totally join. Now I'm off to finally put this to use 127 00:09:37,440 --> 00:09:46,760 for the reason that I obtained it. 128 00:09:41,360 --> 00:09:46,760 Many video editors, one CPU.