WEBVTT 00:00:22.000 --> 00:00:28.720 so thanks for watching guys don't forget to drop a like on a minute 00:00:25.920 --> 00:00:32.800 you can't just do that like we gotta actually talk about this thing 00:00:31.519 --> 00:00:39.280 well i showed them all they needed to know yeah it's faster and that 00:00:36.239 --> 00:00:41.840 okay yes that was pretty definitive but 00:00:39.280 --> 00:00:46.079 there is way more to the story here guys we're gonna give you the rundown on 00:00:43.120 --> 00:00:52.320 AMD's threadripper 3990x the first desktop 64 core processor 00:00:51.280 --> 00:00:56.160 and we're going to give you the rundown on our sponsor glassware instantly see 00:00:54.480 --> 00:01:00.000 your current and past network activity detect malware and block badly behaving 00:00:58.160 --> 00:01:05.450 apps on your pc or Android device with glass wire use offer code Linus to get 00:01:01.680 --> 00:01:10.439 25 off at the link below 00:01:13.439 --> 00:01:20.799 so AMD's 32 core threadripper 3970x 00:01:17.840 --> 00:01:25.360 already beats the pants off Intel's top-end core i9 and xeon products so the 00:01:23.520 --> 00:01:28.720 first question we had to ask ourselves here was 00:01:26.960 --> 00:01:33.920 given that they have no competition in this segment already 00:01:31.600 --> 00:01:38.479 why is AMD even bothering to launch this product yeah it's overkill in pretty 00:01:36.560 --> 00:01:45.200 much every possible way 64 cores 128 threads 288 megabytes of 00:01:42.960 --> 00:01:49.040 cash and official support for up to 3 200 megahertz memory with single rank 8 00:01:47.600 --> 00:01:55.040 Gigabyte sticks even if that doesn't carry over to all the higher capacities and the craziest 00:01:52.880 --> 00:01:59.680 thing about it is that the normal rule where high core count cpus have poor 00:01:57.600 --> 00:02:03.360 performance in lightly threaded applications 00:02:01.280 --> 00:02:06.479 just doesn't seem to apply here we've got about a four percent drop in 00:02:05.040 --> 00:02:12.879 our single thread cinebench score compared to our ryzen 9 3950x that's 00:02:09.679 --> 00:02:15.200 AMD's highest boosting desktop chip 00:02:12.879 --> 00:02:20.400 so from a business perspective one of the obvious concerns when you go all out 00:02:17.840 --> 00:02:24.879 on a desktop processor is that even though this thing is still pricey at 4 00:02:23.120 --> 00:02:31.360 000 us dollars this is essentially the exact same 00:02:27.599 --> 00:02:35.360 silicon as their 64 core epic server 00:02:31.360 --> 00:02:38.480 processor only you get 88 instead of 128 00:02:35.360 --> 00:02:41.599 pci express gen 4 lanes and four of the 00:02:38.480 --> 00:02:44.640 eight memory channels are disabled so 00:02:41.599 --> 00:02:47.920 that's thousands of dollars less for 00:02:44.640 --> 00:02:51.040 what is essentially still 64 cores 00:02:47.920 --> 00:02:52.400 so there's this risk then of competing 00:02:51.040 --> 00:02:57.760 against yourself by giving customers the option of 00:02:55.599 --> 00:03:04.400 getting most of the benefit of your professional grade product with a 00:03:00.400 --> 00:03:06.879 cheaper one yeah cheaper the only two 00:03:04.400 --> 00:03:11.200 possible answers to me are that the they're doing it out of uh share 00:03:08.560 --> 00:03:15.599 vindictiveness to embarrass Intel or it's more of a marketing exercise to use 00:03:13.599 --> 00:03:19.760 their technological advantage right now to pump their stock price relieve some 00:03:17.599 --> 00:03:25.920 of their debt and grab some important mind share the halo effect works yeah 00:03:23.360 --> 00:03:29.280 see top end product it performs better slightly lower in product i can afford 00:03:27.519 --> 00:03:33.280 oh it's probably better blur studio famously rendered terminator dark fate 00:03:31.280 --> 00:03:37.519 on 3rd gen threadripper and giving studios like that twice as many cores is 00:03:35.599 --> 00:03:42.000 probably going to result in a lot more headlines like that that's great 00:03:39.360 --> 00:03:46.400 marketing and besides the thing is epic is still going to be the best tool 00:03:44.560 --> 00:03:51.760 for certain things it's the only way to get support for two 00:03:48.879 --> 00:03:56.959 zen 2 cpus on a single motherboard and as i discovered with our new NVMe server 00:03:54.400 --> 00:04:01.280 both pci express lanes and memory bandwidth can quickly become bottlenecks 00:03:59.040 --> 00:04:06.799 when you're throwing enough data around like in a high-speed storage server now 00:04:03.519 --> 00:04:08.560 to be clear you would never run that 00:04:06.799 --> 00:04:11.599 many drives on a threadripper workstation so i wouldn't expect it to 00:04:10.400 --> 00:04:14.799 be a problem it's all about using the right tool for 00:04:13.599 --> 00:04:18.959 the job and this is gonna be the right tool for 00:04:17.440 --> 00:04:26.880 a lot of jobs it's got a massive 280 watt TDP that's 00:04:21.840 --> 00:04:28.560 25 higher than the 64 core epic 7742 00:04:26.880 --> 00:04:32.160 but it results in a nearly thirty percent higher base clock and a max 00:04:30.800 --> 00:04:36.560 boost as high as four point three gigahertz way higher than epic this 00:04:34.479 --> 00:04:40.800 thing is going to scream but like in what i mean cinebench yeah but do we 00:04:38.960 --> 00:04:45.600 actually have anything that can properly stress a 64 core CPU like look at this 00:04:44.080 --> 00:04:48.960 is that not ridiculous like just look at it 00:04:47.120 --> 00:04:52.880 nobody's ever seen anything like that yeah that's a bit of a problem with 00:04:51.280 --> 00:04:57.199 Windows splitting threads into groups of 64 applications have to explicitly add 00:04:55.600 --> 00:05:02.080 support for spanning multiple thread groups in order to use all 128 threads 00:04:59.600 --> 00:05:05.600 on the 3990x at once that's not a major deal breaker for specialized workloads 00:05:03.919 --> 00:05:10.080 designed with multi-socket machines and massive core counts in mind but it may 00:05:08.240 --> 00:05:15.520 limit us when we try to push it just like we did with our epic 7742 now AMD 00:05:13.360 --> 00:05:19.280 told us that we should expect the best bang for buck with workloads that can 00:05:17.199 --> 00:05:22.800 utilize one to two gigabytes of RAM per thread typical benchmarks like cinebench 00:05:21.280 --> 00:05:26.479 here are so short that they don't give you a clear idea of how much time you 00:05:24.479 --> 00:05:30.240 can end up saving on a large-scale project okay so that was a lot of 00:05:28.080 --> 00:05:34.320 talking but what i'm hearing is we don't have anything 00:05:34.320 --> 00:05:40.400 well one suggestion they had was we could compile a large program like a web 00:05:38.320 --> 00:05:45.120 browser okay why don't we throw what we have at it uh we don't have to let this 00:05:42.400 --> 00:05:47.120 single threaded uh cinebench r20 run no i already know what the answer is yeah i 00:05:46.400 --> 00:05:51.919 know it was about 500 500 exactly 500 okay 00:05:50.479 --> 00:05:56.000 and this thing's impressive and like look at this it's turboing up to i saw 00:05:53.759 --> 00:06:00.479 it go as high as around 4.2 in this load sitting around the 4.05 00:05:58.639 --> 00:06:03.919 4.15 range like 00:06:02.080 --> 00:06:10.160 it's great i love it already so review over then 00:06:07.199 --> 00:06:14.639 um no no okay so let's let's see if it could possibly justify its four thousand 00:06:12.240 --> 00:06:19.360 dollar price all right it's fired up holy crap i have never seen this run 00:06:17.759 --> 00:06:23.520 like this uh yeah no this is supposed to take like 00:06:22.160 --> 00:06:29.520 10 minutes plus we are turboing to three gigahertz on 64 00:06:27.680 --> 00:06:34.720 cores what kind of heat is this thing kicking off it's not even that crazy 00:06:32.319 --> 00:06:38.240 no no it's not even the heat pipes are only kind of warm and it is kind of cool 00:06:36.639 --> 00:06:43.840 here in the warehouse but yeah that's true that's an advantage the craziest thing 00:06:41.280 --> 00:06:47.680 about this is that this benchmark takes two and a half three minutes to even 00:06:46.160 --> 00:06:53.039 start running because it has to go fetch a bunch of resources and it's largely a 00:06:50.080 --> 00:06:58.240 single or two threaded task that means we've only actually been running this 00:06:55.039 --> 00:06:59.440 for about a minute and a half and it's 00:06:58.240 --> 00:07:02.560 half done you know what that might be 00:07:01.280 --> 00:07:06.720 remember how you said like oh doesn't this normally take like a minute a minute and a half 00:07:04.880 --> 00:07:11.440 and it took much longer yeah it's hammering the SSD 00:07:09.280 --> 00:07:15.680 okay we'll talk about that more later should have gone optane oh 00:07:16.479 --> 00:07:23.199 well AMD didn't recommend opting well AMD 00:07:20.479 --> 00:07:27.039 has a [ __ ] agenda i think what i'm gonna have to do is 00:07:24.560 --> 00:07:32.960 time it after the resources have loaded that's the only fair thing to do between 00:07:30.720 --> 00:07:37.120 no i think it's fair for you know if your bullet it doesn't work 00:07:34.560 --> 00:07:40.560 because of other system bottlenecks and you didn't work around that then 00:07:39.360 --> 00:07:44.319 you should have to run the benchmark same way as everybody else i mean we 00:07:42.400 --> 00:07:49.680 could show it in both we could show both ways holy crap 00:07:47.039 --> 00:07:53.360 so from resource fetching being finished to completing this render 00:07:52.000 --> 00:07:59.120 was under three minutes no about three minutes yeah that's 00:07:56.720 --> 00:08:05.199 that's more like titan rtx with the new ray tracing renderer 00:08:00.800 --> 00:08:07.360 running on CPU yeah unreal now let's hit 00:08:05.199 --> 00:08:13.120 it with v-ray more like hit v-ray with it 00:08:10.720 --> 00:08:16.960 yeah look at all those cores light up yeah the crazy thing about watching this 00:08:16.240 --> 00:08:22.240 is it almost looks like a GPU demo yeah 00:08:20.000 --> 00:08:26.400 like how how quickly the image 00:08:23.440 --> 00:08:31.440 goes from totally crappy to like okay it's starting to look photorealistic 00:08:28.960 --> 00:08:35.279 it's not it's not GPU performance but it's 00:08:32.479 --> 00:08:38.560 it's also not CPU performance it's like it's some weird in between just because 00:08:36.640 --> 00:08:42.880 of so many threads like that's kind of the thing with gpus they 00:08:40.560 --> 00:08:46.640 have a lot of compute units that can crunch on little kind of like sub thread 00:08:45.279 --> 00:08:50.959 lit things but like this this is just 00:08:49.279 --> 00:08:56.480 straight up threads wow 00:08:53.360 --> 00:08:58.200 72 752. 00:08:56.480 --> 00:09:03.120 how did our 3970x do uh 00:09:01.040 --> 00:09:08.720 44643 now it's not perfect scaling 00:09:05.440 --> 00:09:13.279 but that is pretty darn good 00:09:08.720 --> 00:09:15.279 yeah and our Intel bench like 36 000. 00:09:13.279 --> 00:09:20.000 NVIDIA except not except actually just running on a CPU 00:09:18.000 --> 00:09:24.200 so what you're looking at right now is crisis the original crisis 00:09:22.880 --> 00:09:29.200 but running completely in software on the CPU our 00:09:27.360 --> 00:09:32.160 GPU is doing nothing 00:09:30.560 --> 00:09:35.760 actually wait no our GPU is not doing nothing why is our GPU doing something 00:09:33.839 --> 00:09:40.160 it's uh doing i think it's doing the media decode oh yeah i think you're 00:09:37.519 --> 00:09:44.640 right oh that's hilarious okay single player you know what's funny 00:09:43.279 --> 00:09:50.880 at first i thought that this wasn't working because last time when i tried doing it 00:09:47.920 --> 00:09:54.160 on the CPU the background was black it didn't load 00:09:52.399 --> 00:09:58.399 it's just working like way better yeah 00:09:56.880 --> 00:10:02.480 wow wait are you sure this looks like it 00:09:59.920 --> 00:10:06.720 might be running on GPU um you'll see here we go all right rock on there go 00:10:05.279 --> 00:10:11.360 though okay hold on wait for it yeah every everything that 00:10:09.680 --> 00:10:14.079 it has to load will take a little a little while but then 00:10:12.800 --> 00:10:19.839 then it gets a little bit better every time 00:10:16.160 --> 00:10:22.000 okay we're good we good we good 00:10:19.839 --> 00:10:26.240 oh we less good this is nothing fuds 00:10:24.800 --> 00:10:30.000 and if you guys want to see it do even more we actually have an upcoming video 00:10:28.399 --> 00:10:34.720 planned where we're gonna put this thing under our completely redesigned subzero 00:10:32.800 --> 00:10:38.320 chiller so make sure you're subscribed so you don't miss that this is crazy 00:10:36.640 --> 00:10:43.600 this is definitely running smoother than last time Anthony yeah 00:10:40.640 --> 00:10:48.320 this is noticeably better actually look it's hitting all the cores 00:10:45.440 --> 00:10:52.560 a little bit yeah yeah this is way more playable than last time like i i can 00:10:50.560 --> 00:10:56.079 play games like this if i have to i used to on my pentium 4. that's right you had 00:10:54.800 --> 00:11:01.519 a pretty crappy computer when you started here didn't you um i know it 00:10:58.079 --> 00:11:03.680 wasn't that bad i had a 6 700k and uh oh 00:11:01.519 --> 00:11:07.519 okay 1080 when i started here yeah like i i got me some kills yo yeah you 00:11:06.079 --> 00:11:10.399 struggled to even hit anything last time yeah like it's to be clear it's not 00:11:09.440 --> 00:11:16.399 great yeah it's not what you'd call a good experience but like 00:11:13.440 --> 00:11:23.360 the fact that it's running acceptably on a CPU unreal yeah like the 64 core i 00:11:19.440 --> 00:11:23.360 mean cryengine cryengine yeah yeah 00:11:23.680 --> 00:11:29.760 really impressive numbers like mind blowing but the thing is if you're 00:11:27.760 --> 00:11:34.079 actually considering purchasing one of these monsters you got to make sure you 00:11:31.600 --> 00:11:38.000 do your research not every load is going to scale the way that some of the ones 00:11:36.000 --> 00:11:43.600 we showed did for example we did a firefox compile benchmark and it only 00:11:40.640 --> 00:11:48.880 outperformed the 32 core threadripper by a very narrow margin now 00:11:46.240 --> 00:11:52.320 some compiles will get a lot of benefit from the additional course it's just 00:11:50.079 --> 00:11:56.160 that not all of them will you also need to make sure that you do your research 00:11:54.240 --> 00:12:01.040 the low volume motherboards for these kinds of bleeding edge platforms can be 00:11:58.480 --> 00:12:05.680 quirky at best so make sure you check out user reviews and 00:12:03.519 --> 00:12:11.680 we got a really counter-intuitive configuration tip from AMD themselves 00:12:08.560 --> 00:12:16.480 they actually recommended against using 00:12:11.680 --> 00:12:17.360 a pci express gen 4 SSD with the 3990x 00:12:16.480 --> 00:12:22.880 like what so get this guys 00:12:19.600 --> 00:12:25.200 most gen 4 ssds are improperly tuned for 00:12:22.880 --> 00:12:30.000 desktop workloads because huge bandwidth numbers look great in the marketing 00:12:27.040 --> 00:12:35.600 materials thing is though it's iops that you actually want for peak system 00:12:32.399 --> 00:12:38.000 responsiveness and with so many threads 00:12:35.600 --> 00:12:44.079 pounding the SSD while you're working on something that becomes even more of a 00:12:40.480 --> 00:12:46.800 concern with a CPU like this one so on a 00:12:44.079 --> 00:12:52.320 poorly optimized drive even a mundane task like a background antivirus scan 00:12:49.600 --> 00:12:57.120 could become crippling so the best choice would obviously be something like 00:12:54.320 --> 00:13:00.720 Intel's optane although AMD didn't specifically mention octane for fairly 00:12:59.279 --> 00:13:05.680 obvious reasons grow up guys all right so then AMD is 00:13:03.360 --> 00:13:10.399 clearly not shying away from the reality that most people who don't want to worry 00:13:07.839 --> 00:13:15.200 about that kind of nonsense should just buy a Ryzen 9 39 50 x which raises the 00:13:13.120 --> 00:13:21.440 question who is this thing actually for 00:13:17.440 --> 00:13:24.399 the answer apparently is the desk side 00:13:21.440 --> 00:13:29.120 compute sector people who need immediate access to large amounts of processing 00:13:26.639 --> 00:13:33.440 power at their desks that means it's for vfx artists who are doing iterations on 00:13:31.360 --> 00:13:38.880 a project or programmers who are recompiling large complex application 00:13:36.320 --> 00:13:43.200 packages on the fly so the kind of stuff that would take a lot longer if you had 00:13:41.279 --> 00:13:48.160 to throw it into the processing queue for a group server every time you tweak 00:13:45.839 --> 00:13:53.680 something i can also see the utility of this kind of a CPU in a virtualization 00:13:50.639 --> 00:13:56.720 environment at 39.90 us dollars you're 00:13:53.680 --> 00:13:59.120 looking at a relatively low cost for a 00:13:56.720 --> 00:14:03.680 lot of cores and with the ability to split everything into four distinct 00:14:01.440 --> 00:14:09.199 pneuma nodes you've got this natural splitting point for a handful of 16 core 00:14:06.480 --> 00:14:13.760 vms with minimal performance loss due to memory access overhead 00:14:11.839 --> 00:14:18.240 except that there's an elephant in the room that puts a bit of a damper on my 00:14:16.880 --> 00:14:24.160 excitement for this product overall AMD and their 00:14:20.800 --> 00:14:27.760 infinite wisdom decided to pull an Intel 00:14:24.160 --> 00:14:30.800 with their total memory capacity support 00:14:27.760 --> 00:14:31.839 by only supporting standard unbuffered 00:14:30.800 --> 00:14:37.680 dimms that means that threadripper even though 00:14:34.639 --> 00:14:39.279 it has an on paper support for up to two 00:14:37.680 --> 00:14:43.839 terabytes of RAM is practically speaking stuck with a 00:14:41.600 --> 00:14:49.199 maximum of gigabytes today so while it smokes the 00:14:47.120 --> 00:14:53.360 xeon w3275 in raw compute 00:14:50.959 --> 00:14:57.839 that chip from Intel supports up to one terabyte of RAM which for some workloads 00:14:56.000 --> 00:15:02.800 is far more important than how many cores you have 00:14:59.279 --> 00:15:04.959 so why doesn't AMD want their flagship 00:15:02.800 --> 00:15:10.800 64 core Threadripper to support registered or load reduced dimms 00:15:07.760 --> 00:15:12.959 well because despite the fact that epic 00:15:10.800 --> 00:15:17.519 supports a whopping 4 terabytes of memory and considering that even 00:15:15.199 --> 00:15:21.360 supporting registered dimms would have doubled threadripper's maximum capacity 00:15:19.680 --> 00:15:24.800 without even surpassing an eighth of epix they simply didn't want to risk 00:15:23.279 --> 00:15:28.880 stepping on their big boy CPU's toes 00:15:26.880 --> 00:15:34.240 it's one of those artificial limitations that in this case 00:15:30.800 --> 00:15:36.880 seriously hurts the 3990x's future 00:15:34.240 --> 00:15:43.279 utility in high-end vfx work because remember 256 gigs is the standard for 00:15:40.160 --> 00:15:44.800 that kind of work in 4k today 00:15:43.279 --> 00:15:49.199 that's to say nothing for when compositions get more complex and higher 00:15:47.199 --> 00:15:53.600 in resolution so this is just the craziest 00:15:51.920 --> 00:15:59.199 4 000 dead end that i sincerely hope can be 00:15:57.040 --> 00:16:03.920 fixed by a firmware update not by waiting around for bigger 00:16:01.519 --> 00:16:09.600 unbuffered dimms to come along like seriously guys you don't need to do 00:16:06.560 --> 00:16:11.120 this epic still has double the total 00:16:09.600 --> 00:16:16.720 memory bandwidth double the theoretical memory capacity 00:16:13.759 --> 00:16:22.320 more pci express lanes and for the non-p skus dual socket support don't you think 00:16:19.680 --> 00:16:26.639 that's enough differentiation or are you starting to regret all the 00:16:24.639 --> 00:16:30.480 big specs that you guys built into the platform now that you're actually 00:16:28.399 --> 00:16:34.639 selling cpus that don't have any competition 00:16:32.000 --> 00:16:37.759 this right here is exhibit a for why you shouldn't fanboy folks 00:16:36.639 --> 00:16:44.800 still a win is a win and i can't deny AMD that 00:16:41.360 --> 00:16:46.480 as inapplicable as this may be for 00:16:44.800 --> 00:16:51.680 almost a hundred percent of you watching today and even much of its target market 00:16:49.360 --> 00:16:56.639 unfortunately it serves its most important function as a halo product 00:16:54.399 --> 00:17:00.000 extremely well and it's an incredibly exciting milestone for personal 00:16:58.160 --> 00:17:04.400 computing not only is AMD putting an unprecedented amount of processing power 00:17:02.399 --> 00:17:07.760 in the high-end desktop market but at this price 00:17:05.520 --> 00:17:14.000 Intel's not even in the same ballpark their closest competitor overall is the 00:17:10.079 --> 00:17:17.919 28 core xeon w 3275 which costs a good 00:17:14.000 --> 00:17:19.760 500 more and while it may support 00:17:17.919 --> 00:17:23.919 practically speaking four times the memory that threadripper does it just 00:17:21.600 --> 00:17:28.319 doesn't come close to the compute performance on display here to even 00:17:26.240 --> 00:17:33.520 approach that you would need dual xeon platinum 8280s and with those costing 00:17:30.799 --> 00:17:38.240 over ten thousand dollars each there's a very real discussion to be had about 00:17:35.360 --> 00:17:41.840 which you need more the horsepower or the memory capacity 00:17:40.080 --> 00:17:45.679 i mean fortunately for Intel they do have a bigger and every way data center 00:17:43.520 --> 00:17:51.039 CPU that might smoke threadripper the xeon 9282 with 56 cores but i have no 00:17:48.960 --> 00:17:54.000 way of knowing if it works really well because it costs so much that Intel 00:17:52.480 --> 00:17:59.280 doesn't even list a price meaning it's probably high enough that i couldn't even trade an autographed kidney for one 00:17:58.000 --> 00:18:02.480 so then that's it AMD is so far ahead in compute 00:18:01.039 --> 00:18:06.559 performance on the desktop right now that the only useful comparison is a 00:18:04.240 --> 00:18:11.200 data center CPU and they've begun purposely kneecapping their desktop 00:18:08.240 --> 00:18:15.200 chips in ways that make them only just barely marketable to their target 00:18:12.880 --> 00:18:19.200 audience seemingly in an effort to keep their profits higher so that they can 00:18:16.640 --> 00:18:23.039 sell epic chips instead which is really frustrating because the epics don't 00:18:20.960 --> 00:18:27.919 turbo the same way and actually might not be as good for this kind of a 00:18:25.039 --> 00:18:32.080 workload it's like bizarro world the situation right here is where Intel was 00:18:30.400 --> 00:18:36.320 when they launched sandy bridge just like kicking AMD while they were down 00:18:34.480 --> 00:18:41.280 you know artificially limiting their best cpus and pulling a big enough lead 00:18:39.280 --> 00:18:45.919 that for their competitor to catch up they would need a whole new architecture 00:18:43.520 --> 00:18:49.840 which Intel is presumably working on and i am very excited to see it those guys 00:18:48.320 --> 00:18:54.960 really do do their best work with egg all over their faces don't they 00:18:52.880 --> 00:18:59.200 and i do my best work in my sponsor segways today's video is brought to you 00:18:56.960 --> 00:19:04.320 by the massdrop object 2 headphone amplifier aka the o2 amp this amp was 00:19:02.640 --> 00:19:09.520 designed and created with the feedback from over 500 mass drop members on what 00:19:07.039 --> 00:19:13.760 they wanted and over 6 000 of them are out there in the wild it serves as an 00:19:11.440 --> 00:19:17.679 ideal baseline reference amplifier and can power everything from in-ear 00:19:15.200 --> 00:19:22.080 monitors to the hd800s you can adjust the input output and power arrangements 00:19:20.160 --> 00:19:26.480 in two different gains medium and standard it delivers big clear and 00:19:24.400 --> 00:19:30.000 accurate sound and it ships for free in the us so check it out at the link in 00:19:28.480 --> 00:19:34.000 the video description thanks for watching guys go check out our AMD epic 00:19:32.000 --> 00:19:37.760 7742 video if you're wondering what to watch next it is the 00:19:35.679 --> 00:19:41.200 big boy pants equivalent to the 3990x 00:19:39.520 --> 00:19:44.880 that AMD clearly wants you to buy instead i'll see you over there