WEBVTT 00:00:00.160 --> 00:00:05.440 how much RAM should i get for my computer so many people ask the question 00:00:03.840 --> 00:00:11.599 and the only answer they can get from the in the no techies and i mean we've 00:00:08.080 --> 00:00:14.080 even done it here before is usually well 00:00:11.599 --> 00:00:18.800 how much are you planning to use get that much which is only useful if 00:00:16.560 --> 00:00:22.000 you know how much your workload requires for gamers it's relatively simple 00:00:20.560 --> 00:00:25.519 there's a number on the back of the box excuse me a digital information page 00:00:23.840 --> 00:00:29.519 because physical media but for content creators and our main 00:00:27.680 --> 00:00:33.520 focus is going to be on adobe creative cloud because that's what we use here 00:00:31.359 --> 00:00:36.719 between adobe's own system requirements and white papers and the numerous 00:00:35.360 --> 00:00:40.960 professionals on the message boards around the web there's no solid answer 00:00:38.640 --> 00:00:45.360 all we really hear is you need multiple fast storage locations more memory is 00:00:43.360 --> 00:00:50.559 good uh faster processors with more cores are good and video cards with like 00:00:47.520 --> 00:00:52.000 speed and memory capacity are good this 00:00:50.559 --> 00:00:56.559 has been the ongoing guidance with seemingly no consideration for recent 00:00:54.399 --> 00:00:59.920 technological changes in the shifting bottlenecks within pcs and workstations 00:00:58.719 --> 00:01:04.720 for years so as part of the process of designing 00:01:02.320 --> 00:01:09.760 the workstations for our editors to find the optimal cost to performance 00:01:07.200 --> 00:01:15.520 compromise for a given workload i asked edzel to run some benchmarks in premiere 00:01:12.080 --> 00:01:17.600 pro after effects and photoshop all the 00:01:15.520 --> 00:01:23.920 creative cloud 2014 versions to investigate the effect of more CPU cores 00:01:20.560 --> 00:01:25.759 more RAM up to 128 gigs and varying 00:01:23.920 --> 00:01:28.400 storage solutions now i was half expecting scraps of paper with notes 00:01:27.280 --> 00:01:34.400 scribbled on them so that we could configure our workstations but what i actually got was a very detailed summary 00:01:32.400 --> 00:01:38.079 with useful charts that we'll publish separately on the Linus tech tip site in 00:01:36.720 --> 00:01:41.360 the link in the video description for folks who want to read it but for those 00:01:40.000 --> 00:01:45.360 of you who prefer videos well we made a video about it too so watch on oh and 00:01:44.000 --> 00:01:48.640 mash that subscribe button if you haven't already because our x99 00:01:47.040 --> 00:01:58.799 overclocking guide another great way of squeezing performance out of your hardware will be coming soon 00:01:58.799 --> 00:02:06.640 Corsair gaming RGB keyboards feature precision cherry mx RGB key switches for 00:02:03.880 --> 00:02:11.360 16.8 million color per key backlighting for virtually unlimited customization 00:02:09.119 --> 00:02:16.400 click now to learn more i'll start by introducing the test bench it's edzel's 00:02:13.200 --> 00:02:18.239 12 core xeon e5 2697 based workstation 00:02:16.400 --> 00:02:22.080 with the same ASUS x79 deluxe motherboard that we use in all of our 00:02:19.599 --> 00:02:26.239 editing rigs two GeForce gtx titans for accelerating 3d rendering a cooler 00:02:23.840 --> 00:02:31.760 master 1200 watt modular power supply a kingston 240 gig SSD and a wd one 00:02:29.360 --> 00:02:37.120 terabyte velociraptor for storage and normally 64 gigs of hyperx quad channel 00:02:34.720 --> 00:02:42.319 RAM but to properly investigate the effects of memory on performance we 00:02:39.360 --> 00:02:47.840 needed more RAM so we used eight 16 gig sticks of intelligent memory ecc 00:02:45.840 --> 00:02:52.280 RAM huge thanks to memphis by the way for providing this to us for testing 00:02:49.760 --> 00:02:58.959 giving us a whopping 128 gigs of RAM to either allocate to 00:02:55.440 --> 00:03:00.879 programs or use as a RAM disk to see if 00:02:58.959 --> 00:03:05.599 we could use excess memory in the system to alleviate storage bottlenecks 00:03:03.519 --> 00:03:09.760 so our premiere pro investigation started with allocating memory to the 00:03:07.440 --> 00:03:15.920 program via the preferences menu and rendering out projects at 1080p and 4k 00:03:12.560 --> 00:03:17.599 resolution with x264 we used x264 00:03:15.920 --> 00:03:22.560 because it was better optimized for multi-core processors which inherently 00:03:19.760 --> 00:03:26.640 utilizes more memory now remember that when we allocate memory to premiere that 00:03:24.800 --> 00:03:31.519 doesn't mean that's the amount of RAM in the system but rather how much of what's 00:03:29.200 --> 00:03:36.000 in the system we can afford to allocate purely to premiere to play with so at 00:03:34.159 --> 00:03:39.760 1080p we found that if you don't want to multitask on your machine while 00:03:37.599 --> 00:03:44.720 rendering eight gigs of system memory is likely to be fine there was no benefit 00:03:42.159 --> 00:03:50.400 to more RAM beyond our lowest tested six gig allocation at 4k though the story 00:03:47.760 --> 00:03:55.200 changes somewhat our rad project render times improved by seven percent when we 00:03:52.720 --> 00:03:59.120 went from six gigs to 24 gigs of allocated memory flattening out 00:03:57.120 --> 00:04:03.280 significantly after that demonstrating that the conventional wisdom of more RAM 00:04:01.840 --> 00:04:08.000 doesn't really help beyond a certain point continues to hold true but where 00:04:06.080 --> 00:04:11.680 that certain point is will depend on the type of projects you're working on and 00:04:09.519 --> 00:04:18.079 we found that our 4k prores renders benefited from up to 64 gigs of RAM on 00:04:14.879 --> 00:04:20.239 Windows 8 but not beyond for larger 00:04:18.079 --> 00:04:23.199 longer projects like movies you might end up needing more memory to keep 00:04:21.519 --> 00:04:26.479 things running smoothly while editing but render times still probably won't 00:04:25.199 --> 00:04:31.120 benefit much now since the usefulness of more RAM 00:04:28.320 --> 00:04:35.520 falls off a cliff past 64 gigs that left us with ha another 64 gigs of RAM to 00:04:33.840 --> 00:04:40.639 play around with in the system so we decided to see if we could utilize a RAM 00:04:37.440 --> 00:04:42.960 disk as a scratch disk and 00:04:40.639 --> 00:04:47.840 okay yeah uh faster storage here doesn't actually do much sort of render times 00:04:45.840 --> 00:04:52.639 aren't improved but a separate physical drive for a scratch disk whether it's 00:04:49.520 --> 00:04:54.960 RAM or or SSD especially SSD versus 00:04:52.639 --> 00:04:59.360 slower hard drives does improve the overall perceived system performance 00:04:57.199 --> 00:05:02.240 with more responsive playback faster file importing and faster project 00:05:01.199 --> 00:05:06.560 opening being some of the most noticeable benefits moving on to after 00:05:04.960 --> 00:05:11.280 effects our investigation changed somewhat to include both RAM allocation 00:05:08.960 --> 00:05:14.880 and CPU core scaling since we couldn't find much information about either of 00:05:12.800 --> 00:05:20.000 these topics RAM allocation in multi-processor rendering mode is a 00:05:16.880 --> 00:05:22.479 particularly confusing thing because 00:05:20.000 --> 00:05:27.840 adobe allows anywhere from one gig to six gigs with the caveat being that if 00:05:25.199 --> 00:05:30.800 you have too many cores for the amount of RAM that you're trying to allocate 00:05:29.440 --> 00:05:35.919 per core in your system let's say you have a quad core CPU 16 gigs of RAM and 00:05:33.440 --> 00:05:42.080 you go and try and allocate six gigs per core to total of 24 gigs derp the 00:05:39.039 --> 00:05:44.240 program supposedly simply won't use all 00:05:42.080 --> 00:05:48.320 of your cores in this case and on top of that it treats logical processors or 00:05:46.320 --> 00:05:53.520 hyper-threaded cores the same as physical ones so finding the balance 00:05:50.560 --> 00:05:56.720 here then is key the largest performance improvement from additional RAM 00:05:54.960 --> 00:06:01.919 allocation was going from one gig per core to two gigs per core so in theory 00:06:00.080 --> 00:06:07.360 then we should ensure that we have at least 48 gigs of RAM for a 12 core 00:06:04.960 --> 00:06:11.600 hyper-threaded after effects workstation but beyond that the law of diminishing 00:06:09.199 --> 00:06:15.360 returns should hit us pretty hard but then we noticed an interesting thing 00:06:13.520 --> 00:06:20.960 here and that was that even though our six gigs per core allocation should have 00:06:17.600 --> 00:06:24.720 required 144 gigs of system RAM limiting 00:06:20.960 --> 00:06:28.639 us to 19 of our 24 threads theoretically 00:06:24.720 --> 00:06:30.319 CPU utilization still hovered around 99 00:06:28.639 --> 00:06:35.280 even in this case so maybe hyper threaded cores really 00:06:33.039 --> 00:06:40.160 aren't affecting performance much which led us to our investigation into CPU 00:06:37.600 --> 00:06:45.600 cores and after effects performance we tackled this in two different ways first 00:06:42.800 --> 00:06:50.160 by using CPU scheduling to isolate our real course validating our hypothesis 00:06:48.240 --> 00:06:54.560 that hyper threading isn't really doing a whole lot in this case and then by 00:06:52.479 --> 00:06:59.280 disabling cores manually within the computer's UEFI BIOS to see how 00:06:56.720 --> 00:07:03.520 performance is affected by more actual physical cores so in that case 00:07:01.599 --> 00:07:08.160 performance improvements from having more horses pulling the cart were 00:07:05.840 --> 00:07:13.120 immediately noticeable and while not linear still demonstrated excellent 00:07:10.400 --> 00:07:18.080 scaling now you might think then what is the point of the CPU scheduling setting 00:07:15.360 --> 00:07:23.360 then why even bother with it well edsel found that reserving CPU cores even 00:07:20.800 --> 00:07:27.840 though CPU usage by the system didn't change much prevented after effects from 00:07:25.520 --> 00:07:31.599 being over scheduled and lagging out the system and we'll wrap things up in 00:07:29.840 --> 00:07:36.560 photoshop where the objective was to determine how much memory we would need 00:07:34.080 --> 00:07:41.039 for manipulating and resizing very large images on a beefy workstation like this 00:07:38.479 --> 00:07:45.599 one and the answer is well really not a heck of a lot while four gigs of RAM 00:07:43.520 --> 00:07:50.319 resulted in much lower performance than eight or more beyond eight gigs we 00:07:47.599 --> 00:07:54.000 didn't observe much of a difference but bear in mind that this will greatly 00:07:51.919 --> 00:07:58.800 depend on the size of your project and the history state setting so the project 00:07:56.000 --> 00:08:02.639 we used took 4.75 gigs of memory or scratch disk and you know the size of 00:08:01.280 --> 00:08:05.280 the project then of course greatly affects the memory usage of your actions 00:08:04.319 --> 00:08:11.360 so in our case while it's not something we can easily benchmark we did find that 00:08:09.039 --> 00:08:14.720 our SSD's performance was a bit of a bottleneck while opening projects or 00:08:13.039 --> 00:08:17.840 performing certain actions things we would have assumed adobe would try to do 00:08:16.160 --> 00:08:22.319 in memory first rather than relying on scratch so we did find that creating a 00:08:20.080 --> 00:08:26.479 RAM disk as the scratch disk could improve the user experience by forcing 00:08:24.000 --> 00:08:30.639 photoshop to write to RAM seven SSD but ultimately for raw performance it's just 00:08:29.280 --> 00:08:35.519 going to come down to how big is your project and how much of it sits in memory so that 00:08:33.519 --> 00:08:39.200 same old conclusion which i guess leads us to the rest of the conclusion 00:08:37.440 --> 00:08:43.599 in the last few years the general guidelines lots of processing cores lots 00:08:41.440 --> 00:08:47.440 of RAM and fast storage haven't really changed 00:08:44.880 --> 00:08:52.399 and they didn't change but i still hope that this video for adobe creators 00:08:49.519 --> 00:08:56.080 anyway helped to clarify which of these things might have a bigger impact on 00:08:54.560 --> 00:08:59.440 them personally speaking of impact we have a brand new 00:08:57.839 --> 00:09:04.240 sponsor on Linus tech tips today that might impact your anime consumption 00:09:01.440 --> 00:09:07.839 habits crunchyroll crunchyroll is obsessed with anime and was built by 00:09:06.080 --> 00:09:12.560 hardcore fans they have a massive selection of shows including the ever 00:09:09.839 --> 00:09:16.160 popular naruto shippuden attack on titan and sword art online and you can stream 00:09:14.640 --> 00:09:20.560 them on a variety of devices like your computer smartphone tablet apple tv or 00:09:18.320 --> 00:09:24.640 xbox their premium subscription which is only seven bucks a month or 6.95 excuse 00:09:23.040 --> 00:09:28.800 me it's very different allows you to stream all of their content completely 00:09:26.480 --> 00:09:32.160 ad free in beautiful hd and you can watch the latest current season episodes 00:09:30.399 --> 00:09:36.720 of shows like fate stay night with professional subtitles as soon as one 00:09:34.480 --> 00:09:40.320 hour after they air in japan but wait there's more crunchyroll is offering 00:09:38.640 --> 00:09:44.800 everyone in the Linus tech tips audience a free 30-day trial if you visit our 00:09:42.720 --> 00:09:48.640 link crunchyroll.com Linus so if you're serious about anime 00:09:47.279 --> 00:09:54.880 and want to binge on some of the biggest shows from japan head over to that link in the video description and start 00:09:52.160 --> 00:09:58.959 watching today guys like this video if you liked it dislike it if you thought 00:09:56.560 --> 00:10:01.440 it just sucked and leave a comment below letting us know 00:10:00.240 --> 00:10:06.240 actually i really want to hear what you guys thought of this video it's a little bit of a departure from our from our 00:10:04.480 --> 00:10:10.480 usual style a little bit more hardcore content creation as opposed to consumer 00:10:08.959 --> 00:10:14.079 thanks for watching guys oh yeah check out the link in the video description if you want to support us and help us make 00:10:12.880 --> 00:10:18.800 videos you can give us a monthly contribution buy a cool t-shirt like this one or just change your amazon 00:10:17.360 --> 00:10:24.880 bookmark to one with our affiliate code embedded in it so we get a small kickback whenever you buy more memory or 00:10:22.160 --> 00:10:29.660 CPU cores or faster storage thanks for watching and as always don't forget to 00:10:26.880 --> 00:10:32.789 subscribe 00:10:38.079 --> 00:10:40.160 you