1 00:00:00,359 --> 00:00:07,759 I have been promising a RAM overclocking guide for seemingly forever so I've got 2 00:00:05,240 --> 00:00:12,880 the Kingston hyperx blue 4 gig kit of DDR3 memory on my test bench today uh in 3 00:00:10,880 --> 00:00:16,920 addition to that I do have a hyperx RAM fan which I did an unboxing of a while 4 00:00:14,719 --> 00:00:20,359 back I found it to be extremely high quality I really like the look of it 5 00:00:18,439 --> 00:00:26,240 overall and the other relevant things that I have today on my test bench are a 6 00:00:22,160 --> 00:00:28,240 core i7 875k unlocked CPU so that's 7 00:00:26,240 --> 00:00:33,800 what's going on under the hood over here and then I'm using a p7p 5de premium 8 00:00:31,160 --> 00:00:36,800 motherboard in case you are wondering so the first thing we're going to do today 9 00:00:35,320 --> 00:00:41,920 is we're going to get everything set up for a baseline reading so that means 10 00:00:39,200 --> 00:00:46,320 I've got my CPU set up at stock frequency and I've actually turned off 11 00:00:44,399 --> 00:00:51,640 turbo mode so we are just going to be running it at peer stock frequency this 12 00:00:48,199 --> 00:00:54,520 is going to allow us to see exactly what 13 00:00:51,640 --> 00:01:00,120 impact just changing the memory has on an otherwise completely stable Baseline 14 00:00:58,079 --> 00:01:04,479 system okay the other thing that I've done is I've enabled the XMP profile for 15 00:01:02,480 --> 00:01:08,040 this memory so that goes ahead and changes all the memory timings for me 16 00:01:06,280 --> 00:01:11,960 Isn't that handy U make sure you buy memory with XMP if you don't want to 17 00:01:10,439 --> 00:01:17,000 fuss around with all of this stuff so it goes ahead changes the frequency to 1600 18 00:01:14,560 --> 00:01:21,920 MHz changes the timings to 9927 these are your primary timings the 19 00:01:20,159 --> 00:01:29,159 most important one is this one right here that is the Cass Cas latency so 20 00:01:26,280 --> 00:01:35,439 here this frequency that is the speed of the memory so that how how fast it 21 00:01:32,000 --> 00:01:38,200 Cycles around these ones are the delay 22 00:01:35,439 --> 00:01:44,079 so higher this and lower this is always better 1 n or 1 t or 2 N or 2T that's a 23 00:01:42,320 --> 00:01:47,640 secondary timing that does have an impact on performance as well and then 24 00:01:46,119 --> 00:01:52,640 these voltages that it's dialed in for me 1.65 that's the memory voltage and 25 00:01:50,360 --> 00:01:57,920 1.3 that is the voltage of the integrated memory controller on the CPU 26 00:01:55,640 --> 00:02:02,439 you should be aware that with core i7 and core i5 CPUs you should never go 27 00:02:00,200 --> 00:02:06,520 over 1.65 volts on the memory because you can damage the memory controller 28 00:02:04,719 --> 00:02:09,080 that is built into the CPU so we're going to go ahead take some baseline 29 00:02:07,880 --> 00:02:14,480 readings and then we're going to come back and play around with some of these settings a little bit and see how much 30 00:02:12,360 --> 00:02:20,440 more performance we can squeeze out of this memory now that we've got the RAM 31 00:02:17,360 --> 00:02:22,599 dialed into our stock XMP overclocked 32 00:02:20,440 --> 00:02:27,120 profile you can see here all of the settings are correct and the CPU is 33 00:02:24,800 --> 00:02:31,000 running at its stock frequency we're just going to take a couple of 34 00:02:28,440 --> 00:02:37,879 benchmarks as a base line here so Super PI is a popular memory Benchmark because 35 00:02:33,560 --> 00:02:42,560 what it does is it gives you uh probably 36 00:02:37,879 --> 00:02:45,000 the largest possible separation between 37 00:02:42,560 --> 00:02:51,159 uh like minute setting basically it's it's a very very fine way of measuring 38 00:02:47,440 --> 00:02:53,440 the the performance of the CPU and RAM 39 00:02:51,159 --> 00:02:57,280 and it'll give you differences that you otherwise wouldn't be able to measure so 40 00:02:55,159 --> 00:03:02,239 you can figure out if what you're doing is beneficial or not so you can see we 41 00:02:59,840 --> 00:03:08,440 calculated pi to 1 million decimal places and we end up with a result of 42 00:03:05,959 --> 00:03:13,519 zooming in so I can see oh no that's not right Calculate 1 million decimal places 43 00:03:11,560 --> 00:03:17,920 it's giving me the wrong one here so now click to start okay and let's find out 44 00:03:16,840 --> 00:03:21,640 what we get this time and we're going to take a quick 3D Mark reading as well and 45 00:03:20,319 --> 00:03:26,480 then we're going to come back to these benchmarks once we've played around with 46 00:03:23,640 --> 00:03:30,040 the frequency latencies of the memory and we're going to see how much we are 47 00:03:28,120 --> 00:03:36,799 able to improve our system performance so there there's the real value is 48 00:03:33,799 --> 00:03:36,799 14305 49 00:03:37,400 --> 00:03:42,360 seconds all right so we're going to run 3D Mark and then I'll be back in a 50 00:03:40,760 --> 00:03:48,040 minute once that's done so there you have it our Baseline 51 00:03:44,599 --> 00:03:51,720 3D mar06 reading is 52 00:03:48,040 --> 00:03:54,840 18534 so 1 53 00:03:51,720 --> 00:03:56,879 18534 3D marks it should also be noted I 54 00:03:54,840 --> 00:04:02,519 am running these benchmarks with turbo mode disabled on the CPU so it is 55 00:03:59,920 --> 00:04:07,840 running only at the default CPU multiplier of 22x it is not scaling up 56 00:04:06,120 --> 00:04:11,680 when I run intensive benchmarks this is just to isolate the RAM as the 57 00:04:10,120 --> 00:04:16,199 performance variable that we're looking at today so the first thing we're going 58 00:04:13,560 --> 00:04:19,400 to try today is leaving our timings at default leaving all the voltages the 59 00:04:17,680 --> 00:04:22,680 same cuz we don't really want to that's outside the scope of this particular 60 00:04:21,079 --> 00:04:26,199 video we don't want to Tinker around with those and what we're going to go 61 00:04:24,199 --> 00:04:33,039 ahead and do is take the CPU ratio and turn it down so we're still very close 62 00:04:28,479 --> 00:04:34,800 to the 2.9 926 Target CPU frequency and 63 00:04:33,039 --> 00:04:38,400 then we're going to turn up the base clock now turning up the base clock is 64 00:04:36,440 --> 00:04:43,639 going to give us a bit of a boost at equal frequency on the CPU and then it's 65 00:04:41,000 --> 00:04:48,600 also going to give us at the same DDR3 speed so go I'm going to go ahead so 146 66 00:04:45,840 --> 00:04:54,000 is where I'm headed so 133 is stock you can see at 133 the options I have are 67 00:04:51,120 --> 00:05:00,360 all of your standard DDR3 speed options but as soon as I go up to 146 that 1600 68 00:04:57,440 --> 00:05:05,240 MHz option turns into 17 50 so we're going to go ahead and see if 69 00:05:03,160 --> 00:05:09,759 we can boot up at these speeds and if we are stable and what kind of performance 70 00:05:07,160 --> 00:05:13,720 impact that's going to make okay I've done some tinkering around and staying 71 00:05:11,639 --> 00:05:19,240 within the same CPU frequency range so about 2.93 GHz I have found that the 72 00:05:16,759 --> 00:05:24,880 highest I was able to go was by lowering the CPU ratio to 17x and increasing the 73 00:05:22,680 --> 00:05:31,280 base clock to 172 so this gives us a DDR3 speed of 2.06 4 GHz the RAM was 74 00:05:29,919 --> 00:05:35,800 stable at this speed so we're going to take some more benchmarks and we're 75 00:05:33,199 --> 00:05:40,639 going to find out what we gained by increasing the memory speed to 2 GHz now 76 00:05:38,479 --> 00:05:44,479 one thing that I do want to point out is that you do not have to run the CPU at 77 00:05:43,000 --> 00:05:48,840 stock speed in fact there's probably more benefit to overclocking the RAM 78 00:05:46,919 --> 00:05:52,680 when you're running an overclocked CPU because the CPU needs more data to keep 79 00:05:51,479 --> 00:05:57,039 it fed but the reason that I did it this 80 00:05:54,960 --> 00:06:01,560 way is because what you'd normally do is find your max CPU frequency in this case 81 00:05:59,479 --> 00:06:05,800 I'm I'm pretending my Max CPU frequency is stock speed so You' find your max CPU 82 00:06:03,919 --> 00:06:10,800 frequency and then you'd tune the RAM from there so uh once you by the time 83 00:06:08,680 --> 00:06:16,479 you're tuning RAM you're already trying to stay within kind of a a fixed range 84 00:06:13,720 --> 00:06:22,000 of CPU frequencies so we'll go ahead fire up CPU Z fire up Super PI so you 85 00:06:19,680 --> 00:06:26,520 can see we're running our CPU at stock speed our RAM is now running 86 00:06:24,360 --> 00:06:33,840 substantially faster this is 1/ half of the actual data rate because it is uh d 87 00:06:30,160 --> 00:06:37,560 DDR class of memory so we're running 88 00:06:33,840 --> 00:06:42,080 start to calculate to 1 million decimal 89 00:06:37,560 --> 00:06:46,240 places so here we are we've seen a 90 00:06:42,080 --> 00:06:48,319 25% increase in RAM speed and let's see 91 00:06:46,240 --> 00:06:54,879 how much of a speed increase we will gain from that so 92 00:06:50,639 --> 00:06:57,000 14.22 7 is our new super 93 00:06:54,879 --> 00:07:02,000 pie 14.22 7 is our new super pie score and 94 00:07:00,560 --> 00:07:07,599 then we'll go ahead and have a look at the 3D Mark score in a moment so we end 95 00:07:05,479 --> 00:07:12,680 up with a score of 18692 which is a slight boost over what 96 00:07:10,720 --> 00:07:17,319 we were recording as our Baseline although it's pretty much within the 97 00:07:14,639 --> 00:07:21,319 margin of error uh so the overclocking method that we used today was just 98 00:07:18,879 --> 00:07:26,400 increasing the RAM frequency uh we found it didn't have much of an impact but 99 00:07:24,280 --> 00:07:33,000 look at it this way you buy the hyperx blue and obviously your results May Vary 100 00:07:29,599 --> 00:07:35,479 you can get upwards of a 25% increase in 101 00:07:33,000 --> 00:07:40,639 speed over what it's actually rated for which even if it only gives you like a 102 00:07:37,319 --> 00:07:43,199 1% increase in a couple of the programs 103 00:07:40,639 --> 00:07:47,280 you use it is a free increase nonetheless so we're going to see how 104 00:07:44,919 --> 00:07:51,960 low we can go with latency overclocking and see if we can get some bigger boosts 105 00:07:49,840 --> 00:07:57,800 that way stay tuned for more by simply adjusting the latencies we were able to 106 00:07:53,639 --> 00:08:00,560 go from 999 27 latencies all the way 107 00:07:57,800 --> 00:08:04,319 down to 77720 you consider the cost difference between kits that are 108 00:08:02,560 --> 00:08:11,360 otherwise basically identical so you look at kits like this where it's a 4 109 00:08:06,360 --> 00:08:14,560 gig kit uh 2x2 GB DDR 3600 and cl9 110 00:08:11,360 --> 00:08:17,080 versus cl8 and you see upwards of a $20 111 00:08:14,560 --> 00:08:20,759 price Delta well it starts to make uh like an overclocking friendly kit like 112 00:08:19,080 --> 00:08:25,400 the blue that we're using today look like a pretty good value if you can get 113 00:08:23,120 --> 00:08:31,599 all of that extra Headroom out of it so let's have a look at our super pie 114 00:08:27,039 --> 00:08:34,000 calculation for our uh 7 115 00:08:31,599 --> 00:08:38,279 overclocking in this case a latency reduction more than like increasing the 116 00:08:36,039 --> 00:08:42,919 clock speed overclocking so it's a little bit different okay so we see a 117 00:08:40,719 --> 00:08:49,440 score of 14 118 00:08:44,760 --> 00:08:49,440 point no I missed it okay we'll try 119 00:08:53,080 --> 00:08:59,680 again I apologize for that I guess I could in theory have just turned off the 120 00:08:56,680 --> 00:09:01,880 camera and gone back and read it but um 121 00:08:59,680 --> 00:09:07,399 I didn't think of that until now and the camera's been running so we might as 122 00:09:03,200 --> 00:09:11,000 well roll with it okay so 14.29 that 123 00:09:07,399 --> 00:09:13,160 seems a little bit higher 124 00:09:11,000 --> 00:09:17,680 290 okay so I'll run my 3D Mark Benchmark and then I will be back in a 125 00:09:18,160 --> 00:09:25,600 moment okay well at 18654 3D marks we don't see a huge 126 00:09:23,279 --> 00:09:33,519 performance difference between our stock 999 27 timings and our tightened up 7 127 00:09:30,040 --> 00:09:35,600 7720 timings uh but this has been my 128 00:09:33,519 --> 00:09:39,320 DDR3 overclocking guide I mean if especially if you're done overclocking 129 00:09:37,200 --> 00:09:43,360 your CPU because that's where you'll see some big performance gains and you want 130 00:09:41,560 --> 00:09:47,640 to Tinker with it a little bit more I highly encourage you to go ahead try 131 00:09:45,240 --> 00:09:51,560 overclocking your memory uh don't forget to do stability checks though all along 132 00:09:50,279 --> 00:09:55,160 the way when you're doing any overclocking and bear in mind of course 133 00:09:53,760 --> 00:09:59,839 that my results today are not necessarily reproducible if you buy the 134 00:09:56,920 --> 00:10:06,920 same RAM I was lucky enough to get A2 5% overclock in terms of frequency or a 135 00:10:03,200 --> 00:10:09,399 significant uh two speed grade uh bump 136 00:10:06,920 --> 00:10:13,680 down in latency so Improvement in latency uh but you may not have the same 137 00:10:11,680 --> 00:10:18,040 results so the best way well not the best way but the way that I use to uh to 138 00:10:16,240 --> 00:10:24,399 check if memory is stable is I use prime 95 blend I use a custom one and then I 139 00:10:21,320 --> 00:10:25,640 go ahead and let it use all or most of 140 00:10:24,399 --> 00:10:31,680 the memory I have installed in the system so as long as this doesn't crash over a period of 8 to 12 hours you you 141 00:10:29,399 --> 00:10:36,120 can assume that your RAM is pretty much stable and then that way even if it's 142 00:10:34,120 --> 00:10:39,240 only a small Improvement you can have some performance Improvement so thank 143 00:10:37,760 --> 00:10:46,240 you for checking out my guide don't forget to subscribe to Linus techtips as 144 00:10:41,000 --> 00:10:46,240 well as ncixcom for all of my videos