|
Oh, the final cost of this i7-3770K 5ghz, 2400mhz CL8 RAM, SSD, NH-D14 build?
$500
- Intel i7-3770K: $209 @ Microcenter
- Gigabyte Z77X-UD5H: $79 @ Microcenter (CPU+Mobo Bundle deal)
- NH-D14 SE2011: $43 @ Ebay (Refurb)*
- GTX 460 768mb Twin Frozr II: $60 @ Geeks.com/Ebay/Amazon*
- Mushkin 2000MHZ CL7-10-8-27 1.65v 2x2GB 996902 PSC IC: Ebay @ $20*
- Intel X25-M G2 80GB SSD: $48 @ Ebay (New)*
- NZXT Source 210: $17 @ Microcenter
- Corsair CX500: $24 @ Newegg*
* Noctua supplied LGA1155 mount for free after showing ebay invoice. Those guys have an above and beyond support. They also sent me my AM3 mount, 3rd fan kit, sticker pack, and badge for free too. From Austria. * I've had my GTX 460 for a long time now, market price (I paid $100 3 years ago for it) * I really ended up paying $15 because I got refunded a bit for the RMA trouble. * I wish I bought a Samsung 830 64GB as my total storage instead. I went with 80GB for more space but that was dumb, as I'll never use that much space and the 830 is noticeably faster, you can find them similarly priced. * I would have bought the cx430 if it was $19 at the time, i wish I got the Antec NeoEco 620w $19 instead.
Bam.
My GPU is a little underwhelming, but a GTX 460 is more than powerful enough to max out SC2 with AA and ridiculousness. I intentionally kept my 768mb over my 1GB model because lower VRAM means higher overclock (which held true in my case as well).
There's a little extra I spent on accessories though, to be fair: 3rd Noctua NF-P14 Fan: $29 @ Microcenter (oucccch) Coollaboratory Liquid Ultra: $16 @ SidewinderPCs via Amazon (I actually paid more like $7 due to some Amazon points I had) 3 x Yate Loon Mediums: $15 @ Petras Tech Shop (I bought 5 for $20 + 5 shipping, 3 years ago) 2 x NZXT Havik 140mm Fans: Free. I paid $30 @ OCN Marketplace for an NZXT Havik 140 heatsink used, kept the fans for this computer, used a 120mm YL on the Havik 140 heatsink. NZXT Sleeved Blue 1m LED: $9 @ Directron NZXT Sentry 2 Fan Controller: $24 @ Newegg (this thing is awesome, the 5 fan probes on it are invaluable) 4 Feet of Rubber Case Trim: $3 @ BestByte for the window 13 x 13 Cell Cast Acrylic: Free @ Local Plasticier
$72 in Accessories.
Sanyo DP19640 19' 1360x768 Monitor: ~$80 (Gift Parents) Ducky DK-1087 Tenkeyless Mechanical Keyboard: ~$80 (Gift GF) Steelseries Kinzu: ~$20 (Gift Best Friend, thanks bobby) Steelseries Diablo 3 Mousepad: Free, Blizzcon 2011 Phillips FW-C250 3 CD Soundsystem (Receiver+2 big speakers): No idea, got it when I was like 10.
~$180 value in HID (same stuff I've used for a few years now)
I also made this i5-3570K build for a little more than $400 (no gpu, 8gb of ram, it's for video editing, yea the nzxt guardian 921rb is a piece of crap case, it made the price go over the 300s. the extreme4? it's a terrible motherboard). Customer is always right ;/
|
United Kingdom20263 Posts
|
When you aren't in a rush to build a computer, you can make sure you buy everything at the right prices.
Also helps to live in the US. Although from what I've looked at UK prices, you don't get bad prices on things. I've seen high end heatsinks for 40-60 pounds on some UK sites. Something to do with your currency?
A smart buyer doesn't buy everything new, and doesn't buy everything used either, but knows which parts are best bought where. Although really I'd only recommend to buy RAM used, everything else is better bought new. And ram only used because it's cheap, and because precious BBSE/PSC is no longer manufacturered, so the only way to buy it, is to buy RAM used.
No reason to buy more than 2x2GB of RAM except for video editing and professional applications.
|
Hmmm okay so what I've been up to the last few days:
So I got my 2400mhz ram overclock stable:
This is after 29 hours of prime95 (custom blend, 80%+ available ram tested, above normal priority, 27.9). Sorry I dont have an aida/maxxmem bench in there, i swore I did one but I think i deleted the picture (its okay, im at 5ghz now ill just do it later).
Now I don't know much about ram timings, but this is what I found out: - It's not about like a mix of ram timings affecting stability, it seems that each timing has it's own max at a given frequency. So it's not like Very Tight A and Very Loose B = Medium A Medium B in stability, there's just a max tightness for A, and a max tighteness for B. However, some timings are related.... - I didn't know exactly what to set my ram timings, like what would be good, so I basically just used the default timings at 1333mhz, which is much tighter than what they default to at 2400mhz. I worked from there, I basically changed a handful of timings at a time to their 1333 default and it either would or wouldn't boot, I'd figure out who the culprit is. Later, I repeated this process but with like 1 hour of prime95.
At first I was having issues with 8-11-10-30 1.8v, which is about where I started, eventually I realized loosening to 12 really helped out, and, I could even tighten everything else up, so I got to 8-12-8-27 1.8v. I realized that i wasn't failing any sooner if I set to 1.75v, but I was at 1.7v, so I set it to 1.75v. I kept failing what I set my secondary and tertiary timings, so I relaxed just the last major timing to 28, and suddenly I could get MUCH tighter timings on secondary and tertiary timings. I also read a few rules at places - twl should be cas - 1, the 4th timing should be like the first 3 added up...thats it really.
I was having a bit of difficulty figuring out how to completely tighten my secondary and tertiary timings (many people might say it isn't worth tuning them, but it is.... you can get faster ram with looser main timings and tighter secondary/tertiary than by just tightening the main ones), so after a couple of fails I made sure that my 2400 cl8-12-8-28 1.75v was 24 hour stable (it was, but this was important to test for first).
It could have been very difficult to optimize my ram timings, as in run 24 hours of prime95 for every single timing I changed, but I posted for help at OCN and a lot of people said my DRAM Refresh Intervals were way too tight, so I set those on auto, and I passed 29 hours of prime95! I mean I've tweaked around with most of the timings enough to know that all the tertiaries are as tight as they can possibly be, and most of the secondaries are as tight as can be.
For example, trfc had to be 80, it wasn't stable at 74. RRSR wasnt stable at 4, had to be 5, RWDRDD had to be 4 not 3.. that and the dram refresh intervals (9360/82 instead of 5200/45) and that was all I needed to be stable really, and the trfc, rrsr, and rwdr staight up couldn't boot so it wasn't like it took long to figure out that affected stability.
I'll post a maxxmem bench later but basically I get something like the following, which is actually
So now I'm working on my CPU overclock! So I need to not only try to make 5ghz 24/7 stable, but do it on offset voltage (I already 'know' i can do 5ghz, even if it takes 1.6v, so really I'll fine tune for the lowest voltage by using offset first. I also tend to work my CPU overclocks from the bottom up, so use a very low voltage, crash immediately, and slowly raise a step at a time until it passes more and more hours, until eventually Im
literallly a single step within the right voltage). I also wanted to make sure my PLL, VTT, and IMC voltages were as low as possible. This will result in lower temps, and well, it may not be super important, but it's important to me - i enjoy overclocking, i want to find the best of my system! It's not all about being practical, although there is a practical goal (maybe it isn't worth the time investment, but hey, i'll be sticking with these settings for years, so i think it's worth it, and its fun to me).
So I wasnt really sure where to begin with PLL voltage, so I set cpu voltage to something that seemed a bit extra, like 1.52v, and reduced PLL voltage to the lowest I've heard someone set it, 1.5v (turns out above 4.5ghz usually needs more than 1.5v PLL though). Bear in mind this is from the stock voltage of 1.8 PLL (i think PLL has to do with BCLK overclocking, which you don't do on ivy bridge except on LN2 runs, so no biggie in reducing it for an air overclock). I was failing overclocks, then I set it 1.6v and it seemed like the overclocks werent failing instantly anymore. 1.6PLL might not be able, but it seemed stable enough and if I have problems, I'll come back to it.
Then VTT/IMC voltage. I was a bit weary of messing with these, my ram overclock of 2400mhz is right on the limit of where people generally need to icnrease vtt/imc voltages. But, whatever. I reduced them both 3 .05 'steps' in my bios, to 1v/.875v, and it seemed stable, I did 2 hours of prime95. I lowered it another 3 steps to .975v/.86, and that was an instant crash. So, I keep 1v/.875v, and i'll come back if I fail p95 later.
As for CPU voltage, I needed to play with LLC settings to see what is really my best voltage. Now, your software voltage is actually an amalgamation, and average, of your vcore, as your voltage is always spiking up and down rapidly. It's just physics, electronics, it is impossible for your voltage to be constant. So what motherboard manufacturers do, is calculate a voltage based on those ups and downs. So the software voltage reading you get, is not at ALL accurate, and is very good at hiding the peaks and dips too you. So while you might see a max vcore of 1.4, in reality your peak offset was 1.42 and your voltage half the time is dipping to 1.35.
Now we all know what LLC is, and that it reduces vdroop, so your voltage is more consistent, but higher LLC settings means a higher peak voltage, and it means much higher temps as your voltage consistently sits at a higher level.
Now the idea with LLC, is that you can use less true voltage, than if you dont use LLC at all. But to really be able to tell what LLC setting is best for you (besides going off what is consistently worked for people who have actually used the tools to test this stuff), is using a digital multimeter to test for true vcore.
So, very quickly, I tested voltages on each LLC setting. I set +.25v offset on High LLC, and it would cause WHEA errors in p95 right away. It took about .34v offset with LLC to be stable. Extreme took about 2.8, and Turbo about .285v. But, what is the real voltage difference? Well, Extreme LLC not only got things very hot on my system, but put my real voltage very high, about 1.52. On High LLC, oddly enough, it didn't raise my voltage that much at all over what I set, but it required an insanely high offset of .32v to be stable.
Now by 'stable' I simply mean not crash and burn with p95 fails or whea errors within 30 minutes of testing, but like +.28v High was VERY MUCH more unstable than Extreme on .27v.
Oddly enough though, Turbo LLC at .285 was the most stable on the lowest real vcore, even though in software, it reports to be one of the highest vcores. It just goes to show, software voltage readings can't be trusted for nuances, only a general picture (set lowest vcore you can, but you dont really know what your real voltage is, and you dont really knwo what llc setting is really best for you). It's funny that Turbo ended up the best for me, as Turbo, the 2nd highest setting, is generally what people recommend when overclocking above 4.5ghz.
I mean I guess I shouldn't be surprised, but I was a little surprised how much better Turbo LLC was than the other settings - it's much cooler than Extreme, and it requires much less voltage than High does. But, the average person would think that Extreme LLC uses the least voltage, because it reports the lowest in software, and because the voltage offset I had to use with it was much lower.
I tried Turbo LLC with .28v, and that was pretty crash and burn right away. But .285v, and I'm already past 16 hours of prime95.
Now granted, if my overclock fails on the 23rd hour, I'll be ripping my hair out trying to figure out the culprit, but this is a very give and take game, and I think I was generous enough with my pll, vtt, and imc voltages. Yea, I'm living a bit dangerously by not testing each change on it's own!
Hopefully, tommorow, I'll be able to post some benches and results, showing you my overclock.
5ghz@1.488v CPU-Z / 1.499v Digital Multimeter
I think I'm actually going to try to do 5.1ghz@1.5-1.55v after 5ghz passes. It might be too hot to handle though, so I'll let this 5ghz finish and see if I can do 5.1 tommorow. I wouldn't consider my chip golden, but I'd definitely say it's an amazing chip.
|
5 ghz was stable! I passed 34+ hours of prime95, with no event viewer whealogger or kernel-whea errors (gotta make sure no errors in event viewer!), no issues at all. Here's some validation pics:
|
Here's some synthetic benchmarks of my system. I'll try to get some real-world benchmarks in a week or two (too busy, don't even have hots).
Benchmarks of my system:
i7-3770K Delidded Gigabyte Z77X-UD5H F15r (beta) Mushkin Enhanced Ridgeback 996902 PSC XDZ Ram IC 2x2GB custom made Intel X25-M G2 80GB GTX 460 768mb MSi Twin Frozr II 922mhz Core/2000mhz Mem 1.087v
Here's IMGUR link as opening image in new tab makes it harder to read: http://i.imgur.com/PqFoyPu.png
http://www.passmark.com/baselines/V8/display.php?id=6110645547 Passmark Rating 3,951 CPU Mark 13,475 2D Graphics Mark 1,100 3D Graphics Mark 3,365 Memory Mark 2,567 Disk Mark 1,323
I got a 42.86% overclock on my CPU (which is almost an identical difference between my CPU Passmark score and stock i7-3770K passmark score). My RAM outperforms ALL ram at stock, even crazy 2000mhz+ ram. For this kind of score on 2x4GB ram you'd have to pay an insane amount of money...
My CPU overclocked out performs the 3970X and has a similar performance score as the $1500 Xeon E5s.
|
United Kingdom20263 Posts
I just cringe when i think of the performance difference in single threaded applications between a 3770k@5ghz and a 3930k at stock (3.2 sandy bridge)
|
Scary numbers.
Could you, when your're done with 5.1 GHz ;P, post a video showing your FPS in SC 2 with something demanding like 50 Broodlords vs. 100 Stalkers?
|
United Kingdom20263 Posts
post a video
Pretty much any kind of recording (fraps, obs, xsplit) hurts performance quite a lot, some benchmarks would be good though.
Min/average
|
Okay, I'll work on some real benches today. Still got some stuff going on!
So my PSU blew out when my NZXT LED failed (it was a faulty LED). It 'derated', as in, the power it could put out was lowered by a lot. I did an advanced RMA with Corsair. Got a new cx500, and then that derated when another LED failed. I don't know if this means the CX500 is a really high quality PSU because my whole system didn't go down with it, or a really poor PSU because it blew when 'just' an LED went out, but I'm going to go with a higher quality PSU now.
Corsair's support has been amazing though, and I wouldn't hesitate to buy from them or recommend them, as what they may lack in quality (on a specifically low end PSU, for super cheap, mind you) they make up for in support. Blew out your PSU? Not a big deal, corsair will cover you. But, I was just a little tired of having to re-run cables every time, it takes 2-3 hours for me to do all of my cables...
So I bought an XFX PRO450w V1(rebadged seasonic s12iiB with jap chemicons) off ebay. I paid $45 for it, it's like new condition with original receipts. I slightly overpaid for it (should've stopped at $35) and then the xfx pro 550w went on newegg for $39 (again) the next day lol. Just got involved with a dumb bid war and already lost too many auctions at the time and needed a PSU.
Then the next day a Rosewill capstone modular 550w for $60 came - but get this - ebay gave me a $20 coupon for being a big seller lol. So I got a rosewill capstone modular 550w, brand new with original receipts, for $40.
As for warranties, the XFX I'm not too sure about. They have a weird policy, you have to register an item within 30 days of purchase or else you don't get their warranty but just a 1 year warranty. This xfx hasn't been registered at all, i'm not exactly sure how much xfx sticks with that registration policy. However, they also support used - original purchaser registers the product, then sells it, the next owner can actually register it and be covered. Which is pretty cool. Maybe they just expect no one to register, I don't know. Not many reports online of XFX RMAs (which could be a good thing, but there's always faulty stuff).
In regards to Rosewill, as long as you have the original proof of purchase (which you do need), you are covered. Which I have.
I've also started to heatshrinkless paracord (colonial blue) sleeve my cables as well. Despite having a semi-modular PSU, I wanted the cables to still be their black, full sleeve from where they come out and then snake behind the motherboard. So, I got some extensions, paracord, a pin remover tool, a lighter, knife & cardboard as a cutting board, and got to sleeving! It's pretty simple once you get the hang of it, just getting the hang took a long time lol. I've currently done all my case, thermal diodes, and front USB connectors. My tool broke though, so will need to wait on the new tool to arrive to finish the rest of the extensions.
On a side note, I also re-did my CLU thermal paste (i think this time around I used a bit more, like a tiny rice grain amount, and did both IHS and heatsink instead of just the IHS) and my temps dropped an additional few degrees, so now my max temp is 81*C (5ghz, 1.5v). I also put CLU on my gtx 460 and at 80% fan speed, dropped from 70*C to 63*C. So that's all pretty insane, actually, considering how high end PK3 is.
I'll throw up pics when sleeving is done, and of some benches.
|
Okay, huge update!
- I got some colonial blue paracord (i messaged lutr0 and an ebay seller about paracord sleeving and got a lot of help from them), 50 feet, and then rosewill pre-sleeved 24 pin, 2x6pin pci, and cpu extension (they were actually cheaper than non-sleeved, used extensions) and heatshrinkless paracord sleeved them all myself, as well as sleeving all my motherboard case cables. It was extremely frustrating first until I found some good russian video on how to do it, basically you dont need to melt down the paracord too much, just a brief melt and as soon as you see it melt, you do a light twist or pinching on it. It took a while because I let a tool drop on the ground and break, very fragile pin remover tools! For the motherboard case cables, I just used glue to hold them on.
In my opinion, paracord looks a million times better than sleeving, even mpdc.
- Well, as you know, I had PSU issues. I sold the XFX pro 450w on craigslist for $40, so I 'lost' $40 but oh well, I got to play with it a bit and it was fun to do that, and not a big deal to lose $5 when I only had to pay $40 for a rosewill capstone 550w modular, which is worth over $70. Now I got my rosewill capstone in. Very quiet, I can't hear it, very nice psu. I've done a bit of multimeter testing on it but I''ll be doing more.
- Got a 2nd monitor. Just some junk monitor to watch chats, skype, and hwinfo.
- I bought 2x2gb more of RAM, for a total of 8gb. This was necessary since I was dual screening. It's not so much that dual screen needs 8gb of ram, but if you are both gaming, streaming, and then pulling up a ton of tabs on chrome and have lots of extra programs, you tend to use more than 4gb of ram. I am also a bit low on VRAM, so my video was paging to ram and then my ram was maxing out, and then paging to my ssd, and i was like, losing 1fps man.
My ram overclock isn't as high anymore, I have to settle for about 8/11/8/28 1.75v. I'm currently on like hour 30 and going to just let this prime95 keep running until I get back from work tommorow, might as well, but hopefully it passes, and then yea that'll be awesome. I mean I can say for sure, there is no sign of degradation on my CPU, as I'm still stable on the same CPU overclock/voltages.
I did have to raise VTT/IMC voltages from how low I had it set before, to stock, in order to be stable though. And I think I was having a lot of difficulty in ram overclocks with the new ram because I was trying to run the good pair tighter than the 'bad' pair, and I guess that wasn't a stable thing to do even if the ram could handle it (like i failed looser on both sticks than what I got them both set to now).
Paid $25 for this PSC RAM that I'm going 2200mhz on though. Unfortunately no amount of work can make them do 2400mhz (well, I'm still going to try... just a little more). Like I tried CL10-15-10-35. Probably has to do with some sub timing, i dunno, but that sucks. I tried every trick in the book ;/
My RAM benches are significantly lower than before - i havent run them on this ram clock yet but I ran it on similar timings, and it's way lower than my 2400mhz CL8-12-8-28 benches. Oh well, I'm still sooo far ahead of even the best samsung timings so it's all good!
Pics:
|
how do u find these deals.
|
|
|
|