Intel i7-2700K has arrived

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
The performance of an i7-2600K 3.40GHz/3.80GHz Turbo vs i7-2700K 3.50GHz/3.90GHz Turbo is (nearly) identical, both are unlocked CPUs so increase the i7-2600K CPU Ratio (+1) = i7-2700K (SAME) CPU.

Further, unless you have Applications that actually use Hyper-Threading, e.g. Rendering, then neither the i7-2600K or i7-2700K will game better than the i5-2500K clock-per-clock. Differences vs i5-2500K are the Default CPU Multipliers (+1) for i7-2600K or (+2) for i7-2700K.

For an extra $55 does it makes no sense to get the i7-2700K? You decide.

NewEgg:
$370 Intel Core i7-2700K - http://www.newegg.com/Product/Product.aspx?Item=N82E16819115095
$315 Intel Core i7-2600K - http://www.newegg.com/Product/Product.aspx?Item=N82E16819115070
$215 Intel Core i5-2500K - http://www.newegg.com/Product/Product.aspx?Item=N82E16819115072

TigerDirect:
$370 Intel Core i7-2700K - http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=1380569&CatId=6991
$315 Intel Core i7-2600K - http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=7073159&CatId=6991
$220 Intel Core i5-2500K - http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=7073161&CatId=6988

MicroCenter:
$___ Intel Core i7-2700K - N/A yet ; will update
$280 Intel Core i7-2600K - http://www.microcenter.com/single_product_results.phtml?product_id=0354587
$180 Intel Core i5-2500K - http://www.microcenter.com/single_product_results.phtml?product_id=0354589

review - http://www.legitreviews.com/article/1751/1/
 
I'm not sure if you made up those numbers or that's acutal sampling, but there is definitely a difference between the 2500K and 2600K from those numbers. How is there not? Your average 2600K will clock 300 Mhz higher on (statistically) equal voltage. How does that not say on average that a 2600K is a better chip?

The way I see it, if you clock each CPU to the maximum stable clock given 75C and plot it in a distribution it should be a normal distribution (Bell curve). I'd expect the mean of a higher model chip to be higher than a lower model chip. (i.e. the average 2600K can achieve 5 GHz while staying under 75C while the average 2500K can achieve 4.7 GHz while staying under 75C)

I do not claim to be the sharpest knife in the drawer, so there's a good chance I'm being dumb and just not understanding you. But if those numbers you just posted above are real chip sampling, then that just shows everything I've been saying.
 
Sorry, making myself clearer requires too much time, I don't feel like writing a book to clearly make what IMO is more nonsense than useful to people.

The data was a copy/paste + average function in Excel. I never 'make stuff up'. One of many sources -> http://www.overclock.net/intel-motherboards/916189-official-intel-p67-z68-motherboard-comparison.html ; raw data.

i5-2500K and i7-2600K - data problem most folks who spend $200~$300+ on a MOBO will get the more expensive i7-2600K therefore the data, sampling, and OC'ing characteristics are skewed to the Higher Phase MOBO's which do deliver lower vCore's. Next, OC'ing vs Stability, IF you're going to Game it makes little sense to RISK the stability with a 5.0GHz CPU OC and playing Crysis for HOURS at a time -> BSOD. A good OC for gaming is 4.2GHz~4.5GHz MAX.

Bell curve: I assume all CPU's will have no problems being 'under-clocked', but I do know there are only a certain percentage that will OC: 4.4, 4.5,...,5.0, 5.1,5.2+ GHz If I took both the i7-2600K and i7-2700K then the vCore per 5.0GHz OC as I posted above would have the same range 1.30v~1.50v with most of the vCores in the 1.35v average.

Reality check, I have a i7-980X so to me ALL of the Sandy Bridge CPU are cheap.

The next Variable, 75C, 75C on what 'HSF' - water then what size radiators, Fan - how many and what CFM fans - limited by what noise factor. PROBLEM - every CPU is it's 'own thing' which is the reason for the vCore differences. Then 75C vs Ambient or Delt Temps, and to complicate more a vCore >1.50v will Thermally Limit you NO MATTER what cooling short of LN2. High vCore = High Temps.

Here's a typical take on OC'ing the Sandy Bridge:
"Results are representative of 100 D2 CPUs that were binned and tested for stability under load; these results will most likely represent retail CPUs.
1. Approximately 50% of CPUs can go up to 4.4~4.5 GHz
2. Approximately 40% of CPUs can go up to 4.6~4.7 GHz
3. Approximately 10% of CPUs can go up to 4.8~5 GHz (50+ multipliers are about 2% of this group)"


ref - http://hardforum.com/showthread.php?t=1578110

---

Things to remember that affect obtainable CPU OC'ing:
1. Physical characteristics (imperfections) for any Sandy Bridge (K) or (X) affect vCore
2. Phases affect vCore, vDroop
3. All MOBO's OC differently: Phases, effective use of Phases, BIOS voltage controls
3. HSF and Type of cooling
4. Ambient temperatures
5. Thermal Paste
6. Air flow
7. (USER): Installation, Settings, Experience
Etc.
 
I am a PhD student in Applied Economics so I do actually understand what sampling bias does to everything. If you regress clockability on processor model only you will get biased estimates of how much model affects clockability because people who buy expensive processors also buy expensive components everywhere else. I'm with you there. That's in a setting where you're pulling heterogenous observations from random real world users, though. You can't really deduce much from that because there's way too much variability in everything else, as you noted above with your list.

I was talking about in more of a controlled testing environment. If Tom's Hardware were to recieve 100 2500K's, 2600K's, and 2700K's each and test each one in a homogenous environment such as a test lab/bench, for example. I was just using a hypothetical experiment setup. Define your maximum CPU temp as whatever, say 75C. It doesn't matter what cooling you're using or how you're measuring it as long as it's identical on each separate trial. Clock each processor up to the maximum you can reach stable while staying under your temperature threshold. Record that and voltage if you want. Then you run all the maximum clocks while staying under 75C in a normal distrubtion. You're not running a range for each processor, just looking for the distribution of the maxes of all chips. You would probably have something like:

2500K: Min-4.4 Mean-4.7 Max-5.1
2600K: Min-4.5 Mean-4.9 Max-5.2
2700K: Min-4.7 Mean-5.0 Max-5.4

That is all hypothetical, but you would get back something like that. You think there would be no deviation between the models in a setting like that? I don't know the answer, so you probably know more about it than me. Do they not choose that top 10% and badge them as their highest models, next 40% something else, and on down, etc.?

 
Not true, water is denser than 'air' so the disposition of heat is considerably greater and the delta-T is increased. Since the increased cooling capacity reduces the affects of vCore relative to CPU temperature water yields higher CPU frequency at lower or equivalent temps (75C).

The next problem is cooling relative to ambient temperature. Clearly, the disposition of heat from water, in the best of circumstances the water can only be cooled to ambient, but not sub-ambient. Exceptions, unless use evaporation of vapor sprayed on the radiator (very risky) or to use some other form of refrigerant.

Bottom-Line, the CPUs are identical in every way except CPUID (Multipliers). No other Lithography changes, there's no increase in efficiency or reduced die (nm). So take a i7-2600K, encode different identifier, put it in a new shinny box and let some fool spend more money.

I don't take you as a fool. You're are kidding yourself if you believe Intel is going to redo millions of R&D, CAD work, testing, development over a +1 multiplier bump. It's like building a car from day 1 that can drive (go) 180 MPH, and then place (artificial) governors for 100MPH, 110MPH, 120MPH, etc and calling them different cars and charging more because the shinny label says anything different. The Sandy Bridge (K) or (X) has a user accessible governor - use it.

edit: IF Intel wanted to do the public a favor then they should have simply bumped the Multipliers on their non-K models. This is the craziest move of gullibility I've seen from Intel in many years and borders on moral turpitude.
 
Fair enough. I'm learning something here. You seem very knowledgeable on the subject. What would you say the maximum safe voltage would be for an i7? What I think you're saying is that under water cooling, which I'm considering, temperature is not necessarily a good measure for a safe overclock.

Also, if all chips are truly the same, what are your opinion on Extreme Editions? Aside from their (usually) higher cache, is there any benefit at all? Or is it really just a marketing ploy completely?
 
De Ja Vu anyone?
970s release when Sandy Bridge appeared. Just bump the cpu a notch up and slap a different tag on it.
Now the the tables reverse. Amd is releasing something lets do the same 2600k slight bump different tag
 
Assuming you are referring to the Sandy Bridge i7's (LGA 1155) with the non-K use Default vCore only, and the K's IMO vCore of 1.35v is reasonably safe for 7/24/365 assuming the temperatures are in the 60C~65C range or less on average. High temperatures do the real long-term damage, to a point, but never go over 1.50v even on water. On water specifically 1.40v~1.45v as long as the temperatures are in the similar 65C~70C range. Most of the time, unless you're running a very long render, you're nowhere near 100% load for any lengthy period as you are when running Prime95. So keep that in mind. Remember 75C is pretty hot, and as the SB temps grow higher so does the instability exponentially.

TIPS:
1. Less is more - I see all kinds of OC'ing Guides with all sorts of IMO CrAzY settings that end-up badly for most folks. Start off with manually setting the vCore, CPU Multiplier all cores e.g. 45, manually set BCLK 100MHz, and I always set my RAM manually and don't use XMP. Great Guide -> http://www.clunk.org.uk/forums/overclocking/39184-p67-sandy-bridge-overclocking-guide-beginners.html
2. Never Trust the Temps - if you ran (3) core temp Apps you'd see all different temps, and what screws-up most folks is they 'see 70C' but the CPU cores are actually spiking (like water boiling - erratic) > Tmax and you get an unexplained Post Failure or BSOD.

The 'problem' with LGA 1155 is its relatively 'locked' BCLK (Base Clock) frequency. So far only the SB/LGA 1155 tied-in everything {SATA, PCIe, USB, etc} (CHEAP) so while you 'could' theoretically use a BCLK of 150MHz you'd corrupt everything. Therefore, you're limited to BCLK of 98MHz~105MHz * CPU Multiplier. Unfortunately, to obtain greater than (34 * 105MHz = 3.57GHz + Turbo) it's required to have an Unlocked CPU Multiplier. In other words Sandy Bridge (K) or Bust!

In contrast, supposedly, the LGA 2011 aka Sandy Bridge Extreme are NOT tying-in {SATA, PCIe, USB, etc} to the BCLK. So those aforementioned restrictions no longer apply. This is similar to the prior LGA 1366 / X58. The i7-3930K (limited multiplier) and i7-3960X (unlimited multiplier) are ideally the most versatile SB-E CPU's, but the overall OC might very well render the same 'max OC'; verdicts still out.

Benefit recap, LGA 1155 an unlocked (K) is required - period to OC. LGA 2011 'seemingly' BCLK OC'ing; see -> http://www.tomshardware.com/reviews/core-i7-3960x-x79-performance,3026-4.html they had problems.

Worth it, depends on what CPU you talking about, the SB-E {6 core / 12 threads} $550 vs $1,000 with i7-3930K 12 MB vs i7-3960X 15 MB L3 probably not for the VAST majority. Frankly, I am very disappointed with the new LGA 2011. The LGA 1155 is a great performer for the cost. Perhaps the next rendition of 'Extreme' 22nm or Haswell will stand out.
 

how do you benchmark a cpu's gaming performance on that high resolution? What gpu were used?
nv-slide.gif
 
I understand the basics of overclocking on Nahelem and Sandy Bridge. I have my 920 right now turned to 4.0 Ghz. Scratch that. I HAD my 920 turned to 4.0 GHz until my motherboard (I think) called it quits earlier this week. I did that by, just as you say, turning the BCLK to 200. That makes it nice because my memory is then turned to 1600 which is it's rated speed, etc.

I really wishSB-E would come out because like I said, my system took a dump earlier this week so I need something soon, and I'd like to at least consider SB-E to replace my Nahelem. Hopefully Asus will RMA my board. I was considering waiting for Ivy because like you said, the SB-E is not that impressive, but there's no way I can live without a desktop for that long (poor me, right?)

What do you think the difference between limited multiplier and ulimited multiplier means in terms of real world limitations? There's no way I'm dropping a grand on a CPU, and if I were it wouldn't be SB-E. Just not impressive enough for me to throw around that much cash. Plus being a grad student doesn't exactly yield unlimited play funds. I was pretty heavily interested in the 3930K, but if it's going to have limited overclocking then no way. I'll buy SB and upgrade to Ivy when it comes out.

I'm with you on the GPU's though. I'm definitely picking up a pair of the GTX 680's or whatever they will be when they come out. My 570's scream a little bit with only 1.25GB of memory when thrown across 5760x1080.
 
So overall, what you're telling me is that Intel does NOT bin their chips? They do not pull their better chips and badge them as 2700K's, their second tier as 2600K's, third 2500K's and on down?

I know there is not architectural differences or lithography differences between any of the SB cpus. That's why they're part of the same family of CPU's. I just thought that their most cleanly produced, most efficient chips were badged as higher models, such as Extreme Editions in the Nahelem or SB-E case, or the 2700K/2600K in SB.

That's why I was saying that if you tested a bunch of each model and then compare the sample of 2700K's against the sample of 2600K's against the sample of 2500K's would you see any difference at all in terms of maximum achievable clocks across the samples? What you're telling me is no. I realize water and air are completely different animals, but as long as you tested all 3 samples on either water or air wouldn't you be able to tell some differences if there were any?

I really always thought the reason people ponied up for Extreme Editions was because they badged their best chips as EE's. If that's not true then you'd have to be swimming in money to buy one.
 


Hmm, aren't you forgetting about process improvements? Generally the longer a fab spends on a particular process node, they tweak it to yield better performance from the transistors. That could be the case here, where Intel with what - 2+ years on 32nm now - have improved it to the point where the bell curve shifts upwards.

And AMD's fab, GF, uses CTI - continual transistor improvement?? - to tweak their process nodes as well.

Finally, if the 2700K is on a new stepping, then there could be litho changes - small tweaks, errata fixes, etc. to bump up performance. Look how the Q6600 improved signficantly with the famous G0 stepping.
 
Yeah I edited my post, it seemed arrogant (980X), but I doubt BCLK 200MHz killed your MOBO. If the problem was immediate then 9/10 corrupted BIOS -> Clear CMOS. The SB-E is supposed to be released November 14, 2011. The initial tests though made me reconsider a complete make over, and correct I'm looking at the next gen GTX. Regarding vRAM - play Metro 2033, Dragon Age, F1 2100, Crysis Warhead and increase your AA 😉 A minimum of 10% and in some cases unplayable with 1.5GB cards; you'll discover a vRAM Bottleneck.

Regarding the SB-E K vs X, I have seen so many conflicting reports I don't know for certain. Purely Guessing - The only idea that makes any sense to me is a combination of CPU Multiplier and BCLK. If that's so I can foresee problems people improperly overclocking the (K) version. Possibly the reason Tom's review and other failed?!

Overclocking i7-2X00K vs i5-2500K comparing Apples and Oranges, my observations have been counter intuitive as I mentioned above.

Bin, no Intel tests but doesn't 'bin sort' to my understanding their official version of 'bin' is series by +1 multipliers, and not as 'quality' as you're suggesting; otherwise you won't be seeing the 50%, 40%, 10%, 2% as above. Similarly, 980X vs 990X you'll also see the assortment of vCores per OC on the same MOBO once you break 4.5GHz, and the obtainable OC seems to be identical (vCore) and same +1 multipliers bin.

@fazers_on_stun - It would be nice to think it were so, but the simplest example of overlapping AMD lithography is the CPU's with 'hidden cores' - one common print with similar CPUID encoding. Spread some fear that oooh not as good but those whom I know have been running 4 cores on their oddball 3 core CPU's with no problems and once core is unlocked CPU-z shows it as a completely different CPU...(comments) - http://www.cpu-world.com/CPUs/K10/AMD-Phenom%20II%20X4%20840%20-%20HDX840WFK42GM.html
 
Yea 5760x1080 just kills the memory on those cards. Crysis 2, Dragon Age 2, GTA 4, all memory bottlenecks. I was considering upgrading to 3GB 580's, but I'm scared that in a year or so even 3GB won't be enough across 3 monitors. That's why I decided to wait for the 600's. I'm hoping they have more memory and I've heard rumors that it will be a new kind of memory that is very fast.

You are obviously incredibly knowledegable on the subject. I was always under the impression that Intel bin sorted and put their best chips as their highest models. There are a lot of false rumors out there on the internet about that if it's not true.

On the mobo issue....I was just cruising along and my pc dies. Then I can get the powre to flicker on but then back off. I got it to actually boot one or two more times, but now it's basically dead. You think I just need to clear my CMOS?
 
What aggravates me the the screwing with 1.5GB vs 3GB GPUs! The cost of GDDR5 ($100 1.5GB) cannot be that much!

RE: MOBO - Post Failure suggests CMOS, but running midstream as described is a component failure and often that of a blown PSU. Especially if it 'tries' to reboot. IMO - go to the local 'Best Buy' if in US (good return policy or an 'open box') try a new PSU. Worst IF your PSU has a bad rail you're risking the: GPU(s), MOBO, CPU in particular.
 

The purpose was to demonstrate Hyper-Threading in relation to FPS. The conditions were the same, and I know most games have an assortment of limitations: Cores used, CPU/GPU (components), RAM, vRAM, OS, etc - or any variable you can imagine. The best to demonstrate is HT=on & HT=off which were posted. As I said, I know some games 'do' benefit from Hyper-Threading.

IMO - all of the Sandy Bridge are good. The cost to performance of the i5-2500K vs i7-2600K (+$100 USD) isn't justified for the 'typical' budget minded gamer. Instead for the same cost outlay using the (+$100 USD) with a superior GPU is a better choice.

Otherwise, those with unlimited budgets build an SR-2 and pour the money into it. I make that example to illustrate a point (+$100 USD) to many is considerable and my example is maximizing fixed/limited resources.

Which will Game faster?
Choice A - i5-2500K + GTX 580
Choice B - i7-2600K + GTX 570