OC Report :: Intel i5 660 :: The Life & Times of Clarkbar

[3oh6]

Thank-you, thank-you...thank-you...alright that’s enough. Cut it out. Welcome to the Original OC Report I'm your host overclocking forum whore...3oh6. Don't you just love when you get woken up in the middle of the afternoon by your UPS driver because he knows you well enough to understand you are likely home, just sleeping off the hangover from two days ago. So he pounds on the door so loud that the Jumanji stampede in your dream wakes you up. That's a sign that you get too many deliveries. Perhaps when your FedEx driver invites you to his Christmas party and upon arrival you are asked to leave your keys in a fish bowl at the door is a stronger sign you have a computer hardware addiction. However you look at, I'm tapping a main vein today because I have a loaded needle and an itch that needs to be scratched.

To kick off the launch of the new Intel 32nm Clarkbar dual core processors, I have decided to do a type of OC Report. Consider this a benchmarker’s review. The goal, to show how good - or how bad - the new Clarkbars stack up to the common benchmarks that us losers run. Essentially the early Clarkbar numbers have been coming in over the past few weeks and the numbers aren't particularly good...with the exception of a small selection of benches. So today, that is what we look at, just how far off these highly hyped processors are.


If you are looking for architecture diagrams, insightful commentary on the cache structure, or other witty anecdotes regarding Intel's tick-tock crack pot marketing campaign that will undoubtedly spew from the mouths of countless 'media'; then go read those reviews. This is about one thing and one thing only, overclocking. I am here to piss some people off, cut some people down, and provide information that is actually useful to benchmarkers. Here's how this is going to play out...

Directory:

• Specs & Pricing :: Speculation & Hype
• Test Bed :: It's Classified...Still
• Boorriing :: The Meat & Potatoes
• Nothing Stuffy Here :: Real Benchmarking Results
• Conclusion :: Shallow Thoughts...By 3oh6

As always; there are a few caveats and disclaimers that need to be issued, lawyers and agents orders. Since my last OC Report almost five months ago I have not only taken on an agent but a lawyer as well, this overclocking is becoming big business and having both seems to be necessary. It is a good thing HWBot came up with a revised algorithm that will make overclocking cheaper to compete in because lawyer’s fees are adding up and my agent takes too big a cut of the millions I make from overclocking. Needless to say, I can talk whatever smack I want because Intel did not send me this processor. This literally fell out of the sky, landed in a big brown truck, and somehow managed to get my address taped to the top. But because I am such a nice guy, I have adhered to the Intel NDA which is why you are not reading this until now...whenever this is.

[3oh6]

As mentioned, I am such a big time professional overclocker that I managed to get my hands on some Intel specifications sheets regarding Clarkbar. I also have the secret to unlock fixed multi CPU's, proof the moon landing was faked, who really killed JFK, and the reason Dinos22 thinks the Geocities graphics that are posted in his threads are cool. Actually, I simply typed Clarkbar into Google, clicked on some links, and did a bit of work for your lazy asses. Needless to say, the information is still scarce as this is being typed before NDA - AKA not in the future - but will be updated with pricing as it starts to show up...aside from in Germany where you could have bought a processor for the past couple weeks. Oh wait, M.Beier said you couldn't buy them legally so if you did buy one, police will be outside your door shortly...MP5's locked & loaded no doubt.


The processor I am beating like a rented mule today is near the top of the initial offerings from Intel, the Core i5-660. The 660 clocks in at 3.33GHz at stock meaning it has a 25X multi, 26X with turbo enabled on both cores, and 27X with a single core enabled. Here is a brief chart of the rest of the Clarkbar processors that Intel has scheduled for launch on January 7.


They just couldn't let the Pentium name die, could they. I get that you have to leave a bottom bin for all the junk chips to fall into along with price point and all that other garbage...but did you have to keep the Pentium name around for yet another generation? Just let that crap die already. As for pricing; I won't speculate with pre-release prices, and you know what, I am not going to update this section as I initially planned. Screw it, go find the prices yourself dammit!

[3oh6]

There is no dedicated photo section because, well, it’s a processor. If you've seen one, you've seen them all. The setup for the benchmarking is going to be an EVGA P55 Classified. For those that know me, you know that P55 and myself aren't the best of friends. I mean, I say hi when I see it on the street or passing in the halls at work...but it isn't like I am inviting P55 out for drinks on the weekend. Okay, maybe that is being a little politically correct. I actually dream of beating the snot out of P55 in a dark alley with no one watching and no repercussions looming...I hate this platform. With that said, recent BIOS updates from Pete and the crew have started to turn me, specifically this motherboard. Here are some photos...

click for full size...

These Clarkbars are able to handle extreme sub-zero temperatures so I wasn't sure how my insulation was going to hold up as I normally don't bench below -110C (can't find a CPU to handle it). Needless to say, I thought I would go with my tried and tested kneaded eraser then go from there if I ran into issues. Learning from the X58 Classified I poured the insulation over into the first DIMM slot. As you can see in the first photo above, I insulated with eraser in-between the first and second slot. I then stuffed paper towel into the pins of the first DIMM slot seen in the second photo above. This has worked for me to keep moisture out of the first DIMM slot on the X58 Classy when going below -120C, so hopefully it works here. The reason it is a problem is that there is just no heat generated by the board east of the CPU socket so the cold spreads easily when going real cold...AMD styles.

click for full size...

These photos are just to indicate the insulation with the single stage mounted and my MMouse Rev 3. Now you may be wondering why I chose the MMouse Rev3 over the F1 but it is pretty simple. The F1 has a lot of mass and the LN2 isn't close to the base of the pot where as the MMouse Rev3 has a very thin base with LN2 nearly sitting on top of the IHS. With the thinner base, the MMouse was deemed to provide colder temps to the CPU with the pot full. There were also worries of a wicked cold boot bug and the idea of torching the F1 from -180C up to -60C every cold boot had my head spinning with the smell of propane. The mass of the MMouse is that of Paris or Nikki Hilton and would facilitate this process much better as well. Here's a full list of hardware used in the results...

Hardware used:
MB:            EVGA P55 Classified 200 (BIOS A45 / A51)
CPU:           Intel i5 660 ES (L934B369) // Intel i7 860 (3925B478)
CPU Cooling:   Chilly1 single stage (tuned by Ruffus) // MMouse Rev3 CU & LN2
GPU:           Biostar 8600GTS (SPi) // 3 x MSI GTS 250 (comparison benchmarks)
Memory:        OCZ Blade 2x2GB PC3-17000 8-9-8-24 (Elpida Hyper)
NF200 Cooling: Stock
PCH Cooling:   Stock
PWM Cooling:   Stock
PSU:           Corsair HX1000W (dual EPS connector used)
HD:            Seagate SATAII 80GB 8MB NCQ
OS:            Windows XP Pro SP3 (n'lightened by 3oh6) // Vista SP2 // Win 7

Ambient Temperature: 22-23C

There really isn't a lot to explain above. For the comparison benches in the Boorriing :: The Meat & Potatoes section, the 3 x MSI GTS 250's were used for all benching. This is not the most powerful GPU setup out there but it was the best I could get my hands on in the short time frame I was working with. If you want to see what three or four HD5870's do in this same setup, do the comparison yourself. I did what I could, don't get greedy.

[3oh6]

***Important Notice***
All of the results contained below were recorded during the very early days of Clarkbar...before NDA was even lifted. BIOS developments have drastically changed things that allow for much higher QPI ratios to be used, and exponentially increasing performance in a number of benchmarks below.

At this time I do not have the ability to update the benchmarks with higher QPI so this notice will remain in place until I do. Just keep this in mind when looking at the results.
***/Imptortant Notice***

This is basically the whole point of this thread. It is boring, somewhat useless, but someone had to do some empirical research so the community had a starting point with these processors. The speculation of how the Clarkbar performs can end here. For benching purposes, it is pretty useless. Well...there is no point in trying to paint a tin can as a shiny aluminum master piece. I am a reviewer and not some idiot who runs different OS's for a direct comparison or anything stupid like that. Testing was held to a strict scientific standard running identical clean OS's for each setup, imaging is my friend. The settings for each setup are listed on the charts but it came down to three configurations:

click for full size...

Hardware Setups:
White bars :: Clarkbar :: 238x21 @ 5.00GHz w/714MHz 5-5-5
Red bars   :: Clarkbar :: 238x25 @ 5.95GHz w/714MHz 5-5-5
Black bars :: i7 860   :: 209x22 @ 4.60GHz w/836MHz 6-7-6

For the Clarkbar runs I wanted to keep things the same with just a CPU frequency change so the only adjustment was simply the CPU multi. For the Lynnfield configuration I wanted to give Clarkbar a fighting chance in the memory intensive benches so I went with the 2:8 ratio and low 16x uncore. So keep this in mind when looking at the results...the Lynnfield config is crippled in the memory department. Let's start with the three 2D benches I ran, SPi 1M / SPi 32M / Hexus PiFast. The results done on phase (white / black) were run three times with results averaged for the charts below. I wanted to run the Lynnfield rig at 5GHz as well but didn't have enough LN2 so you get what you get .


Alright, so fittingly I start with pretty much the only benchmark worth running with this setup...SPi 1M. PiFast isn't too bad but the efficiency isn't great and it will take 7GHz+ runs to catch the best E8600's. With 1M, the WR has already fallen to Clarkbar and we have only gotten started. The biggest thing is memory bandwidth. The Clarkbars separate memory controller is an absolute dog forcing everyone to run 4:12 at any BCLK what so ever...Clarkbar is not memory clocking friendly. This is evident in the 32M times. It just happened to turn out that a Clarkbar at 5GHz is equal to a retarded, one legged, paraplegic Vietnam vet, i7 860 at 4.6GHz. Put it this way, I was messing with BCLK around 250 and with the CPU multi making up for the BCLK difference, at the same CPU frequency (5GHz) a Clarkbar with 60MHz faster memory was almost 20 seconds faster in 32M. That is a CPU that is absolutely begging for memory bandwidth...like an empty pocket crack head pulling on a dealer's pant leg.


Like sending sheep in to get slaughtered by wolves, AM3 eats Clarkbar for breakfast then hunts down its family and takes them home for lunch and dinner. There, the cat is out of the bad, AM3 likes memory bandwidth...if you are just learning that now, you are weak because it is not a secret for anyone who can go through a few screens at HWBot. As for 01, it is what it is. Comparing Clarkbar to an i7 isn't much of a comparison in 01. They both suck compared to C2D but the fact that Clarkbar isn't worse than Lynnfield at 4.6GHz means it isn't going to be absolutely horrible if you can run 01 at 7.5GHz .


This is where the 3D picture brightens slightly...by slightly I mean as dimly as the lights in the head of Forrest Gump, we are talking shoe size IQ dim. In 03 Clarkbar might have a shot at surpassing C2D if the clocks can get high enough. It is still early and only time will tell if 7GHz benchable comes with regularity on Clarkbar retails. As of right now, the consensus seems to be 6.5GHz range isn't a problem but even at that clock speed, 03 world records won't be on the bedpost of Clarkbar any time soon...I don't think they will anyway. But I have no real GPU's to test so what do I know. In 05, the story is the same as AM3, a beat down of epic proportions. Considering Bloomfield can bench 05 at 5.5GHz with regularity and Bloomfield is stronger than Lunnfield, Clarkbar is a waste here as well. Again, perhaps with better GPU's and at 6.5GHz it can outpace a substantial Bloomfield overclock, but I doubt it.


Now I know what you are thinking..."he is comparing a quad core with HT to a dual core with HT...that isn't fair", well guess what, I don't care. The CPU score doesn't make this big a difference in the final score of 06 anyway. Going through my spreadsheet, the game test results get spanked all up and down by the Lynnfield processor in 06 except Canyon Flight, which is weird because the other SM3.0 test really hates Clarkbar. In fact, Clarkbar at the 5.95GHz clocks scores 166.xxx FPS while the Lynnfield only averages 164.xxx FPS at 4.6GHz in Canyon Flight. That is about as close to a victory as we are going to see in 3D with Clarkbar though, and that isn't saying much. In Vantage the game tests are all close but game tests don't really scale with Vantage that much which is evident in the very little scaling going from 5GHz to 5.95GHz if you take CPU score into consideration. And if you are wondering, PhysX is disabled...no one here benches Vantage with PhysX enabled, right?

So, clock for clock it isn't even a contest...as they say, don't show up to a gun fight with a knife. But considering the higher clocks that Clarkbar can run there were questions about whether it could compete with Bloomfield. Considering it gets sent home like a dog with its tail between its legs by Lynnfield, I would say that train has left. So much for asking questions. Of course, this is just with 3 x GTS 250's but still...the numbers are there and they don't lie. Clarkbar will own the 1M world record as of the time this is posted, possibly PiFast, but the party ends there unfortunately. If someone can get a CPU that can run decent memory clocks at 6.75GHz+, there might be a chance for a 32M challenge with Clarkbar until Gulftown is released...but that scenario is highly unlikely IMO.

[3oh6]

Now that we know what Clarkbar is - and mostly - is not good at, time to exploit that. This section will likely be updated as I post respectable Clarkbar results, which for now are limited to 1M and PiFast. Although as you can see below, 32M isn't that terrible...especially if I can get memory clocks up even slightly with the same CPU clocks. My results are from a single ES, no hand picking or crap like that...just random luck of the draw like it is with any overclocking. You can trust that Intel isn't sponsoring this thread . This is a photo of the setup as it was for all the benchmarks posted here. Stupid P55 Classified doesn't have a PCI slot so I had to use a PCI-E card...how cutting edge of me.

click for full size...

I will try to keep this section updated with results going forward.

click for full size...

I had multiple sessions, some lasting as long as 4 hours without raising temps above -120C. At one point, I went at least 2 hours without having to cold boot so the temp on the pot probe never went above -165C. Needless to say, these chips seem not to have a cold bug with the proper BIOS or mod - depending on board manufacturer. I did have a wicked cold boot bug on the EVGA P55 Classified at -60C. That all changed when a mod was provided which lowered cold boot bug to a manageable -120C. Some people have experienced no cold boot bug at all with the mod so it will be a chip by chip basis. I also thought I would throw up some photos of how the insulation held up.

click for full size...

As expected, my insulation technique worked flawlessly so all those eraser haters out there, go to hell . When done properly it can handle a full pot of LN2 for hours no problem...perhaps you just don't know what you are doing. The first DIMM slot also held up very well but I will want to extend my paper towel layer on top just a bit so it completely covers that slot. Overall though, running with a pot full of LN2 for the first time took a bit of getting use to. Pretty boring to be honest. It was nice being able to chat in forums, and really concentrate on tweaks and times while benching though...no cold bug gets my

[3oh6]

There you have it, the burning question answered 10 minutes after NDA has lifted. Is Clarkbar worth all the hype? Absolutely not in my opinion. Sure, cold bug is finally gone on Intel but cold boot bugs will still creep up. Yes the chip overclocks really well but the downside is that it scales with cold, not voltage. So on a single stage, clocks will be limited by temps, on air, the same thing. Will air/water guys have as much fun overclocking these little dual cores as the C2D or Bloomfield/Lynnfield’s of more recent...it is hard to say. My gut tells me there will be a lot of 'hard walls' hit that Intel clockers aren't use to and will be frustrated by. Of course this is all speculation on my part because this i5 660 hasn't seen two minutes of ambient cooling...unless you consider when I am testing contact with my LN2 pot to be ambient cooling.


The fact of the matter is that the LGA1156 platform isn't the ideal platform for the extreme overclockers at the top of their game. The performance just isn't there. Keep in mind, the comparisons were done on the P55 platform, an already inferior platform to LGA1366. That makes these results even that much more telling. Clarkbar was the little kid on the block that was to bring LGA1156 that world class performance because of the incredible frequencies. It does clock high, no question, but efficiency just isn't there mostly due to memory limitations as I have advised throughout this post. The reason we have only seen 1M and CPU-Z validations thus far is because that is all it is good for. If retail chips help with memory clocks, or somehow blast past 7GHz in benchable clocks, that could change. But as of this moment, the graphic above sums up my feelings on Clarkbar.

[nickshih]

wow good result , detailed !! all the result done with 2gbx2 right ??

[Splave]

disappointing.

Poor sample, or are the newer chips dogs?

Think the mem is a cpu design issue or board?

[Hondacity]

excellent job jody

[Gautam]

I am a reviewer and not some idiot who runs different OS's for a direct comparison or anything stupid like that.

Best. OC Report. Ever.

[hiwa]

damn nice and clear report
thanks for sharing jody

[sandy]

OMG
nice

[LardArse]

There should be a Joey Clarkbar show on TV, this guy's just so entertaining

[Expat GriZ]

Jody; you've done it again. Great Report, and even though I have a 659 in the mail somewhere, I feel the same way about this platform.

Keep em coming...

[Oki]

I've enjoyed every thought, sentence, pics with candy as well.
Little boy will be a SPi 1M breaker aka gold taker - it can be saw from the pics above - lol.

[kiwi]

Damn it sucks hard in 3D benches I wonder if the same situation is in real world gaming

Nice review mate

[scooter.jay]

Very good review Jody well done lots of info cant wait to get me hands on one of these

[Vivi]

haha nice one jody, sick write up i loved it.

pitty the efficiency is so bad ;/

[George_o/c]

Hehe, you had me laughing really hard at some points, the text was just so damn' entertaining.
Good job all in all though Jody, great overclocking report, proper benchmarks testing and moreover clear thoughts

... You filthy full-time professional multi-million overclocker

[K404]

OMG the numbers are horrible... but thanks for sharing them

[qballe]

realy GOOD report
my compliments

[carpo93]

very nice man

[angoholic]

Very nice work, written in a funny way and nice results!

[massman]

[RPGWiZaRD]

You nailed it, no need to check other reports, this one shows it all. It's a shame about the 3D performance.