Intel "Devils Canyon" Launch - Reviews, OC Info, Availability

[Cali3350]

Microcenter still showing 30th for in store pickup, other places showing June 20th (Amazon) to 25(Newegg). Who knows when Intel will deliver these things right now.

 

[Homeles]

I've tried to simulate the performance of an i3 using an i7, but it's an imperfect test because cache size can't be altered on the fly... can it?

In what sense? Intel's LLCs are dynamic, so it's definitely possible. It's just a question of whether or not the BIOS or whatever will allow you to change it.

 

[crashtech]

In what sense? Intel's LLCs are dynamic, so it's definitely possible. It's just a question of whether or not the BIOS or whatever will allow you to change it.

Sorry for the continued OT, but I didn't know that functionality could be exposed to outside control. Pretty cool if so, I'd like to know which boards support it.

 

[DrMrLordX]

Also silent data corruption is just one possible problem with overclocking. I specifically highlighted it as an example because Prime/Linpak (et al) can't test for it.

True. But what could test for it, at least in the short term?

Bear in mind, I do understand that overclocks do break down over time. I've got some possible data corruption from my piece-o-crap Sempron 140 getting cranky, though that was after several years of (ab)use. I also suspected my x2-3600+ of corrupting some transcodes, which it probably did. Most of that was obvious enough (to me, anyway) that I could reasonably point the finger of blame on the least reliable parts of the system: that is, anything overclocked.

But let's say you put together a fresh system, overclock it, and want to know if you're going to get any "silent data corruption" right this instant . . . say, flaws in files edited in Photoshop or what have you. How do you propose people would test for that? I'm not trying to be rude or contentious; rather, I'm genuinely interested in knowing how you'd bench for that.

Do a bunch of sequential and random file copies and look for errors? Code a loop that builds a massive text file and look for erroneous characters? Do a bunch of double-precision FP calculations and look for more error than is the norm?

 

[TreVader]

I think over clocking is something that really is only truly useful to gamers. The few seconds you might gain encoding would easily be surpassed by going with a 8 core Xeon.


I don't know why anybody else would over clock. Spreadsheets loading ms faster I guess.

 

[DrMrLordX]

Overclocking used to turn hours long encode/transcode batch jobs into . . . slightly faster hours-long encode/transcode batch jobs. Sometimes significantly faster.

Cutting your encode time down by 33% or 40% is a big deal, or at least it was when it could take a looooong time to get it done. Most video encoding/transcoding is now fast enough that I think people would notice it less.

There are still situations where extra CPU horsepower, memory bandwidth, etc. can come in handy (3d rendering comes to mind).

 

[TerryMathews]

I think over clocking is something that really is only truly useful to gamers. The few seconds you might gain encoding would easily be surpassed by going with a 8 core Xeon.


I don't know why anybody else would over clock. Spreadsheets loading ms faster I guess.

overclocking is cheap/free. an 8 core chip is far from it.

 

[jj109]

I've tried to simulate the performance of an i3 using an i7, but it's an imperfect test because cache size can't be altered on the fly... can it?

I wouldn't worry about the L3 cache size... if going from 1.5 MB/core to 2 MB/core doesn't really change the performance, going up to 4 MB/core isn't going to change it either. With cache, you either don't have enough and adding cache decreases miss rate almost linearly, or you have enough and miss rate goes to near 0.

 

[StinkyPinky]

How long does it take to encode a movie these days? I remember circa 2001 it would take me about 7 hours on whatever CPU I had back then (P3-650 I think). By memory I think I upgraded to a AMD Athlon 1800 and it took about 1.5 hrs.

 

[Bman123]

Over clocking isn't what it use to be and users are just holding on to memories now. You can't get that cheap dual core and double the speed anymore. I'd rather build a stable system then tinker with that shit for hours trying to avoid bsod

 

[crashtech]

I wouldn't worry about the L3 cache size... if going from 1.5 MB/core to 2 MB/core doesn't really change the performance, going up to 4 MB/core isn't going to change it either. With cache, you either don't have enough and adding cache decreases miss rate almost linearly, or you have enough and miss rate goes to near 0.

Yeah, I think so too, but you'd be surprised at the minutiae people will fixate upon if you are telling them something they don't want to hear. I'd like to come up with an acceptable way to simulate various CPUs without actually needing to remount the HSF a bunch of times.

Anyway, again apologies for the OT digression.

 

[TerryMathews]

Over clocking isn't what it use to be and users are just holding on to memories now. You can't get that cheap dual core and double the speed anymore. I'd rather build a stable system then tinker with that shit for hours trying to avoid bsod

I suggest you read the L5639 thread here. I'll grant you it's harder than it used to be (Thanks Intel) but there can still be value extracted if you have proper motivation.

 

[Homeles]

I suggest you read the L5639 thread here. I'll grant you it's harder than it used to be (Thanks Intel) but there can still be value extracted if you have proper motivation.

It's gotten easier since Nehalem, no? Sandy Bridge and Ivy Bridge are a simple multiplier and voltage bump, then you run Linpack for an arbitrary amount of time before you decide it's good. Haswell complicates things slightly with the system agent being decoupled, but most people don't even tweak that from what I gather, and just keep them running at core clocks.

My Lynnfield rig was a pain in the ass to overclock, and I never bothered with it after my initial tinkering.