Tom's Hardware i7-7700K OC Preview

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Headfoot

Diamond Member
Feb 28, 2008
4,444
641
126
Agreed, 4.4GHz is not a typical overclock for the 5820K.

On the early ones maybe, I built 3 rigs exactly the same as the one in my sig and every single one hit 4.4ghz. Some were easier than others to get there. These were all early-2015 models.

Mine can do 4.5 and maybe 4.6 if I give it a ton of voltage and step down RAM clocks and fiddle with BCLK+multiplier to get there.
 

dullard

Elite Member
May 21, 2001
24,998
3,326
126
We have quad cores for ~10 years now,what exactly is keeping the software from using at least these 4 threads?
Reality.

In specific cases, adding threads is easy and helpful for the software. But in far more cases, adding threads is neither easy nor helpful.

As a crude example, please write your next four replies that you'll have to my next four replies, in full. Oh wait, you can't do that until you actually hear the other side in the conversation? More than one thread (that is more than one reply) doesn't help in that situation. You can only start a conversation and then you wait. Having four, six, or even eight Anandtech forum windows open to write multiple different replies to me doesn't help at all until you get something from me to reply to.
 
  • Like
Reactions: VirtualLarry

krumme

Diamond Member
Oct 9, 2009
5,952
1,585
136
We have quad cores for ~10 years now,what exactly is keeping the software from using at least these 4 threads?
That the majority sold is 2 core models.
The incentive to program for more threads have been to little. Now the new consoles kind of force games using 4-6 threads.

Clearly there is a process and ipc wall. And its steep. With that wall and hopefully a compettive zen and zen 8c/16t in new consoles this deadwater can change.

The marginal cost for Intel giving i5 and i7 more cores for same price is minor. I know compettition from before. If zen is compettitve we will see a huge change. Intel have a lot more to offer than what they get away with now. They can easily step up. There isnt a single technical barrier its just a decision already laid out in scenarios.
 
Mar 10, 2006
11,715
2,012
126
The only people that should be interested in increasing IPC is shareholders be it IBM or Intel, and even AMD.
The only purpose it have is monopolizing the market by keeping the software from using more threads and making it more expensive for entry by new competitors and competing products. Classic top management strategy.
Its a shame. And what we have now is just pathetic improvement of performance per year. Like 10% improvement what we had 10 or 25 years ago. And ofcource at high prices. Pretty obvious looking at the margins and record profit.
An Arm a73 cpu core is 0.65mm2 on 10nm. It just shows how idiotic this monopoly race for IPC have been. Next step 5 wide x86 cores used with AVX 512 bit. Way to go for the shareholders.
Unfortunately we had the crappy bd core, and Moar cores is still parroted. But what a bunch of selfinflicted pain it is.
Looking in hindsight a oc sandybridge perf is about excactly the same performance than a oc KL. Same stuff with a lot of PR lipstick.
Go 10 or 20 years back and people expected more from 6 months progress. Not 6 years.
X86 is now a ugly cashcow with pathetic performance for the customers instead of beeing a place of progress and innovation.

So, you think that software developers aren't threading their software because most consumer CPUs don't come with tons of cores?

This explanation has no basis in reality. Games are primarily designed for systems with eight hardware cores (Xbox One and PS4).

I also recommend reading this: https://en.wikipedia.org/wiki/Amdahl's_law
 

TheELF

Diamond Member
Dec 22, 2012
3,967
720
126
That the majority sold is 2 core models.
The incentive to program for more threads have been to little. Now the new consoles kind of force games using 4-6 threads.
2012 game that was supossed to come out in 2009!
60 threads...if software devs wanted they would have,or as dullard correctly explained if there where anything to gain from it they would do it.
https://en.wikipedia.org/wiki/Max_Payne_3#Development

iY8IsKn.jpg
 

krumme

Diamond Member
Oct 9, 2009
5,952
1,585
136
So, you think that software developers aren't threading their software because most consumer CPUs don't come with tons of cores?

This explanation has no basis in reality. Games are primarily designed for systems with eight hardware cores (Xbox One and PS4).

I also recommend reading this: https://en.wikipedia.org/wiki/Amdahl's_law

As andrei have shown android and chrome can use 4 or even up to 8 cores for even stuff like browsing.
Thats what make surfing okey on a a53 core. Because there is 4 of them. And the core production cost mm2 is pretty much next to nill. And the small cores is sipping power on a cheap leaky 28nm process.
Its happening all the time just not under x86 reign.
But ofcource its highly expensive to write for more threads. And the consoles uses the 6 here and it have takes years to get here. But the alternative is just worse. Process cost is exploding and ipc have hit a wall.
Amdahl "law" and all the usual talk is to be interpreted in their historical context and time and have to pave way for new technology in programming. Eg. Games without a main thread.
Dice made bf3 use 6 cores when others was stuck in single thread hell and the game was a blast. They are resting on those laurels today raking in profit for ea.
There will be a day when someone releases game that uses 16 cores and then it will be a blast again.
 

2is

Diamond Member
Apr 8, 2012
4,281
131
106
There is one thing though, this struggle to get this one arch anywhere, you got Core ranging from 2C/4t tablets in ~3 watts to desktop class 4C/8T ~95watts to HEDT ~140watts. Maybe they're stretching it a bit thin and maybe the design decisions made both on the arch and the process node hurts potential peak performance somewhat. One core to rule them all may be too thin of a gamble come competition.

Yeah, I don't doubt that is partially the reason why we see relatively lack luster performance gains, but I think the reason Intel went this route goes back to demand. There just isn't enough of it in the desktop space for them to dedicate/develop an entire product line towards.
 
  • Like
Reactions: cytg111

krumme

Diamond Member
Oct 9, 2009
5,952
1,585
136
HEDT is just like a normal desktop processor with a tacked on gpu.
Its the i7 920 replacement!
But since there is zero compettition a new segment with a new price is created. It sell in small numbers so its taken from server dies.
If amd can somehow get a zen on tsmc 10nm or gf/tsmc 7nm higher up in freq then HEDT as segment and word is gone again.
 

superstition

Platinum Member
Feb 2, 2008
2,219
221
101
my i5 6600K needs 1.44V for 4.8Ghz, 1.35V for 4.7Ghz
the i7 7700K needs 1.3V for 4.8Ghz, and will likely clock a bit higher at 1.35V and 1.4V respectively
Perhaps your i5, being an i5, isn't particularly great silicon?

Also, Intel may be using a TIM that's less of a bottleneck.
 

VirtualLarry

No Lifer
Aug 25, 2001
56,227
9,990
126
You want to see "not particularly great silicon"... my i5-6400 (the "Gimp bin", 2.7Ghz), requires 1.410V vcore, just to hit 4.455Ghz stable. I tried 4.51Ghz at that voltage, but it just wasn't fully stable.

One advantage that the "K" CPUs have, is that they can leave the uncore clock multi at something lower, like 42x (4.2Ghz uncore), while being able to push the Core clock higher, like 4.6-4.7Ghz.

With my locked i5, doing a BCLK OC, pushes the uncore to the same freq. as the core, which may in fact be the weak point for Skylake CPUs.
 
Mar 10, 2006
11,715
2,012
126
You want to see "not particularly great silicon"... my i5-6400 (the "Gimp bin", 2.7Ghz), requires 1.410V vcore, just to hit 4.455Ghz stable. I tried 4.51Ghz at that voltage, but it just wasn't fully stable.

One advantage that the "K" CPUs have, is that they can leave the uncore clock multi at something lower, like 42x (4.2Ghz uncore), while being able to push the Core clock higher, like 4.6-4.7Ghz.

With my locked i5, doing a BCLK OC, pushes the uncore to the same freq. as the core, which may in fact be the weak point for Skylake CPUs.

4.45 on the uncore is a LOT :)
 

cytg111

Lifer
Mar 17, 2008
23,049
12,719
136
You want to see "not particularly great silicon"... my i5-6400 (the "Gimp bin", 2.7Ghz), requires 1.410V vcore, just to hit 4.455Ghz stable. I tried 4.51Ghz at that voltage, but it just wasn't fully stable.

One advantage that the "K" CPUs have, is that they can leave the uncore clock multi at something lower, like 42x (4.2Ghz uncore), while being able to push the Core clock higher, like 4.6-4.7Ghz.

With my locked i5, doing a BCLK OC, pushes the uncore to the same freq. as the core, which may in fact be the weak point for Skylake CPUs.

You guys are walking a thin line in the "fully stable" category IMO.
I would do what you did, Larry, and then dial it down another at least 200Mhz as a margin, cause, have you thought of everything? ambient temps, loadworks, a little degradation, case and fans collecting dust obstruction airways? Just saying :).
 
  • Like
Reactions: Arachnotronic

asendra

Member
Nov 4, 2012
156
12
81
My 4770k only does 4,2~3 stable at 1,3v. Now that was a disappointment...
I could put more v but then I'm limited by temperatures even with my H90 AIO. (If I could drop ~20° by delidding i could maybe push it more?? but then, is it even worth the risk if I can only get 100-200mhz more?)
 

Lepton87

Platinum Member
Jul 28, 2009
2,544
9
81
On early batches or low voltages, maybe. Last batch before BDW-E was averaging like 4.7Ghz at about 1.3V i believe.
I got mine as early as possible when it has just become available so it's one of the first batches. If the last samples over-clock to 4.7GHz then I might just replace mine with one of those. Very impressive gains from the process maturity.
 
Last edited:

Majcric

Golden Member
May 3, 2011
1,369
37
91
My 4770k only does 4,2~3 stable at 1,3v. Now that was a disappointment...
I could put more v but then I'm limited by temperatures even with my H90 AIO. (If I could drop ~20° by delidding i could maybe push it more?? but then, is it even worth the risk if I can only get 100-200mhz more?)

Honestly, it not worth the risk. Exception being you love to tinker and have money to spare.
 

WhoBeDaPlaya

Diamond Member
Sep 15, 2000
7,414
401
126
Honestly, it not worth the risk. Exception being you love to tinker and have money to spare.
It'd be less trouble to swap the 4770K for a 4790K (assuming you have to stick with Haswell).
I'm happy that both of my launch 4770Ks manage to hit 4.6GHz, but a 3rd one that a friend bought at the same time was not so lucky (~4.3GHz IIRC).
 

2is

Diamond Member
Apr 8, 2012
4,281
131
106
It'd be less trouble to swap the 4770K for a 4790K (assuming you have to stick with Haswell).
I'm happy that both of my launch 4770Ks manage to hit 4.6GHz, but a 3rd one that a friend bought at the same time was not so lucky (~4.3GHz IIRC).

I don't think it would be less trouble, nor would it even be worth the expense.
 

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
My 4770k only does 4,2~3 stable at 1,3v. Now that was a disappointment...
I could put more v but then I'm limited by temperatures even with my H90 AIO. (If I could drop ~20° by delidding i could maybe push it more?? but then, is it even worth the risk if I can only get 100-200mhz more?)

My 4770k must have come of the same crappy wafer as yours!
 

pj-

Senior member
May 5, 2015
481
249
116
On the early ones maybe, I built 3 rigs exactly the same as the one in my sig and every single one hit 4.4ghz. Some were easier than others to get there. These were all early-2015 models.

Mine can do 4.5 and maybe 4.6 if I give it a ton of voltage and step down RAM clocks and fiddle with BCLK+multiplier to get there.

I got a 5820k in mid 2015 and 4.4ghz is not exactly easy for it. I forget the exact voltages but the difference in heat and temps under load between 4.2 and 4.4 was obvious and not worth it. I wouldn't be surprised if it was pulling an extra 30-50w for those 200mhz.
 

Headfoot

Diamond Member
Feb 28, 2008
4,444
641
126
I got a 5820k in mid 2015 and 4.4ghz is not exactly easy for it. I forget the exact voltages but the difference in heat and temps under load between 4.2 and 4.4 was obvious and not worth it. I wouldn't be surprised if it was pulling an extra 30-50w for those 200mhz.
Yours is probably below average. Every single one of those 3 machines I built on the 5820k hit 4.4 fine. I hit the wall where I had to jack up voltage and heat dramatically starting at 4.5, so I dropped all of them back down to 4.4 I could do 4.4 at 1.25v+vdroop correction to 1.3v+vdroop correction vcore on all of them. All of these were on the cheapest Asrock X99 board, which is still pretty nice. I find it hard to believe I got 3 above average chips in a row

At 4.2 you have to start looking at BCLK, RAM dividers, uncore, etc. It definitely gets more complex from 4.2 and up
 
Last edited:

WhoBeDaPlaya

Diamond Member
Sep 15, 2000
7,414
401
126
I don't think it would be less trouble, nor would it even be worth the expense.
Less trouble than delidding I mean. Expense is subjective I suppose, depending on how much he can pawn his 4770K for and how much he can buy the 4790K for.
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
Reality.

In specific cases, adding threads is easy and helpful for the software. But in far more cases, adding threads is neither easy nor helpful.

As a crude example, please write your next four replies that you'll have to my next four replies, in full. Oh wait, you can't do that until you actually hear the other side in the conversation? More than one thread (that is more than one reply) doesn't help in that situation. You can only start a conversation and then you wait. Having four, six, or even eight Anandtech forum windows open to write multiple different replies to me doesn't help at all until you get something from me to reply to.


You can have a conversation with 4 different people (4x cores) concurrently (MultiTask/Throughput).

That is also what Async Compute does in GPUs.