Should Intel Have Killed OCing With Sandy?

[poofyhairguy]

Right around the Sandy time period Intel locked down OCing and made it so only a couple of higher-end models could OC. This had the effect of preventing lower-end models from cannibalizing sales of upper end models, but recently I have been wondering if maybe they should have gone further.

No chips have really OCed further than Sandy could (some post-Sandy chips were worse actually), so because we have OCable Sandy chips the whole question is "what is the IPC improvement?" every generation and that question always has a disappointing answer. Instead we justify upgrades based on RAM speeds, or board features, or something else like that.

If Intel never would have made the K series, then newer chips could be upgrades in IPC and clock speed. I know that I would have upgraded a 2600 by now for pure frequency increases but I won't upgrade my 2600k because the IPC increases alone aren't enough. It seems like as long as Intel sells K chips they have lost the ability to give meaningful upgrades without greatly increasing their own costs (by providing more cores for example) which has sucked a lot of enthusiasm out of the segment.

I know Intel likes the positive PR of extreme overclocking, and board partners love OCing (otherwise why buy a fancy board?), but would Intel have been better off just killing all OCing at the end of 2010?

 

[Arachnotronic]

No.

 

[sm625]

There is no competition, so it makes the most business sense to lock the clocks on all their SKUs. However, had they done this then I believe we would have had workarounds by now which would have allowed the overclocking of all SKUs. So I for one wish they had gotten rid of all K models.

 

[Sable]

No.

Is the correct answer. /thread

 

[monkeydelmagico]

No chips have really OCed further than Sandy could

What? Ivy then Haswell both obtained higher OC frequencies than Sandy. Broadwell didn't do too well but the IPC made up for it. Skylake is almost back to holding high frequency crown for Intel processors (excluding P4's)

When you combine skylake higher frequency AND IPC it's a fair amount more computing power than Sandy. Just because some programs don't take advantage of it isn't Intels fault nor should it spell the death of OC'ing.

GPU's are making leaps forward. At some point they will "catch up" and the ability to OC your chip will make a realy difference again. My crystal ball is foggy on the exact date of this event.....

 

[poofyhairguy]

What? Ivy then Haswell both obtained higher OC frequencies than Sandy.

I have seen some people clock Sandy up to 5ghz on air. I don't think anything since has broken this 5GHz on air wall was my point.

When you combine skylake higher frequency AND IPC it's a fair amount more computing power than Sandy. Just because some programs don't take advantage of it isn't Intels fault nor should it spell the death of OC'ing.

It seems like the death of OCing more has to do with the fact that the massive headroom on the Sandys (2600k- 3.8 GHZ to 5 GHZ) has been eaten by Intel trying to sell upgrades (6700k 4.2GHz to 5 GHz).

GPU's are making leaps forward. At some point they will "catch up" and the ability to OC your chip will make a realy difference again. My crystal ball is foggy on the exact date of this event.....

For there to be some sort of sea change with overclocking we need chips that go beyond 5GHz on air, but thanks to smaller dies (and therefore more concentrated heat) this seems almost impossible.

Is there some innovation coming down the road to blast past the 5GHz wall that I am missing? I thought it was basically a physics limit.

 

[BonzaiDuck]

I have seen some people clock Sandy up to 5ghz on air. I don't think anything since has broken this 5GHz on air wall was my point.



It seems like the death of OCing more has to do with the fact that the massive headroom on the Sandys (2600k- 3.8 GHZ to 5 GHZ) has been eaten by Intel trying to sell upgrades (6700k 4.2GHz to 5 GHz).



For there to be some sort of sea change with overclocking we need chips that go beyond 5GHz on air, but thanks to smaller dies (and therefore more concentrated heat) this seems almost impossible.

Is there some innovation coming down the road to blast past the 5GHz wall that I am missing? I thought it was basically a physics limit.

There is at least a handful of chip design features since Sandy which enhance performance. Clock-speed is not the accurate measurement of performance or the potential for it.

I have a paraplegic friend who is a physicist. His computing activity is more and more limited because of a progressive difficulty getting from bed to wheel-chair. We'd thought of all sorts of solutions: a monitor mounted to his ceiling, for instance.

He had speculated back at the advent of Ivy Bridge that further reducing the lithography would indeed lead to problems associated with quantum physics -- for instance, current leakage or electron migration. But it apparently hasn't occurred. Another mutual friend and electronics tech suggested that this derives from the lower power requirements of each chip generation.

But moving from 14nm to something smaller brings us down to the "atomic" level. I'd only guess that 7nm is a limit. But who the hell knows for sure? It only makes sense, though.

 

[frozentundra123456]

I think Intel at one time predicted they could reach 5nm. But they have not done well with 14nm, so who knows??

 

[LTC8K6]

I think Intel at one time predicted they could reach 5nm. But they have not done well with 14nm, so who knows??

They apparently have done well with 14nm.

http://forums.anandtech.com/showpost.php?p=38372021&postcount=203

 

[ClockHound]

I have seen some people clock Sandy up to 5ghz on air. I don't think anything since has broken this 5GHz on air wall was my point.



It seems like the death of OCing more has to do with the fact that the massive headroom on the Sandys (2600k- 3.8 GHZ to 5 GHZ) has been eaten by Intel trying to sell upgrades (6700k 4.2GHz to 5 GHz).



For there to be some sort of sea change with overclocking we need chips that go beyond 5GHz on air, but thanks to smaller dies (and therefore more concentrated heat) this seems almost impossible.

Is there some innovation coming down the road to blast past the 5GHz wall that I am missing? I thought it was basically a physics limit.

You're only missing about 32nm. ;-)

Intel's 32nm process was the last great OC'ing platform. With all the low hanging IPC fruit gobbled up, we're lucky that they gave us 5-10% IPC boost per gen since then. Impressive, really - in a depressing sort of way.

Since the mobile revolution of 2012 (blame the Mayans), its been about performance per watt. Higher and higher clocks, sadly, is not a step on the power efficient ladder.

Post 32nm, Intel set themselves on a course to carve out a huge market in mobile, which worked about as well as clocking a 5775c to 5.5Ghz on air.

Could they release a 14nm chip that clocks at 5Ghz on air? Yes, they could. But, the tradeoffs would be uneconomic. The desktop enthusiast market isn't near big enough to warrant it. Unless you're prepared to pay $50,000 per chip with a MOQ around 10,000.

 

[poofyhairguy]

What I (we?) really want is a 4GHz chip that performs the same as the 5GHz Sandy and still has OC headroom.

Failing that, 6 current cores at what the 2600k was new. That would be a real upgrade that doesn't depend on IPC. I figured we would be there by now.

 

[jpiniero]

Could they release a 14nm chip that clocks at 5Ghz on air? Yes, they could. But, the tradeoffs would be uneconomic. The desktop enthusiast market isn't near big enough to warrant it. Unless you're prepared to pay $50,000 per chip with a MOQ around 10,000.

There are a few other groups (namely the NSA and Hedge Fund types) that would also love to see higher clock speeds. That might be enough, along with desktop enthusiasts to make it worth Intel's while. But you won't be seeing anything cheap.

 

[YBS1]

What I (we?) really want is a 4GHz chip that performs the same as the 5GHz Sandy and still has OC headroom.

Failing that, 6 current cores at what the 2600k was new. That would be a real upgrade that doesn't depend on IPC. I figured we would be there by now.

I'm not so sure that a 4GHz Skylake doesn't already perform as well or better than a 5GHz Sandy Bridge. What you're asking for is likely already here.

 

[SPBHM]

before sandy bridge I used to overclock all my CPUs, so for me they killed OC with sandy bridge...

having to buy more expensive chipsets and CPUs just for not being completely locked out from OC? nope.


but sure, a 4790K would look even better at 4.2GHz if 2600K/3770K were locked to 3.5GHz or so...

 

[Arachnotronic]

I'm not so sure that a 4GHz Skylake doesn't already perform as well or better than a 5GHz Sandy Bridge. What you're asking for is likely already here.

yep, if you assume a ~25% IPC boost in going from SNB to SKL, a 4GHz SKL = 5GHz SNB. And the 6700K has OC'ing headroom to boot.

 

[greatnoob]

What you are saying is "I want something that reaches 5ghz" yet you haven't realised that clock speeds themselves are worthless as any Skylake 4ghz chip would most likely be much faster than your 5ghz Sandy Bridge chip.

"I want to have a chip that clocks to insane numbers to show off even if it has lower performance than a chip that's clocked lower" isn't a good enough excuse to waste R&D resources and release a product to a niche of a niche group for obvious reasons.

 

[Sheep221]

What I (we?) really want is a 4GHz chip that performs the same as the 5GHz Sandy and still has OC headroom.

Failing that, 6 current cores at what the 2600k was new. That would be a real upgrade that doesn't depend on IPC. I figured we would be there by now.

You are aware that 6700K stock is way faster than 2600K OC?
And...
You are mixing enthusiast/workstation six core i7s with mainstream quad core ones.
Product line with 2600K goes like this 875K>2600K(2700K)>3770K>4770K(4790K)>5775C>6700K

 

[IntelUser2000]

Is there some innovation coming down the road to blast past the 5GHz wall that I am missing? I thought it was basically a physics limit.

Quite likely there's a fundamental limit that won't be breached.

http://hwbot.org/benchmark/cpu_frequency

FX-8370 overclocked to 8722.78MHz is the fastest. And you see Celeron D(Prescott) chips there. Pretty much all frequency-focused chips are on the top 10 list.

If that's what you can do with exotic cooling with extreme pipeline stages, more sane and realistic designs will fall short of that, and air of course much further.

There were talks about Prescott chips at 5GHz with 400MHz speed bumps every quarter or two until 6-6.5GHz. Even the engineers at Intel clearly didn't know everything about what they were doing, if the same architecture with exotic cooling with simplest arrangement(Celeron) can only do 8.7GHz.

Air overclock limits are interestingly all at 5GHz too, Prescott, FX chips, Sandy Bridge. Different architecture, different company, different process, all same result. That's quite telling. Did you hear that folks? Screech... that's the sound of CPU advancement and process coming to an abrupt stop.

 

[BonzaiDuck]

before sandy bridge I used to overclock all my CPUs, so for me they killed OC with sandy bridge...

having to buy more expensive chipsets and CPUs just for not being completely locked out from OC? nope.


but sure, a 4790K would look even better at 4.2GHz if 2600K/3770K were locked to 3.5GHz or so...

Can we call this the "K-chip issue?" I've been buying Intel CPUs since before the '90s. As the product line developed and changed, the relative prices have not. I remember -- I think -- when a Pentium III* had something close to a $500 price-tag. I'm sure that was it, because my cousin and I bought three or four of them at lower prices which turned out to be counterfeit -- partly Intel's fault for making the difference between CPU models dependent on disabling a connection. All the chips rolling off the production line were manufactured for top-end speed specs, and then they chose to disable the full chip speed and market as lesser models to a less-demanding market segment.

Further inquiry presented us with an "Intel Authorized Dealer" application, and a clear understanding that genuine unmodified CPUs would only be sold by those participants at certain prices and never less.

So . . . if you want to play this game with intel as part of an enthusiast market segment, what is so forbidding about $350 in 2016 dollars for a Skylake K chip?

* Correction: the counterfeits were mature-model Pentium II's. But that's not pertinent to this as a factor of anything.

 

[superstition]

Electromigration becomes an increasing issue as process nodes shrink, eh?

The question posed in this topic is valid enough, though, since AMD's FX, even overclocked to 5 GHz, cannot challenge recent Intel designs in single thread performance.

Intel really doesn't have to offer overclocking. However, one big thing is the delidding thing, which Intel promotes by using inefficient TIM. (Note: The Devil's Canyon TIM was sold to the enthusiast community with the promise of improvement but analyses found nothing significant.) Delidding, which can quickly destroy a chip — prompting an additional chip sale, is a beautiful thing to an Intel executive. Why wouldn't it be?* The "two sale" strategy has been rather big lately, like selling the Anniversary Pentium to people who will realize it doesn't really cut it in some games because of just having two threads. Intel also made delidding more risky by putting capacitors near the cutting area and then by thinning the substrate. Both of those changes were just coincidental, I'm sure.

As long as delidding can offer more sales of Intel chips it is a significant incentive for Intel to keep overclocking alive.

*It also offers people an additional layer of the most addictive of the reward schedules by amplifying the silicon lottery gamble. Instead of just the silicon lottery, one gets the delidding lottery (will my chip survive it?, how much will the improved transfer increase the overclock?). It gives people a feeling of success (and bragging rights), not just an improvement in thermal transfer. So it also serves as additional marketing.

 

[cytg111]

Is there some innovation coming down the road to blast past the 5GHz wall that I am missing? I thought it was basically a physics limit.

Yea and it is surprisingly simple, the idea is to perform the calculations in a quantum gravity well where time moves much faster and thus performance. Problem is getting the results out. I am sure they will figure something out, they always do.

 

[Ajay]

No.

^n!

 

[BonzaiDuck]

Where did I read about a 7nm limit? How thick is an atom?

Many of us gathered in these forums have been immersed in our IT obsession for years, and we develop habits of comfort. So I think we keep anticipating a path for the future which is defined by those habits and may not really be in the cards.

Also, I haven't come across a single person telling me they killed their Sandy Bridge with an overclock, provided they avoided pushing the VID or VCORE well above 1.4V. I had seen a CPU deteriorate or thought I did (it might have been the motherboard), but it had a fixed OC with EIST disabled.

My oldest SB-K has been running (mostly) non-stop 24/7 for the past five years. I've not had to make any adjustments for "aging" in the settings. Just takes a licking, keeps on ticking.

 

[Phynaz]

Electromigration becomes an increasing issue as process nodes shrink, eh?

The question posed in this topic is valid enough, though, since AMD's FX, even overclocked to 5 GHz, cannot challenge recent Intel designs in single thread performance.

Intel really doesn't have to offer overclocking. However, one big thing is the delidding thing, which Intel promotes by using inefficient TIM. (Note: The Devil's Canyon TIM was sold to the enthusiast community with the promise of improvement but analyses found nothing significant.) Delidding, which can quickly destroy a chip — prompting an additional chip sale, is a beautiful thing to an Intel executive. Why wouldn't it be?* The "two sale" strategy has been rather big lately, like selling the Anniversary Pentium to people who will realize it doesn't really cut it in some games because of just having two threads. Intel also made delidding more risky by putting capacitors near the cutting area and then by thinning the substrate. Both of those changes were just coincidental, I'm sure.

As long as delidding can offer more sales of Intel chips it is a significant incentive for Intel to keep overclocking alive.

*It also offers people an additional layer of the most addictive of the reward schedules by amplifying the silicon lottery gamble. Instead of just the silicon lottery, one gets the delidding lottery (will my chip survive it?, how much will the improved transfer increase the overclock?). It gives people a feeling of success (and bragging rights), not just an improvement in thermal transfer. So it also serves as additional marketing.

Intel sells a million CPU's a day. Do you think they give two farts about the 10 people that delid?

 

[StrangerGuy]

You are aware that 6700K stock is way faster than 2600K OC?
And...
You are mixing enthusiast/workstation six core i7s with mainstream quad core ones.
Product line with 2600K goes like this 875K>2600K(2700K)>3770K>4770K(4790K)>5775C>6700K

Shhh...Stop making sense with sound logic.

And frankly I could care less about OCing when the top-end i7-Ks are already 4+GHz out of the box, and the landscape now is completely different from the times when the bargain basement and the uber expensive CPUs are different only by clocks and cache capacity.