Haswell overclocking

[Piroko]

I believe the threshold for Intel to solder the IHS is somewhere in the 80-90w range. Anything over they solder, anything under they do not. Thus sandy bridge (95w) got soldered and ivy bridge (77w) did not. Perhaps someone else can verify this, I thought I read something about that awhile back.

I don't think there's a set TDP line beyond which they solder. Die size, elevated operating temperatures and maybe added complexity to the soldering process from Haswell GT3e might play a role.
There were rumours that Intel had an IB GT2e (so to speak) ready for production, perhaps that led them to thermal paste.

 

[Hulk]

This is kind of my point.

We can all see intel drip feeding the market. We dont need to deny it to make our selves feel better when we have to buy intel. Its clear to my eyes that intel isnt pushing performance like it was.

As i said this is most likely because of AMD's lack of performance and Arms low power. Which is why Haswell isnt any faster but uses less power... Coincidence is it Homeles?

Okay here's how I see it.
Intel has no competition over 10W and virtually none of the market under 10W. They are moving into the sub 10W arena because as Anand astutely pointed out this is where the market it moving and if they are to keep their fabs at capacity they need a significant presence in that market.

Now, keeping in mind that Intel has no competition at the top end they have devoted the majority of their resources into 1. Improving power efficiency, 2. Increase GPU performance for SoC purposes, and finally 3. Improving overall performance.

This makes sense from an economic point of view in my opinion. Why should Intel devote enormous resources to making their parts even more "faster" than the non-existent competition. Not only will it not sell any more parts, but it will allow ARM to continue to dominate the sub 10W market.

In addition, as has been discussed in many, many threads in this forum it is becoming increasing hard to increase IPC. There is only so much Instruction Level Parallelism to be exploited.

So it's not so much that Intel is dripping out performance but more that they are focusing their efforts where there will likely be the biggest payoff. Moving into the ARM space, SoC, better integrated GPU. Last on the list is better overall performance. For better or worse that's where us consumers have driven the market.

The fact that Hyperthreading reappeared a few years ago was a bit of a wake up call that Intel was having trouble increasing IPC. HT basically uses "unused" execution pipes in the already wide and getting wider every generation Core architecture.

TSX and AVX/AVX2 are what I believe nudges to the programmers that it's time to really start focusing on multi-threaded performance. At this point the best way to increase performance is to increase the core count and make the software better able to utilize all of the available cores. It would be nice for software manufacturers of compute intensive applications starting to advertise how close to linear (1:1) their software scales with increasing core count.

 

[tweakboy]

Or wait for Ivy E 6 core 15MB mem. Im grabbing the 4930k which will be 600 dollars.

 

[frozentundra123456]

As long as you tell them that you were overclocking the CPU when it died, and if they choose to replace it, I don't see what the issue is.

OK, we are getting into semantics and ethics, but if you return a defective CPU, do they actually ask you if you overclocked it.

It is a fine line, but if they dont ask, I see no reason to volunteer that you were overclocking. If they ask and you deny it, that is more questionable.

Tangentially, I actually had a sales person tell me that if I put a replacement GPU in an off the shelf system and it overloaded the power supply, to put the original card back in and return the computer. It never happened because I put in a HD7770 and the system is more than able to handle it, but still.....

 

[frozentundra123456]

I don't think there's a set TDP line beyond which they solder. Die size, elevated operating temperatures and maybe added complexity to the soldering process from Haswell GT3e might play a role.
There were rumours that Intel had an IB GT2e (so to speak) ready for production, perhaps that lead them to thermal paste.

What about ivy bridge E? A six core Ivy E would certainly have a higher TDP than 80 or 90 watts. Will they solder those?

 

[Fx1]

Intel plays the oldest trick in the book.

Unlocked FSB
Unlocked Multi
Solder TIM
Hyperthreading
Triple Channel Memory

These are just the things that i can remember which have been added and then taken away and then added back depending on what intel wants to give us. This is how they control the performance and the upgrade cycle.

If AMD was competitive you can bet we would be seeing 15% YOY gains in performance... i dont buy this "its getting harder" bull. We have faster ram and SSD's feeding our CPU's now and more cores that are faster than ever. Intel have just shifted their focus to low power instead of IPC performance.

Which means desktop Sandybridge owners wont have a reason to upgrade for a very long time.

I find it funny actually because intel is focusing on mobile but i think laptops are going to suffer way worse than desktops as a cheap tablet will replace a low end laptop but a cheap tablet wont replace a desktop they are too different. Not forgetting that PC gaming is growing significantly and Gamers are most likely to upgrade long before they really need to. They are also the most vocal and influential customers on the web.

Desktop users wont have a single reason to upgrade because performance isnt any better than Sandy. Laptop users are moving to ARM tablets which are cheaper. You can see why intel is betting the farm on the Ultrabook... Which wont be successful long term.. its too expensive BECAUSE of intel.

Surface Pro is a total bust. Windows 8 tablets wont take off. Ultrabooks are not ultra enough. They have poor GPU's but cost £1000+ which makes them very expensive Powerpoint machines.

Now you can see why PC sales are suffering 16% YOY declines. Intel and MS are helping the decine.

 

[bleucharm28]

loved my 920 D0, it was great OCker.

 

[AtenRa]

AMD fanboys. Point in case: Bulldozer.

Really ???

http://www.anandtech.com/bench/Product/203?vs=434

 

[AtenRa]

Why are you more important that 99% of the consumers out there?

Only selfish enthusiasts like yourself consider it to be a "screw up." Stop thinking about you. You are not more important than the vast majority.

Even if overclocking was worse, this goes back to the whole 99% thing. Why are you more important than the vast majority?

Your selfishness and ignorance are insufferable.

You are writing on an ENTHUSIAST forum in the CPU & OVERCLOCKING section, next time join another forum if you dont care about OC, the rest of us do.

And that makes YOU being the selfish and rude (middle spoken) around here.

PS: also, read the Forum rules.
http://forums.anandtech.com/showthread.php?t=2192303

 

[Maximilian]

You are writing on an ENTHUSIAST forum in the CPU & OVERCLOCKING section, next time join another forum if you dont care about OC, the rest of us do.

And that makes YOU being the selfish and rude (middle spoken) around here.

PS: also, read the Forum rules.
http://forums.anandtech.com/showthread.php?t=2192303

Yeah, i always think these people who try and take the average joe's point of view would be happier over at mumsnet or something :\

 

[inf64]

Really ???

http://www.anandtech.com/bench/Product/203?vs=434

It's amazing how facts (like above in AT bench) tell us one thing and forum "experts" claim something completely opposite . 8150 is just a better chip all around, the benchmarks don't lie. But hey, it's "faildozer" how other experts on the web in various "reviews" labeled it. It has "lower IPC" (a thing that varies wildly from one workload to another) therefore "it sucks".
Similarly, at one point I presented here on a forum a chart from hardware.fr which showed PD and BD headsup at same clock (everything else being equal). Another "forum expert" (wutaxl) disregarded it as "fail test" since he decided for himself that PD is no faster than BD and it's all the "same crap". A lie repeated 1000 times becomes the truth and here we have users who confirm this saying almost every day.

As for Haswell OCing, we are on a forum where most users don't run stock settings on a CPU. If anyone wants to discuss "Haswell without overclocking", an offtopic section is a great place for that .

 

[Borealis7]

if non-K CPUs get the 3 BCLK settings i'm seriously going to consider not getting a K chip.

i could get an "ordinary" 4770 for the HT+TSX and use a 125Mhz BLKC for a 25% OC, up the trubo (4 bins) and end up with a 4.8Ghz 4C/8T CPU which is more than enough for any game out there today including Crysis 3 (which i don't plan on playing...), plus most games are GPU bound these days anyway and will continue to be for the foreseeable future.

heck, i'd be happy if i could get 167Mhz BCLK stable on air without any multiplier change.

 

[Fx1]

if non-K CPUs get the 3 BCLK settings i'm seriously going to consider not getting a K chip.

i could get an "ordinary" 4770 for the HT+TSX and use a 125Mhz BLKC for a 25% OC, up the trubo (4 bins) and end up with a 4.8Ghz 4C/8T CPU which is more than enough for any game out there today including Crysis 3 (which i don't plan on playing...), plus most games are GPU bound these days anyway and will continue to be for the foreseeable future.

heck, i'd be happy if i could get 167Mhz BCLK stable on air without any multiplier change.

I have a feeling my 4770k is going to last me a very very very long time. Given Intels performance id expect it be no more than 15% slower than a CPU in 5 years time.

May as well get the K given how long ill be keeping it.

 

[Borealis7]

Given Intels performance id expect it be no more than 15% slower than a CPU in 5 years time.

quoted for checking back in 5 years

 

[Fx1]

quoted for checking back in 5 years

If you bought a 2700k then you would be easily ok for 5 years.

Broadwell wont be anything other than a haswell die shrink..

Are we betting on big performance or overclocking performance in 2015?

It seems 5ghz on Air is some kind of magical limit which i doubt 14nm is going to move us past.

 

[WhoBeDaPlaya]

You are writing on an ENTHUSIAST forum in the CPU & OVERCLOCKING section, next time join another forum if you dont care about OC, the rest of us do.

That's pretty much what I wrote :|

 

[Rezist]

Buying a K series processor hardly comes at a higher premium. And unless I'm mistaken, you can't hit as high of overclocks unless you have access to both the multiplier and the base clock. I.e., (I believe) you can push a black edition Phenom II harder than you can a non-black edition Phenom II, as you have more ability to tweak and tune. So there is still incentive for overclockers.

I'm not drawing any conclusions. But it would be nice to be able to overclock an i3 again.

Yeah but then you need a Z77 board as well. Your looking at usually a 230$ cpu (3570k) and the cheaper Z77 boards range from 110 to 130.

If a B75 (70$) board could overclock and you could use the cheapest quad 170-180$. You could save 70-100$ depending on the boards you go for.

Obviously the k and z77 boards will overclock further for sure. But then your gonna need a CPU cooler as well (30$) I'm sure every Haswell quad could hit 3.5Ghz w/stock cooler if they let us just overclock them.

 

[Idontcare]

OK, we are getting into semantics and ethics, but if you return a defective CPU, do they actually ask you if you overclocked it.

It is a fine line, but if they dont ask, I see no reason to volunteer that you were overclocking. If they ask and you deny it, that is more questionable.

Tangentially, I actually had a sales person tell me that if I put a replacement GPU in an off the shelf system and it overloaded the power supply, to put the original card back in and return the computer. It never happened because I put in a HD7770 and the system is more than able to handle it, but still.....

In terms of ethics, I would hope it is pretty clear.

If I burgle your house while you are away at work, and I never get caught (and I never confess to the theft), have I committed burglary?

Stealing from a company is the same as stealing from people because ultimately the employees have less money for bonuses and salaries, and the shareholders have stock that is worth less (however incrementally less is not the point) than it would have otherwise been had fraud not been committed in the first place.

All the semantics in the world don't change the crime, just changes your conscience.

"I wants what I wants and I wants no one to tell me I'm a bad person for doing whatever it takes to get it." (said pretty much every dishonest person to walk the planet)

 

[fixbsod]

Just to hazard a guess but I want to say that a good chunk of the reason for the perceived perf deficit is that when BD launched it did underperform due to some issue w/W7 scheduler (?) I'm assuming those numbers are post patch which did bring up the avg BD results like 5-10%?

Really ???

http://www.anandtech.com/bench/Product/203?vs=434

 

[AtenRa]

Just to hazard a guess but I want to say that a good chunk of the reason for the perceived perf deficit is that when BD launched it did underperform due to some issue w/W7 scheduler (?) I'm assuming those numbers are post patch which did bring up the avg BD results like 5-10%?

Original FX8150 review here in AT, BD is faster than Phenom II 1100 even without the new scheduler.

 

[Hulk]

If AMD was competitive you can bet we would be seeing 15% YOY gains in performance... i dont buy this "its getting harder" bull. We have faster ram and SSD's feeding our CPU's now and more cores that are faster than ever.

I'm not a microprocessor expert by any stretch of the imagination but I have been following architectural developments for the last 20 years pretty closely. The Core architecture seems to be highly optimized. I really, really hope you are right and Intel could make significant strides in IPC if they wanted to.

 

[lagokc]

I'm not a microprocessor expert by any stretch of the imagination but I have been following architectural developments for the last 20 years pretty closely. The Core architecture seems to be highly optimized. I really, really hope you are right and Intel could make significant strides in IPC if they wanted to.

Even if Intel couldn't improve IPC much, they have had a huge amount of clockspeed headroom for the last few generations. The only reason they don't have CPUs that boost well over 4GHz is because they want to leave more room for the next generation to be a tiny bit faster.

 

[Fx1]

In terms of ethics, I would hope it is pretty clear.

If I burgle your house while you are away at work, and I never get caught (and I never confess to the theft), have I committed burglary?

Stealing from a company is the same as stealing from people because ultimately the employees have less money for bonuses and salaries, and the shareholders have stock that is worth less (however incrementally less is not the point) than it would have otherwise been had fraud not been committed in the first place.

All the semantics in the world don't change the crime, just changes your conscience.

"I wants what I wants and I wants no one to tell me I'm a bad person for doing whatever it takes to get it." (said pretty much every dishonest person to walk the planet)

Your not stealing from a company when you RMA 20$ worth of silicon when you pay $300 for the CPU.

This is why intel will replace the CPU. Because the value of that CPU is nearly nothing to them. Keeping you happy is worth more.

 

[Hulk]

Even if Intel couldn't improve IPC much, they have had a huge amount of clockspeed headroom for the last few generations. The only reason they don't have CPUs that boost well over 4GHz is because they want to leave more room for the next generation to be a tiny bit faster.

You think they have a huge amount of frequency headroom? Right now the fastest parts turbo up to 4GHz. With mainstream parts and cooling I think they might be able to set base clocks at 4GHz and turbo up to perhaps 4.6GHz or so. That's about 15%. And with mobile they are TDP constrained already.

As I said above I'm not an expert but how much do you mean by "huge?"

 

[Enigmoid]

Original FX8150 review here in AT, BD is faster than Phenom II 1100 even without the new scheduler.

perf/watt is very similar which is sad considering they went from 45 nm to 32 nm.