Intel Haswell i7-4770K IPC and Overclocking Review @ [H]

has anyone had an issue with ssds on the new asus mobos? a friend of mine just went to a 4770 today, and he may have just lost his 2 ssds. the first one kept rebooting at windows boot and looping. and then he lost all trace of it in the bios.

once he booted into windows on a mechanical drive, his other ssd only lasted about 2hrs before that one dropped out of windows. i know he gave up on the system after that point, so we dont know if that drive is gone yet too. but it seems odd that he would lose 2 ssds in one day on a new system (just cpu, mobo, and ram), when hes had these things running problem free in his old system for the last 18+ months.

Do those SSDs work fine in another (older) system? Aren't the new boards much pickier about strict SATA protocol adherence? Those old Sandforce drives might not work out so well.
 
has anyone had an issue with ssds on the new asus mobos? a friend of mine just went to a 4770 today, and he may have just lost his 2 ssds. the first one kept rebooting at windows boot and looping. and then he lost all trace of it in the bios.

once he booted into windows on a mechanical drive, his other ssd only lasted about 2hrs before that one dropped out of windows. i know he gave up on the system after that point, so we dont know if that drive is gone yet too. but it seems odd that he would lose 2 ssds in one day on a new system (just cpu, mobo, and ram), when hes had these things running problem free in his old system for the last 18+ months.
What controller do those SSDs use? If it's SandForce 1xxx, there's your problem.
I've run headlong into this problem with my 34nm Vertex 2 and MSI Z87-GD65, and Kyle/Dan encountered this in the ASUS Z87-A review as well.

Looks to be a general Z87 issue that everyone has to patch.

That said, MSI did issue a new BIOS in late May (retail is late April). Haven't tried the Vertex 2 with this just yet. Newer SandForce controllers (using a SanDisk Extreme) work fine with both.
 
That is just unbelievable. The SF1200 line of SSD drives were very common several years ago and it's just hard to swallow that at no point during the validation of Z87, were these drives tested. How exactly does that happen? At least the SF2200 line of drives seem to be okay - all of the Asus videos i've seen were using the 2nd generation sandforce with seemingly no issues.

In any case, the fact that this problem exists across all Z87 motherboards is inexcusable.
 
That is just unbelievable. The SF1200 line of SSD drives were very common several years ago and it's just hard to swallow that at no point during the validation of Z87, were these drives tested. How exactly does that happen? At least the SF2200 line of drives seem to be okay - all of the Asus videos i've seen were using the 2nd generation sandforce with seemingly no issues.

In any case, the fact that this problem exists across all Z87 motherboards is inexcusable.

The root of the problem is Sandforce, not Z87. Also notice that Haswell can be pickier about PSUs. These are non-zero risks when self building your own systems.
 
Oh give me a break. Are you going to defend Z87 at all costs with this ridiculous hyperbole?

Didn't SF have to do some non-standard stuff to their chipset to get the speeds they had early in the game? IMHO, it's lucky that the MB makers/Intel were even willing to patch in kludges to get SF's non standard setup to work in the first place.
 
If you're going to throw ridiculous claims out there just cut the nonsense and present proof. What you're stating is not true W.R.T. SF1200 drives. I will also add, SF2200 drives work fine on Z87. Let me say that again: SF2200 works fine on Z87. SF1200 does not.

Take chill pill. There's no reason to get so worked up over it.
 
IF you're going to exaggerate present some proof. And not circumstantial evidence. Hard evidence. Let me state that again: HARD EVIDENCE that SF1200 is directly responsible for the issues when in fact, SF1200 was COMPLETELY FINE ON EVERY PRIOR AND SUBSEQUENT CHIPSET produced by intel. Why does SF1200 work fine on P67? x79? x58? Z77? Oh wait, according to you the issue is NOT with Z87 but with SF1200. Okay pal, if you say so.....since SF1200 works fine on every other chipset known to man, but not Z87, the isssue is CLEARLY with the SSD. :rolleyes: Stop defending Z87 at all costs, are you an intel shareholder or something?

Doing a bit of digging we found that some 1200 and 1600 series SandForce controllers are not fully SATA specification compliant and the previous Intel chipsets allowed for this out of spec behavior. It seems that the Z87 will not allow for this as it requires 'strict adherence to the SATA specification.' We did test the workaround BIOS that ASUS sent us without issues using the older Corsair F120 SSDs that employ the Sandforce 1200 controller. So obviously, this will have to be solved at a motherboard BIOS level and something to keep in mind if you have older SATA 2 SSD drives.
From the Take chill pill. There's no reason to get so worked up over it.[/QUOTE]

Yeah, it's got rather heated in here in a few posts :p
 
Well, at least this will let me upgrade my GPU this round. 2600K still going strong at 4.6 GHz. :)

I was looking forward to this CPU, but I didn't really NEED an upgrade...
 
Games can be HEAVILY multithreaded.
---
Basically there is a lot of parallelizable workload in games, and only lack of talent and incompetency of current game developers prevented to use efficiently 6+ cores.

Actually there is an image of recent game that work well on 6-cores. http://chipreviews.com/press/wp-content/uploads/2013/02/headsuphtoff.jpg

Maybe it's me - and I only play fast FPS games - but I was under the impression that increasing threads inevitably leads to more latency. BF3 obviously scales nice but that's a rather snailish game.
 
What controller do those SSDs use? If it's SandForce 1xxx, there's your problem.
I've run headlong into this problem with my 34nm Vertex 2 and MSI Z87-GD65, and Kyle/Dan encountered this in the ASUS Z87-A review as well.

Looks to be a general Z87 issue that everyone has to patch.


thats exactly what the problem was. hes running 2 vertex 2s.
 
Wow! If you had told me 2.1 years ago that the 2500K that I bought for $199 would still be in the top 5 CPUs for OC enthusiasts in mid 2013 I would have said: "You're crazy!"

At 1080P there are so few games that necessitate more than a 4.5ghz 2500k and a GTX 670. I guess I'll buy a new game console or smart phone, the PC market is stagnant.

The worst part is that I play Civ 5 so buying a second video card is a total waste of time. Borderlands 2, CS, and BF4 will not necessitate a second GPU @ 1080p.
 
Oh give me a break. Are you going to defend Z87 at all costs with this ridiculous hyperbole?

SF1200 drives have been on the market forever and I actually still have a SF1200 drive in my HTPC that has been humming along since early 2010. Yes, 3 years. I know of others, as well, who are still using the first generation sandforce drives with no issues. So to say that the issues are with sandforce without presenting evidence of such is laughable to say the least. Additionally, do not confuse the first generation SF1200 drives (which are having issues) with the second gen SF2200 drives. The SF1200 drives were wildly popular circa 2009 and had no major reliability issues.

IF you're going to exaggerate present some proof. And not circumstantial evidence. Hard evidence. Let me state that again: HARD EVIDENCE that SF1200 is directly responsible for the issues when in fact, SF1200 was COMPLETELY FINE ON EVERY PRIOR AND SUBSEQUENT CHIPSET produced by intel. Why does SF1200 work fine on P67? x79? x58? Z77? Oh wait, according to you the issue is NOT with Z87 but with SF1200. Okay pal, if you say so.....since SF1200 works fine on every other chipset known to man, but not Z87, the isssue is CLEARLY with the SSD. :rolleyes: Stop defending Z87 at all costs, are you an intel shareholder or something?

The SF1200 was a hokey controller with dramatically variable performance and higher failure rates than today's models. So that kind of throws a wrench into your argument, though it's technically correct that Z87 is probably at fault here.
 
Take chill pill. There's no reason to get so worked up over it.

Yeah, you're right of course. Over-reaction, my mistake.

I do still firmly believe that attributing an issue with chipset validation to the SSD itself is a major flaw in thinking - the SF1200 does work with every chipset out there except the Z87. And I say this as someone who is an overall fan of the Z87 platform and planning to buy a haswell system based on the Asus Z87 Deluxe in the near future - luckily, I am not using SF1200 in anything *but* my HTPC so any desktop I assemble will be free of the issue. However, In my opinion - there was some type of lapse in validation somewhere along the line, because SF1200 should work; therefore the lapse in validation is an inexcusable error on intel's part IMHO. I just don't get why some others are deflecting any possible blame on intel's behalf for this, these drives should work. Consider being in the shoes of someone who just purchased a new Haswell upgrade and threw a SF1200 in for a OS drive - I would be absolutely furious w.r.t. the data loss. I mean, how would you feel in that situation? Like I said though,I definitely did over-react earlier. Apologies if I offended anyone.
 
Last edited:
I'm constantly surprised at how many people don't understand the basic physics here. While the core itself may be hotter due to the die size and the method used to bond it to the heat spreader the chip still puts out less heat, you cannot generate more total heat with lower wattage.

The thing is, the power consumption has gone up compared to Ivy. Compare 4770k to 3770k - note that these are clock for clock.

55330.png


As you have pointed out though, it is also due to the method to bond (versus the soldered method on Sandy).


Yeah, you're right of course. Over-reaction, my mistake.

I do still firmly believe that attributing an issue with chipset validation to the SSD itself is a major flaw in thinking - the SF1200 does work with every chipset out there except the Z87. And I say this as someone who is an overall fan of the Z87 platform and planning to buy a haswell system based on the Asus Z87 Deluxe in the near future - luckily, I am not using SF1200 in anything *but* my HTPC so any desktop I assemble will be free of the issue. However, In my opinion - there was some type of lapse in validation somewhere along the line, because SF1200 should work; therefore the lapse in validation is an inexcusable error on intel's part IMHO. I just don't get why some others are deflecting any possible blame on intel's behalf for this, these drives should work. Consider being in the shoes of someone who just purchased a new Haswell upgrade and threw a SF1200 in for a OS drive - I would be absolutely furious w.r.t. the data loss. I mean, how would you feel in that situation? Like I said though,I definitely did over-react earlier. Apologies if I offended anyone.


You may be right. If you want to be certain, wait a couple of months before buying. It's possible that something like the bug on Sandy Bridge could reveal itself. On an unrelated note, this chipset did after ship with the USB 3.0 bug. Personally, if you own a Sandy Bridge system, performance wise, you're not missing out on much. Let's face it - if the Asus numbers are right, then only 10% of these chips are going to 4.8 GHz on air, and perhaps as many as a third struggle to reach 4.5 GHz.

Another reason to wait may be because there may be a better stepping down the line, kind of like the q6600 - G0 compared to B3. Unless you are totally getting the upgrade itch (and I have to admit, even I with my 2600k somewhat am because the new motherboards are awesome), there's every reason to wait. It's become kind of like when the new generation of GPU comes. I always wait, because companies like MSI will have something better in a few months (ex: the Lightning series) - that and by then the driver issues are usually gone, plus the new generation should by that point have gotten full SLI/Crossfire support (which tends to be often buggy when a new generation of GPUs is out).
 
Last edited:
Of course the power consumption went up compared to IVB. The chip now has an integrated VRM and the iGPU has more die size dedicated to it - so it's not really surprising that the TDP is actually slightly higher than IVB. Indeed if you look at the specs you will see that the 4770k does have a slightly higher TDP - it is 84W TDP compared to IVB 77W TDP. That amount is trivial and not something even worth considering, IMHO.
 
Yeah the consensus is the FIVR has something to do with the raised power consumption and temperatures. It isn't the TDP that really matters - it's the real world power consumption under load, along with the temperature of the highest core that makes the big difference. Compared to Sandy, this chip runs HOT, even if Sandy does use less power.

Causes:
1. FIVR
2. Possibly the new PCI/DMI bus
3. Intel's crappy paste versus the soldered Sandy caps
4. The fact that more power is being put through a smaller area than in a 32nm process

3. May solvable through delidding, but the others ... there's no easy solution.
 
I'm constantly surprised at how many people don't understand the basic physics here. While the core itself may be hotter due to the die size and the method used to bond it to the heat spreader the chip still puts out less heat, you cannot generate more total heat with lower wattage.

I was talking about Overclocking the chip. I guess I wasn't clear.
 
Whatever SKU I end up getting, I'll be delidding it for sure. I don't see why I shouldn't as a lot of people have had good success with Ivy. My 3570K will make a good test subject. :D
 
Whatever SKU I end up getting, I'll be delidding it for sure. I don't see why I shouldn't as a lot of people have had good success with Ivy. My 3570K will make a good test subject. :D

But a lot of people also killed/damaged their chips in the process. Only delid, if you have the funds to buy another one in case of failure.
 
The thing is, the power consumption has gone up compared to Ivy. Compare 4770k to 3770k - note that these are clock for clock.

55330.png



.

You failed to read properly:

Here I’m showing an 11.8% increase in power consumption, and in this particular test the Core i7-4770K is 13% faster than the i7-3770K. Power consumption goes up, but so does performance per watt.
You could easily gimp the frequency of the chip to perform like an older CPU while also reducing power draw. Getting work done faster and using less power while at it is what Haswell does.


and here's where the majority of the computer's time is spent:
55329.png
 
What thermal paste do you guys recommend for haswell/4770k?

What about for de-lidding?
 
What thermal paste do you guys recommend for haswell/4770k?

What about for de-lidding?

Depends, on what cooling solution you are going with and what frequency you are trying to achieve.
 
What thermal paste do you guys recommend for haswell/4770k?

What about for de-lidding?

Any tim will be fine (though I recommend pk-1). For delidding, collaboratory liquid ultra or pro should be your only options.
 
Any tim will be fine (though I recommend pk-1). For delidding, collaboratory liquid ultra or pro should be your only options.

Where is the best place to get Liquid pro in the US? All I can find is amazon for ~$37 shipped from Germany or Japan.
 
Well it looks like I'm not getting any further than 4.6, the amount of voltage I need to maintain this speed (1.285 + 1.65 mem) can put me into thermal throttling territory with prime95 loads.
 
Well it looks like I'm not getting any further than 4.6, the amount of voltage I need to maintain this speed (1.285 + 1.65 mem) can put me into thermal throttling territory with prime95 loads.

True, but you're leaving performance on the table in my opinion. I don't calculate prime numbers with my computer, so I'm willing to accept high P95 load voltage and temps. I haven't come close to those voltages/temps in games for instance.
 
I'm surprised 1.285V is in thermal throttling territory. Is that common on the new chips?
 
I'm surprised 1.285V is in thermal throttling territory. Is that common on the new chips?

The throttling only occurs in prime when doing 8k work loads, everything else lives around the high 60's low 70's. It's a completely unrealistic workload but a true stability test, AIDA on the other hand is a much better real world representation of a normal full loading and doesn't spike the CPU anywhere near as hard.
 
I'm surprised 1.285V is in thermal throttling territory. Is that common on the new chips?

On Prime95 and others that use AVX instructions, yes. Also in the case of P95, increasing DDR speed from 1600 to say 2133 will increase temps by 10C. With DDR at 1600 and a nice closed loop cooler, you might be able to hold 1.285V during P95.

The throttling only occurs in prime when doing 8k work loads, everything else lives around the high 60's low 70's. It's a completely unrealistic workload but a true stability test, AIDA on the other hand is a much better real world representation of a normal full loading and doesn't spike the CPU anywhere near as hard.

This.
 
On Prime95 and others that use AVX instructions, yes. Also in the case of P95, increasing DDR speed from 1600 to say 2133 will increase temps by 10C. With DDR at 1600 and a nice closed loop cooler, you might be able to hold 1.285V during P95.
.

And I'm running 2400mhz memory, bonus warmth :D
 
The throttling only occurs in prime when doing 8k work loads, everything else lives around the high 60's low 70's. It's a completely unrealistic workload but a true stability test, AIDA on the other hand is a much better real world representation of a normal full loading and doesn't spike the CPU anywhere near as hard.

Realistic maybe, but Aida64 ran for 12 hours on my chip at 4.4 Ghz and 1.295 volts with full stability and no throttling, and folding@home SMP client crashed within 8 hours.

Prime95 is still a good bet for full stability.

I have a sub-par 4770K. It needs 1.33 volts for 4.4Ghz stability and even with a de-lid/CLU application and use of a decent Antec 920 I'm getting full load temps near 87 Celsius on P95 with large workloads.

Chip is a joke. My de-lidded 3770K is running 4.9Ghz at 1.38 volts on a crappy Kuhler 620 at mid 70s with the same coollaboratories Ultra application.

Big step back for Intel IMHO.

I'm also thinking that they sent a bunch of golden samples to reviewers who are getting these easy 4.5+ Ghz overclocks.
 
My 4770k from SabrePC $310 shipped seems pretty decent, 1.2v @ 4.4ghz with 4 sticks of 30nm samsung ddr3 1600 @ 2133 using adaptive voltage so idles at 800mhz and .714v :p

Prime gets pretty warm high 70's and AIDA64 gets about 70c with H100 on low setting in a 550d
 
I wouldn't trust Aida64 for stability testing Haswell. Motherboard manufacturers are pimping it because it makes them look better than they are. I ran a 4.8GHz overclock that was "stable" for a couple of hours. Then I fired up Handbrake and the machine crashed instantly. Without Aida64 running I tried launching some other benchmarks and running them only to find out that my "stable" overclock was anything but stable.

Kyle has the same result here.
 
I wouldn't trust Aida64 for stability testing Haswell. Motherboard manufacturers are pimping it because it makes them look better than they are. I ran a 4.8GHz overclock that was "stable" for a couple of hours. Then I fired up Handbrake and the machine crashed instantly. Without Aida64 running I tried launching some other benchmarks and running them only to find out that my "stable" overclock was anything but stable.

Kyle has the same result here.

True and that was more or less my point, prime is still the best way to test for true platform stability however it's workloads aren't really anything close to real world.
 
Anywhere I can find a more comprehensive overclocking guide? I want to see how far my 4770K can go on normal voltage before backing back down to near stock levels on my NH U12S. (Yes, I know this is the [H] but my 4770K at stock is mighty overkill for the things I do at this point.)

The Asus guide was a good read but it seems like some forums have been suggesting tweaking more values than the Asus's "leave this at auto" standard recommendation.
 
Last edited:
Anywhere I can find a more comprehensive overclocking guide? I want to see how far my 4770K can go on normal voltage before backing back down to near stock levels on my NH U12S. (Yes, I know this is the [H] but my 4770K at stock is mighty overkill for the things I do at this point.)

The Asus guide was a good read but it seems like some forums have been suggesting tweaking more values than the Asus's "leave this at auto" standard recommendation.

http://www.overclockers.com/3step-guide-to-overclock-intel-haswell

as far, its all new and I would wait a few weeks until the OC community does its job.
many wait for the next costa rica batch due to Haswell dont oc well for them unless on ln2.

I tweaked 3 settings as far and I run 4.6ghz without issue.
 
Back
Top