Haswell build - H87 vs Z87

[Bouncing Zabaglione Bros.]

BZB,
Do you know a way to cap adaptive volts on ASUS boards to something manageable when running AVX (IE, 1.2000V at most), as I feel my board's UEFI volts too high when it's not even OCd to 4GHz with 4-core multiplier. I get like 1.275 or maybe more with prime95 stresstest (which is far higher than any other app admittedly; I still don't like it going so high.)

I tried setting manual volts in UEFI, but ended up pegging the CPU at that voltage at all times even when not loaded at all, drawing power and belching extra heat needlessly.

Yep, that's pretty much why you need adaptive voltage and not the standard VID.

My motherboard is the Z87-Pro, so depending which motherboard you have, the instructions may be different, but you are looking for the same things, and a lot of the Asus UEFI are very similar in most respects.

1. Boot to the UEFI.
2. Switch to Advanced mode.
3. Go to AI Tweaker.
4. Scroll down to CPU Core Voltage and change it from "auto" to "adaptive". A new adaptive voltage field should appear.
5. Enter 1.200 into the new adaptive voltage field. Don't forget to hit enter after you've put the numbers in.
6. Save and exit.

If you use AI Suite, you should see the difference on the TPU page under the CPU Core Voltage pane. You can do testing in this same part of AI Suite to find the OC adaptive voltage that you like by changing this setting. However, manual changes like this are not then saved to the UEFI, and are lost between boots. You need to go into the UEFI and make the changes there to make them stick. Only when you run the 4-Way auto-tuning are the settings then saved to the UEFI. Everthing else you change manually in AI Suite is not saved between boots.

Rather than using Prime95 (for the reasons I outlined above), I've been using MediaCoder X64 to encode MP4s using their experimental segmented encoder. This is a real world application that is trying to load the CPU as much as possible for efficient encoding, it won't pull as many volts as a torture test, because it's actually doing other encoding calculations at the same time, not just trying to heat the CPU.

I've found that dropping the adaptive voltage to 1.200 makes the encoding run massively cooler, and even Prime95 behaves itself a lot better.

 

[smw]

Linpack is a testing-to-destruction stress test though. It's not really a proper application. Even Intel engineers cringe at how people use it (like running it for hours on end). They say it's like putting your car on a rolling road and then pegging the engine into the redline for hours on end - don't be surprised when something breaks.

It's one thing to be using it as a maths library, another to be using it to see if it breaks the CPU.

I was under the impression that the benchmark uses the same routines as the ones from the math library. So if you use the routines from there and run a simulation for a couple of days it would still be possible to overload your CPU. It's granted that not all of you time will be spent on solving the matrices itself, so the fans will have time to cool off the CPU, but this is why I remarked if it will show the same loading patterns for other optimized code (thus also heating up during the rest of the program).

Also a small note (not that small though ) to other haswell owners - you should test if all of the subsystems work okay. Today I ran the memory benches from Aida64 and while the Write speed and Latency were as expected, memory copy and read were way off (also the aggregated memory bandwidth in SiSoft Sandra2013). There was a UEFI update on Asrock's site from just 2 days ago, which didn't even mention memory performance in the notes, but sure enough after the update the speeds were ok. Actually, Sandra even reported (in the result page for the Cache bench), that supposedly there is also microcode update available for my CPU, but after the UEFI update this was still reported, so I don't know if this is true or not. Also for some reason after the update, the CPU fan speed is no longer reported in any software that I currently have available.

 

[Bouncing Zabaglione Bros.]

I was under the impression that the benchmark uses the same routines as the ones from the math library. So if you use the routines from there and run a simulation for a couple of days it would still be possible to overload your CPU. It's granted that not all of you time will be spent on solving the matrices itself, so the fans will have time to cool off the CPU, but this is why I remarked if it will show the same loading patterns for other optimized code (thus also heating up during the rest of the program).

I think it does use the same libs, but in real world usage you're unlikely to be doing nothing but loading every aspect of the chip all of the time. When you drive, you're not on the redline all the time even if you touch it now and again, and that's what these stress tests are doing.

Also a small note (not that small though ) to other haswell owners - you should test if all of the subsystems work okay. Today I ran the memory benches from Aida64 and while the Write speed and Latency were as expected, memory copy and read were way off (also the aggregated memory bandwidth in SiSoft Sandra2013). There was a UEFI update on Asrock's site from just 2 days ago, which didn't even mention memory performance in the notes, but sure enough after the update the speeds were ok. Actually, Sandra even reported (in the result page for the Cache bench), that supposedly there is also microcode update available for my CPU, but after the UEFI update this was still reported, so I don't know if this is true or not. Also for some reason after the update, the CPU fan speed is no longer reported in any software that I currently have available.

I'd recommend updating UEFI regularly on new boards like these Haswell ones. Manufacturers will be putting in fixes and changes as they get to know the hardware and how it handles in the real world. I'd expect all kinds of stability and compatibility fixes to be going in for the first few months.

 

[swaaye]

I have a cheap Asrock Z77 board that can update its BIOS while in the BIOS via a internet connection. We are undeniably living in the future. It didn't even brick itself.

 

[smw]

I have a cheap Asrock Z77 board that can update its BIOS while in the BIOS via a internet connection. We are undeniably living in the future.

Indeed! I've been discussing this selfsame thing just the other day with my android butler!

 

[homerdog]

Oh yeah? I just updated my BIOS while still in Windows.

It makes me proud to be an American. Even though this was all designed by Taiwanese people.

 

[swaaye]

BIOS flashing in Windows is a sure sign of courage. True test of manhood.

 

[Grall]

6. Save and exit.

Thanks. I'll go have a look-see. Although CPU temps have improved slightly after switching CPU heat sink to the NH-12S, it still hits upwards of ~92C, and I suspect it would do that regardless of what air cooler I use due to the TIM intel is using and these crazy adaptive volts. Maybe even these simple all-in-one water coolers wouldn't budge temps much...

Rather than using Prime95 (for the reasons I outlined above), I've been using MediaCoder X64 to encode MP4s

Oh, I know. I only use prime95 as a stress test, not as an all-in-one wonder problem finder (because like I've mentioned, P95 didn't see any issues with my RAM @2400, while other software bombs out at that speed.)

Spent most of the afternoon and all friggin evening tearing my PC down and re-building it again, due to having to switch mounting hardware for the new CPU cooler (no cut-out on the rear plate), and also replacing the case fans to higher flow/quieter units. These fans really are really quiet. With 5 case fans (and 4 more fans on the video boards) going all I hear is a really low hum at no-load situations. PSU fan doesn't even spin up. The loudest component when just browsing the web or such is now my 3.5" 7200RPM Hitachi HDD from around four years ago which I only use for bulk storage. I will prolly switch it for a WD green drive instead as performance won't matter. It will run cooler and quieter.

Experimenting with the Corsair Lighting Kit by the way. It is somewhat amusing, although the whole setup causes a lot of cable clutter. I'll have to do something about that somehow...

 

[homerdog]

BIOS flashing in Windows is a sure sign of courage. True test of manhood.

BTW after posting that I rebooted, and my Windows installation is hosed. I'm posting this from my laptop as my PC reinstalls Windows.

I tried everything to revive it, but alas, my BIOS had reverted to default settings, which in this case meant switching the SATA ports from AHCI to IDE mode, and my Win7 install became very, very, very angry.

I am no longer proud to be American. I have been humbled, and now many days of reinstalling drivers, games, and various programs await. Not to mention redirecting all my libraries to the data drives and reconfiguring my BIOS. Oh yes, I must pay for my sin of expecting things to work right, or at least without catastrophic failure.

Edit: Now I realize, I had sinned against the PC Gaming Gods by coveting the inferior hardware of the PS4. I have learned my lesson oh great ones.

 

[Grall]

I'm surprised that just switching SATA ports back into AHCI didn't fix the problem (possibly along with rolling back to previous restore point if necessary). I know I have rebooted with drives back in IDE mode by mistake after flashing BIOS on my Win7 system with no ill effects whatsoever, the OS re-detected my storage devices/controllers, and asked me to reboot again. Doing that, then rebooting again to flip back to AHCI (requiring another re-detect and reboot) and everything would be back to normal again.

 

[Bouncing Zabaglione Bros.]

I tried everything to revive it, but alas, my BIOS had reverted to default settings, which in this case meant switching the SATA ports from AHCI to IDE mode, and my Win7 install became very, very, very angry.

Why didn't you just go into the BIOS and switch the ports back to AHCI? I've done that to put a RAID disk into a new machine. I just had to reconfigure the BIOS for RAID, and it booted right up.

 

[Bouncing Zabaglione Bros.]

Thanks. I'll go have a look-see. Although CPU temps have improved slightly after switching CPU heat sink to the NH-12S, it still hits upwards of ~92C, and I suspect it would do that regardless of what air cooler I use due to the TIM intel is using and these crazy adaptive volts. Maybe even these simple all-in-one water coolers wouldn't budge temps much...

Those AIO water coolers don't seem to have particularly better performance than a high end air cooler, and they cost more. There's also a lot more to go wrong, and some horror stories around the net for what can happen when they fail and spill water into your case.

IIRC, all the AIO water coolers are made by Akasa, and just rebadged for various companies. Once you have the fan(s) on the radiator and the pump running, they are no quieter than a good aircooler with a big fan on it.

Personally, I don't think the AIO's are quite ready for prime time compared with the effectiveness, price and near zero maintenance of a quality air cooler. It's especially the case with it being so difficult getting heat out of Haswell and into the heatsink in the first place.

 

[homerdog]

My computer booted after the BIOS update, and I noticed that the first time it tried to load Windows, it failed, then rebooted again, then went back into Windows on the 2nd boot attempt just fine.

I can only assume that after the first failed attempt to boot after the BIOS update, Windows switched the registry key back to expect the SATA ports to be in IDE mode. When I clumsily went into the BIOS and switched to AHCI mode, I did not take that into account. The next time I tried to boot, it failed and went into startup repair, which failed.

I tried switching the SATA ports back to IDE mode at this point, but Windows still would not boot. I tried going back to a restore point, which not only failed, but informed me that the restore point had been destroyed in the process. LOL!

 

[swaaye]

I've found that once you have Windows working with AHCI you can switch between legacy IDE and AHCI whenever you want to. The AHCI driver just needs to be enabled manually or automatically during initial Windows install (the legacy IDE driver always is). So I'm not sure what was going on there. Maybe the UEFI update did something else that blew up Windows. Or that startup repair made a mess.

 

[Silent_Buddha]

Sweet, just noticed there are quite a few consumer level Z87 boards with 10x SATA. That's almost enough to max out my Silverstone Raven RV03 that I use for a home server (10x 3.5" internal bays + 2x 2.5" internal bays) without using an SATA PCIE card. Theoretically that should help to reduce power draw by a bit in addition to Haswell power savings over my existing Sandy Bridge Core i3 in there. The much improved video playback on Haswell should allow me to ditch the Radeon 5450 that I have as well leading to more power savings. Full ATX MB is likely to draw a bit more power than my current Mini-ITX board. But then I'll also be able use more memory more easily if I wanted.

Now I just need Intel to hurry up and release Haswell based Core i3's into the retail channel.

Regards,
SB

 

[homerdog]

I've found that once you have Windows working with AHCI you can switch between legacy IDE and AHCI whenever you want to. The AHCI driver just needs to be enabled manually or automatically during initial Windows install (the legacy IDE driver always is). So I'm not sure what was going on there. Maybe the UEFI update did something else that blew up Windows. Or that startup repair made a mess.

I do suspect that startup repair somehow made things much worse, but that's a hunch. I don't really know what caused such unrecoverable disaster. Throughout the whole I didn't lose any data, it's just been a major PITA.

I'm back up and running now, and with all the hundreds of GBs of games I'm downloading, I may be personally responsible for the Steam errors some of you guys have been experiencing

 

[joker454]

Just curious since this is a haswell build thread, has anyone experimented with faster ram and seen any major speed increases? I just tried 2133mhz ram with my 4770k cpu and it was barely faster than my previous 1600mhz ram on some video encoding tests. I think many haswell boards support up to 2800 speed ram, has anyone tried more extreme ram and seen actual benefits?

 

[Bouncing Zabaglione Bros.]

Just curious since this is a haswell build thread, has anyone experimented with faster ram and seen any major speed increases? I just tried 2133mhz ram with my 4770k cpu and it was barely faster than my previous 1600mhz ram on some video encoding tests. I think many haswell boards support up to 2800 speed ram, has anyone tried more extreme ram and seen actual benefits?

I'm running 2400 using the standard XMP 1.3 profile, and I'm not seeing massive differences in real world use. Other people have tried overclocking RAM and it doesn't seem to help much on the Z87/Haswell platforms.

I have seen claims that you can get better and more stable overclocks if you run your RAM at lower speed and disable the onboard GPU, but it doesn't seem to make any difference to me. Cranking up the CPU is what gives the significant gains.

 

[joker454]

I'm running 2400 using the standard XMP 1.3 profile, and I'm not seeing massive differences in real world use. Other people have tried overclocking RAM and it doesn't seem to help much on the Z87/Haswell platforms.

I have seen claims that you can get better and more stable overclocks if you run your RAM at lower speed and disable the onboard GPU, but it doesn't seem to make any difference to me. Cranking up the CPU is what gives the significant gains.

Bummer, it was worth trying at least. Guess this will do until Haswell-E rolls around.

 

[Grall]

Just curious since this is a haswell build thread, has anyone experimented with faster ram and seen any major speed increases?

From what I recall of an Anandtech investigative article, the most gains are seen from pressing down latency, not increasing bandwidth. This was with ivy bridge, but haswell is largely identical from an architectural standpoint, so the conclusions should be the same.

Of course, this was from a CPU standpoint, IGP useage may/will (probably) have other needs.