New toy: i7-5775C

[mackerel]

Installing the i7-5775C turned into a manifestation of "fail to prepare, prepare to fail".

I decided to install it into my existing Haswell system. Had some Gigabyte Z97 mobo, 2x8GB Corsair Vengeance 2400, GT9600, and an optimistically rated "600W" PSU that weighs as much as an Intel stock cooler. Still, it was working fine for a long time as is.

I went straight in. Out came the i5-4570S, in went the i7-5775C. New thermal paste on CPU, pop the Hyper 212 back on and hit the power button. The system came to life, then turned off, and repeated this cycle until I pulled the plug. Ok, diagnosis time.

As a quick attempt I replaced the ram with older Vengeance 1600C8, which made no difference. Maybe it's the PSU? The only one I had outside of a system was the HX1000i. Overkill, but will do for testing. Didn't help.

Well, the CPU is the only thing that changed... off came the cooler. Took out CPU. Underside looked good. Socket pins looked good, but there was a single fleck of white stuff on edge inside the socket. I couldn't make out what it was even with more light, and had no magnification. Blowing didn't shift it. I got a pin and tried to move it. Whatever it was, it moved elsewhere in the socket, but at this point I was more concerned about damaging pins than removing it. It was kinda between pins so decided to leave it was safer option for now. I think it was either dust or maybe a solid lump of excess thermal paste.

Ok, put old CPU back in. Booted first time. Did I get a duff CPU? Or could it be the bios. Checked the version and sure enough, the mobo was on one version older than needed to support Broadwell. Doh! I had assumed given the board was so old, it was already up to date in bios versions. Well, I was wrong. Apply update, swap CPUs once more, and finally it was alive!

At this point, I should point out how much I hate the Hyper 212 when it comes to repeatedly fitting and removing it. To reach the 4 mounting screws you have to remove the fan. To remove the fan I had to remove the ram stick closest to the socket. One time, the mobo mount post came up with the screw so I had to reapply that afterwards. It was more than adequate as a cooler for running this CPU stock.

So I'm done? That's when the next set of problems began. On first boot I noticed the display looked weird, so the CPU swap seemed to have changed something and the video driver maybe needed to be updated. Windows tried to do that for me (it had the drivers from before) and screen went black. This is normal. It stayed black for a long time. This is not normal, then it rebooted. Leaving it, it just kept going through reboot cycles. Fine, I pulled the GPU and decided to run iGPU for now. That was ok, apart from Intel's driver update utility offering me a newer driver, that after install reported no driver was installed. It was working regardless, so I left it at this point, as I'll put in a different real GPU later.

That's where I'm leaving it for now. I've decided to refresh the system tonight. I'll put in my old 280X as the best cosmetic match for the case. I've ordered a Corsair CX550M which just arrived so I can recover the HX1000i for more serious uses. Also in process of cleaning the raid 0 SAS storage array, moving it instead onto a WD Green 1TB that was my SteamOS test install. I wouldn't mind the array so much if only it didn't take forever to boot and detect drives.

Then I might be able to start benching. I only did quick stress tests to make sure nothing obvious was wrong. Some aida64 stability as well as a bit of Prime95 small FFT. On the latter point, the CPU seemed to clock down to 3.0 GHz running that, which is below base clock. I need to check limiting settings at some point. In similar situation (but different mobo) the 5675C ran at normal turbo clocks.

 

[mackerel]

I've done more tinkering, and my conclusion is moving towards my mobo is rubbish. It's a Gigabyte Z97-D3H-CF. No matter what I try in bios, the power limit (PL1) doesn't go over CPU rated 65W. If I set it higher, it remains at 65W. I've tried values like 100 to 4000. If I reduce it to 45, it limits to 37W, which I now see if the CPU's optional TDP down mode.

I've tried turbo on and off, power saving features on and off, upping current limit, turning on/off "K OC" function whatever that does. Basically nothing in bios is allowing me to go above 65W. Clocks drop like a rock as soon as that power is reached.

Enter XTU. Installed it. Turned up power limit. Immediately performance increased. So it is not the CPU is itself limited in some way, just the mobo wont honour the power limit setting for whatever reason. Dunno what to do at this point. I could just fight on with XTU but that wont be a good solution for 24/7 running. I could swap it with the i5-5675C in an Asus board which I know OCs fine. Or I could look at getting another non-Gigabyte mobo, but due to age I don't rate my chances too good.

 

[mackerel]

Gonna call it a night on this one... using XTU and relaxing power limits, it runs all cores 3.7 GHz with P95 small FFT, which is essentially like MCE without the extra voltage. In that state the CPU reports 120W consumption, and temps get a bit scary with the Hyper 212. The CPU responds to XTU set power limits pretty closely. I can set 95W for example but then it bounces on and off the power limiter all the time. Interesting in XTU the slider all the way to the right is 4096W and it indicates "unlimited". The bios only lets me select up to 4090W but I've not seen the CPU in states other than 37W or 65W PL1 regardless of that setting. It is also interesting the per-core turbo clocks in XTU are different than shown in bios, and XTU also shows a +0.1v offset on cache which eats into the power limit a little. At 65W PL1, 0v cache offset is 3.0 GHz P95 small FFT, and 2.9 with +0.1v.

I think for serious play with OC I'm going to have to swap it to the existing i5 system. That also has a Noctua S14 in it which should help with the thermals also.