Traveling Retro Gamer - Brickless S4M w/ 7600K, 1070

[SilverJS]

Update : undervolted both CPU and GPU. GPU to 80% as per @Thehack 's recommendation. CPU was with following settings (Asus Z270i), and was stable on P95 for a good 20 minutes (quick and dirty, I know, but good enough for this test) :

Voltage : Adaptive Mode
Negative offset
Turbo voltage available : 1.152
Offset : 0.08

With those settings, Heaven seemed to run without issue. I let it run for 10-15 mins, and on it went. So, that's a clear improvement.

Then, tried the big kahuna : heaven + P95. Crash within seconds.

I know this is worst case scenario, and artificial, but still - the math just doesn't add up...

Really wish the hdplex 400w ac-dc was out. I think I'd just go out and buy one, then get the dc-atx as well (or maybe the KMPKT 360, seeing as I've already got his dc-atx) and call it a day. I really thought my current combination would be enough, but clearly not.

That, or I sell the 1060 and just get a 1050Ti... But I don't really want to do that. Would rather future-proof power section instead of pampering to current one...

 

[Thehack]

Update : undervolted both CPU and GPU. GPU to 80% as per @Thehack 's recommendation. CPU was with following settings (Asus Z270i), and was stable on P95 for a good 20 minutes (quick and dirty, I know, but good enough for this test) :

Voltage : Adaptive Mode
Negative offset
Turbo voltage available : 1.152
Offset : 0.08

With those settings, Heaven seemed to run without issue. I let it run for 10-15 mins, and on it went. So, that's a clear improvement.

Then, tried the big kahuna : heaven + P95. Crash within seconds.

I know this is worst case scenario, and artificial, but still - the math just doesn't add up...

Really wish the hdplex 400w ac-dc was out. I think I'd just go out and buy one, then get the dc-atx as well (or maybe the KMPKT 360, seeing as I've already got his dc-atx) and call it a day. I really thought my current combination would be enough, but clearly not.

That, or I sell the 1060 and just get a 1050Ti... But I don't really want to do that. Would rather future-proof power section instead of pampering to current one...

You don't lose as much performance as you'd think by power limiting, it runs much more efficient. A 1060 at 80% power limit still has 90% of its stock performance.

Can you try reduce the max multiplier, by 10%, and -100mV offset (.1V)?

We still haven't determined whether this is a KMPKT 160 issue, or the AC-DC, but at this point I may believe it is the AC-DC. The Mini 160 based on most user reports should be able to handle your components.

 

[SilverJS]

Can you try reduce the max multiplier, by 10%, and -100mV offset (.1V)?

On the CPU, I'm assuming you mean? Sure - I'll give it a shot. I can also undervolt the card more significantly. I guess my question would be though : to what purpose? Will that help us determine which of the two (AC-DC, or DC-DC) units is the weak link? Or just in a bid to make the system useable? Because I can just use my RM550x in the meantime if my current power solution is not satisfactory, that's not an issue.

I was really hoping to have the whole Mini built and complete by the 9th of September, as that's when my heavier travel starts, but I don't know anymore if that is feasible...I'm starting to think that I'll just return my Mouser stuff (Meanwell and connectors/power cords), and use that to buy a Dell 330W brick for now (with my KMPKT DC-DC), and call it a day. I might revisit in a year or two to see the status of the latest AC-DC components. I wanted brickless, but not THAT bad. =) And that 400W AC-DC won't be ready in time, methinks.

 

[Thehack]

On the CPU, I'm assuming you mean? Sure - I'll give it a shot. I can also undervolt the card more significantly. I guess my question would be though : to what purpose? Will that help us determine which of the two (AC-DC, or DC-DC) units is the weak link? Or just in a bid to make the system useable? Because I can just use my RM550x in the meantime if my current power solution is not satisfactory, that's not an issue.

I was really hoping to have the whole Mini built and complete by the 9th of September, as that's when my heavier travel starts, but I don't know anymore if that is feasible...I'm starting to think that I'll just return my Mouser stuff (Meanwell and connectors/power cords), and use that to buy a Dell 330W brick for now (with my KMPKT DC-DC), and call it a day. I might revisit in a year or two to see the status of the latest AC-DC components. I wanted brickless, but not THAT bad. =) And that 400W AC-DC won't be ready in time, methinks.

Just to see it running. I'm interested in why it is failing as well, as I intend to make a kit using the UHP 200. The 7350k should be only using 65W and the 1060 120W. UHP is 200W with 300W peak.

 

[aquelito]

Have you tried cooling down your PSU with a fan during your tests ?

 

[SilverJS]

Just to see it running. I'm interested in why it is failing as well, as I intend to make a kit using the UHP 200. The 7350k should be only using 65W and the 1060 120W. UHP is 200W with 300W peak.

That's what I was thinking too...! Although, I wasn't tracking those bigger spikes on the 1060 until KMPKT brought them up.

Have you tried cooling down your PSU with a fan during your tests ?

I haven't, but that's easily remedied! =) That said - I'm not entirely sure that would have much of an effect, as the MeanWell is quite cool to the touch if I put my hand to it right after the crashes. The DC-DC unit, however, is definitely not - and I guess that's why I myself, for what it's worth, am more inclined to suspect the DC-DC unit, but I've got two people who know WAY more than me (not hard to do!) speculating otherwise. =) I need additional parts to properly diagnose - we'll see!

 

[Choidebu]

@Kmpkt , I think lending OP a dynamo 360 could test if the dynamo mini is at fault here, since it means gpu power is handled by the 360 not the mini. If it still crashes then he can return the meanwell.

@SilverJS , with gpu power limiting do you still see crashes in real world scenario, i.e. running games?

 

[SilverJS]

@Kmpkt , I think lending OP a dynamo 360 could test if the dynamo mini is at fault here, since it means gpu power is handled by the 360 not the mini. If it still crashes then he can return the meanwell.

@SilverJS , with gpu power limiting do you still see crashes in real world scenario, i.e. running games?

Actually, that's exactly the scenario we've worked out. I'll just buy the 360 regardless - if it turns out that the dc-dc wasn't the issue, I'll just have extra power potential down the line if I want it.

I actually haven't had time to test since I last posted. I might this afternoon - I had intended on seeing how the rig performed in real world scenarios.

 

[SilverJS]

Quick update - played about 20 minutes of Diablo 3 just now. Didn't further undervolt yet (as per @Thehack 's request - maybe later this afternoon). So, this was just at stock clocks (4.2GHz) but with 0.08v undervolt, and GPU at 80% power limit. Exactly the settings I had during my last benchmark run, I hadn't touched a thing since then.

Just wanted to see a bit of real-world use - and, from the little I saw anyway, no issues at all. CPU and GPU utilization both had reached a max of 100% utilization at some point during the session as reported by HWMonitor (after the fact - just looking at the historical peak) and GPU memory had actually reached 85% (Buffer 49%). Memory seems to typically be in the low 30's or so when running my Heaven sessions (of course, I only know this after having run them using my RM550x). But - no issues, no crashes, nothing : just solid, stable gameplay, during one of the more demanding GPU scenarios (4-player with two DH's, for those who know).

I will say this though : the DC-DC heatsink was quite warm to the touch - I only put my finger on the top portion of the heatsink, and couldn't hold my finger there for more than a second or two.

I'll play a bit more, and also play SC2 - that game seemed to be less forgiving of unstable BIOS settings and such, based on my experience. But still - even if it all works out, I want the peace of mind to know that the rig can take the worst-case scenario, just so I can go and play the games I want later down the line. My honest gut feeling is that the addition of the 360 will solve my problems, but we'll see. =) @Kmpkt should be sending me one shortly.

 

[SilverJS]

Another quick update : sightly longer D3 session today - crashed. MeanWell only sightly warm, dc-atx inhumanly warm.

Already set up 360 purchase with @Kmpkt, who will also loan me a 330w Dell brick, so that way we totally troubleshoot. Stellar individual right there, ladies and gents.

 

[sebplane]

Could you double check the cable you are using to power the cpu? Black and yellow 4 to 4 pin or completely black 4+4 pin?
I was having a similar issue and was using the wrong cable... See my post:
https://smallformfactor.net/forum/t...roubleshooting-thread.6897/page-2#post-106291

 

[SilverJS]

Thanks for the suggestion! But, I'm using the completely black 4+4. In fact, that's one of the things I had to make sure to inform Craig of : to make sure to include a 4 to 4 yellow/black when he ships me the 360. According to his write-up, one should have been included with the 160, but I never received one, so I don't even have one - precluding the possibility of that error, I suppose!

 

[SilverJS]

So I received a package from KMPKT today. Was so excited I went home at lunch to try quick. =) But first : a big, BIG thumbs-up to Craig for supporting me through this. Customer support I got through him was fantastic across the board.

I just decided to test my theory about the 160 being the bottleneck. So I chose to go for broke and use the 360+160 combo paired with the Meanwell, with the 1060 at 100% power cap (in other words, no cap) and my processor at my preferred overclocked settings of 4.8@1.31v...

...and...

...no issues!

Well - this is VERY preliminary, as I only ran one loop of Heaven. =)) It's all I had time for - I have to run back to work now. =) I know this was an exceedingly short run, but I can say with certainty that :

1. I'd never gotten this far on the Meanwell+160 alone;
2. Temps for all units were much saner. The Meanwell was barely warm to the touch; same for the 360; and the 160, while definitely warm, was nothing like it was before.

More extended testing is required, of course - and, just to make very sure I use all that that 360 can give me, I might just make it power the SATA too (only one SSD, but hey), just to relieve the 160 as much as I can.

But so far, that's what I've got. More testing to come shortly. Oh - and, after having seen that Dell brick in the flesh - man, that thing is BIG! And heavy! =) I used to think that going brickless wasn't such a big deal after all, until I saw that beast...Looks like I MIGHT be able to go brickless after all! =)

Cheers!

 

[SilverJS]

So I ran the full torture test yesterday : full power 1060 with the overclocked 7350k. Ran P95 at the same time as I ran Heaven - did that for just short of 45 minutes. No issues. MeanWell was certainly warm by this point, and the 160 was again uncomfortably warm to the touch. 360 was room temp. So - all seems to work now. . Bit, I might try powering the CPU off the 360 (in addition to the gpu). The 360 seems to have a lot of headroom and I'm still concerned with how warm the 160 gets.

45 minutes isn't what I'd call sufficient to determine stability on an overclock necessarily, but I think it's enough to determine if the power solution is sufficient...I figure if the power solution couldn't handle it, it would have given up the ghost after 45 minutes. But, I'm open to pushback - thoughts?

 

[Thehack]

So I ran the full torture test yesterday : full power 1060 with the overclocked 7350k. Ran P95 at the same time as I ran Heaven - did that for just short of 45 minutes. No issues. MeanWell was certainly warm by this point, and the 160 was again uncomfortably warm to the touch. 360 was room temp. So - all seems to work now. . Bit, I might try powering the CPU off the 360 (in addition to the gpu). The 360 seems to have a lot of headroom and I'm still concerned with how warm the 160 gets.

45 minutes isn't what I'd call sufficient to determine stability on an overclock necessarily, but I think it's enough to determine if the power solution is sufficient...I figure if the power solution couldn't handle it, it would have given up the ghost after 45 minutes. But, I'm open to pushback - thoughts?

Are you running the cpu power off the Dynamo 160?

 

[SilverJS]

I am - as per KMKPT's recommendation in his 360 Combo setup thread. He does say that, although not the preferred way, one can use one of the 360's two PCIe outputs to power the EPS on the motherboard, and I think I might try that next.

 

[Thehack]

I am.

I would move the cpu 12V to the Dynamo 360 to reduce the load on the 160. It may be a combination of the load and oven effect of the S4 mini that is causing the heating issue. Better off using the beefier unit to do the power regulation.

It is a strange issue though since usually the 160 mini is known to be able to handle around 200W.

 

[SilverJS]

Yep, my thoughts exactly...plus, the fact that I'm still running this open-air (haven't actually installed anything in the Mini yet), so that any heat build-up will undoubtedly be made worse when all is inside the Mini...understand the S4M has a vaunted ventilation system, but still, it can't help but be worse when inside a case as compared to just open air.

 

[Choidebu]

My 160 comfortably handle up to its rated wattage (I assumed - it pulls up to 180W from the wall running heaven and prime95)

Uncomfortable to touch is not a good indication of thermal problem. 50 deg is uncomfortable yet most electronics are fine with it.

Why don't you finish the build and see if there's any issue cropping up? At this point it's just strawmanning - what good does swapping cpu eps input do anyway if no actual issue exists?

 

[SilverJS]

Fair point - but, my response would be : why not? If I have the option to do it, and temps on components were reduced as a result, I'm not entirely sure what I have to lose (other than compromising my ability to go with a huge graphics card later, but I hardly have that option right now with my current ac-dc solution).

But yes - next step now is to put it all together, which I'm in the process of doing right now. I'm certain of the components I'll be using - how exactly I wire them up is trivial at this point, and easily amended if need be.

Speaking of putting together - looks like there won't be that much dremeling needed to fit the C6 backend into the C8 square hole after all. I should find out shortly.