Ah! The Ebay stuff? He has 'Kareon' in his username, I guess that's what Josh was talking about.
-
Save 15% on ALL SFF Network merch, until Dec 31st! Use code SFF2024 at checkout. Click here!
Traveling Retro Gamer - Brickless S4M w/ 7600K, 1070
- Thread starter SilverJS
- Start date
So - GPU settings that, I think, should be fairly close to final :
Power limit : 75%
Core clock : +200
Memory clock : +400 (didn't try higher yet)
Benchmark comparison (Unigine Superposition):
Everything stock, right out of the box : 8868
With settings above : 8736
So, 98.5% performance for 75% of the power draw. Pretty thrilled!
I'll run longer benchmarks later, but so far, tested for 15-20 minutes each of Heaven, Superposition, and both the GPU and memory burn tests of Kombustor.
Power limit : 75%
Core clock : +200
Memory clock : +400 (didn't try higher yet)
Benchmark comparison (Unigine Superposition):
Everything stock, right out of the box : 8868
With settings above : 8736
So, 98.5% performance for 75% of the power draw. Pretty thrilled!
I'll run longer benchmarks later, but so far, tested for 15-20 minutes each of Heaven, Superposition, and both the GPU and memory burn tests of Kombustor.
In SFF, we don't tune for more FPS, we tune for lower power!
Legit, though +200 mhz is a solid undervolt.
Wow - all I can say is, I spoke WAY too soon. Who knew that D3 could be so demanding...? Either way, had to scale everything back A LOT. Right now, I'm about to try 65% power limit, as I just had another power cycle at 70%. I either had instant GPU instability at my previous settings (from post above) or, even after I turned everything back to stock save power limit, I was getting power cycling after like 5 minutes of play. I suspect it's the peaks the MeanWell can't handle...
At 65%, a quick benchmark gave me 7518 - a 1060 Amp does roughly 6600 or so, so I'm starting to get down there... (I mean - I think it's still worth it to keep the 1070, for the still higher performance if hobbled, higher gddr count, and the trouble of selling/buying, but still ....)
Amazing the difference between the "benchmarks" and "stress tests" and the real thing!
At 65%, a quick benchmark gave me 7518 - a 1060 Amp does roughly 6600 or so, so I'm starting to get down there... (I mean - I think it's still worth it to keep the 1070, for the still higher performance if hobbled, higher gddr count, and the trouble of selling/buying, but still ....)
Amazing the difference between the "benchmarks" and "stress tests" and the real thing!
Also certain games peg cpu differently so you can try reducing the power budget on the cpu more. It may also be your undervolt (core clock increase) or mem overclock that may be unstable so don't just count that as a power issue.
CPU is already undervolted - ask I could get was 0.04v. Better than nothing, I suppose. 
EDIT : I meant, all I could get to maintain stock clocks. I suppose I could always underclock as well as undervolt.
I had already turned off all GPU overclocking (core and memory) at that 70% power spike, so I think it can only be power. Plan now is : I'll find the power limit that works, THEN start applying overclocks to the GPU.
So far, 65% works fine.
EDIT : I meant, all I could get to maintain stock clocks. I suppose I could always underclock as well as undervolt.I had already turned off all GPU overclocking (core and memory) at that 70% power spike, so I think it can only be power. Plan now is : I'll find the power limit that works, THEN start applying overclocks to the GPU.
So far, 65% works fine.
Last edited:
Alright, so... Real world results now. Stable after 1.5 hours of D3.
Power Limit : 65%
Core clock : +175
Memory clock : +400
New benchmark : 8083. So, 91.1% of the performance of a stock 1070 for 65% of the stock power draw.
EDIT : I'm kinda tempted to just stick the 7350K back in there just to see what happens. I was running a pretty healthy undervolt on that (0.08, as I recall?). I wonder if the two additional physical cores are just sucking a fair bit of power under load... The T processor is clearly easy to cool, but I'm now more skeptical than ever with regards to their supposed low power consumption.
Power Limit : 65%
Core clock : +175
Memory clock : +400
New benchmark : 8083. So, 91.1% of the performance of a stock 1070 for 65% of the stock power draw.
EDIT : I'm kinda tempted to just stick the 7350K back in there just to see what happens. I was running a pretty healthy undervolt on that (0.08, as I recall?). I wonder if the two additional physical cores are just sucking a fair bit of power under load... The T processor is clearly easy to cool, but I'm now more skeptical than ever with regards to their supposed low power consumption.
Last edited:
Well. THIS was interesting.
Plopped the 7350K back in. No issues at all running with 80% power cap. So that 7600T drank a lot more power than I thought.
With 80%, had clock +140 and my now-usual mem +400. Anything more on clock would sometimes give me the occasional GPU restart in D3. Benched 8808, so essentially full stock 1070 performance.
So now I have a decision to make...
One last option is to plop in that 7600K I have sitting around for my sister's next build, just to run further tests. It's ready to go in its motherboard, with its cooler and RAM and all, waiting to be installed in her computer (she's running an Athlon II X3 445...ouch). Wonder if I could turn it into a T processor, but with an AVX offset, adaptive voltage mode available (instead of straight offset only for the T), etc...
Plopped the 7350K back in. No issues at all running with 80% power cap. So that 7600T drank a lot more power than I thought.
With 80%, had clock +140 and my now-usual mem +400. Anything more on clock would sometimes give me the occasional GPU restart in D3. Benched 8808, so essentially full stock 1070 performance.
So now I have a decision to make...
One last option is to plop in that 7600K I have sitting around for my sister's next build, just to run further tests. It's ready to go in its motherboard, with its cooler and RAM and all, waiting to be installed in her computer (she's running an Athlon II X3 445...ouch). Wonder if I could turn it into a T processor, but with an AVX offset, adaptive voltage mode available (instead of straight offset only for the T), etc...
So this is even more interesting. Tried my sister's 7600K. Just took the whole shebang I have for her (it, Asus Z270-E, 2x4gb Corsair lpx 2800), hooked it up to my Mini power supply solution, and went to town undervolting. The thing was an average clocker - 4.8 stable at 1.30v - but looks to be quite the undervolter. I was running it at stock clocks but with the same rough vcore as the T - 0.92ish under load.
First indication was that, in P95, I was pulling in 30-35W on HWMonitor. Recall that both previous processors were pulling 40-45W as indicated by the software. Again, I'm not using that as an absolute measure, but as a relative one. And here, it was clearly telling me that, despite clocking 0.5GHz higher than the T, it was drawing less power.
So, I was able to run the 1070 at its higher settings from earlier - 80% power, etc - but with a full-fledged 7600k in tow. And, my MeanWell was noticeably cooler to the touch than it had been in either of my two past configs.
I don't see how I lose here - I gain everywhere. Higher clocks, lower power draw. But, I won't draw definitive conclusions yet - been bitten by that before.. Tomorrow, I'll test it in the actual Z270i mobo, with my actual 3466 ram, etc. Can't see any of that making a difference, but... Stranger things have happened.
First indication was that, in P95, I was pulling in 30-35W on HWMonitor. Recall that both previous processors were pulling 40-45W as indicated by the software. Again, I'm not using that as an absolute measure, but as a relative one. And here, it was clearly telling me that, despite clocking 0.5GHz higher than the T, it was drawing less power.
So, I was able to run the 1070 at its higher settings from earlier - 80% power, etc - but with a full-fledged 7600k in tow. And, my MeanWell was noticeably cooler to the touch than it had been in either of my two past configs.
I don't see how I lose here - I gain everywhere. Higher clocks, lower power draw. But, I won't draw definitive conclusions yet - been bitten by that before.
. Tomorrow, I'll test it in the actual Z270i mobo, with my actual 3466 ram, etc. Can't see any of that making a difference, but... Stranger things have happened.
LOL! Haven't forgotten about her, worry not. =) In fact, already have a line on not one, but two 7600K's that are semi-local. (That's another advantage with the K-SKU's - they're fairly ubiquitous...compared to the T's, anyway.) Selling the T will (rather easily) fund the purchase of the second 7600K.
So tried the 7600K in the Z270I. Had the EXACT same BIOS settings - down to the RAM being downclocked and downvolted to match the 2800 RAM in my sister's future setup - and, to my surprise, the 7600K pulled ~50W in HWMonitor on P95 26.6. That was really strange...
...so I decided to just call the motherboard's bluff and try a bit of D3 just for kicks. GPU was at the 80% power limit settings (+140, +400). No issues. Tried a bit of Heaven + P95 26.6 - for 5 minutes or so, looked good (would usually power cycle within the first 5 minutes if it was going to, in my experience). MeanWell was again noticeably cooler than it usually was.
Not final, of course, will have to run more rigorous stability tests, but I'm inclined to think :
1. Z270I's wattage consumption reporting is a bit flawed (I really can't see the Z270E somehow shaving off 15W of consumption..unless the Load Line Calibration, which was at Level 3 in both cases, is run differently between the two mobos);
2. For whatever reason, the MeanWell seems to like the 7600K better. Can't say why.
So - anybody want to buy a 7600T? =) I'll give you guys a good price. I really think I'll go K - at the worst, I can downclock further if I eventually find I need it, but I'm quite likely to still be faster than the 3.5GHz all-core turbo of the T.
So tried the 7600K in the Z270I. Had the EXACT same BIOS settings - down to the RAM being downclocked and downvolted to match the 2800 RAM in my sister's future setup - and, to my surprise, the 7600K pulled ~50W in HWMonitor on P95 26.6. That was really strange...
...so I decided to just call the motherboard's bluff and try a bit of D3 just for kicks. GPU was at the 80% power limit settings (+140, +400). No issues. Tried a bit of Heaven + P95 26.6 - for 5 minutes or so, looked good (would usually power cycle within the first 5 minutes if it was going to, in my experience). MeanWell was again noticeably cooler than it usually was.
Not final, of course, will have to run more rigorous stability tests, but I'm inclined to think :
1. Z270I's wattage consumption reporting is a bit flawed (I really can't see the Z270E somehow shaving off 15W of consumption..unless the Load Line Calibration, which was at Level 3 in both cases, is run differently between the two mobos);
2. For whatever reason, the MeanWell seems to like the 7600K better. Can't say why.
So - anybody want to buy a 7600T? =) I'll give you guys a good price. I really think I'll go K - at the worst, I can downclock further if I eventually find I need it, but I'm quite likely to still be faster than the 3.5GHz all-core turbo of the T.
I'm really surprised. Did various benchmarks and stress tests (Kombustor + P95, etc), no issues. Just did a 1hr D3 session - all good. (Did have a single GPU restart at some point at +140 - turned it down to +135, all was well.)
This 7600k just seems more...graceful than the other two chips. It's got more grunt yet seems to consume less power. I'm glad I tried that out!
Have a local buyer for 7350k, and I'll grab another 7600k while I'm out. I'll keep the more efficient one of the two I'll have, and give my sister the other one. (Hey - she doesn't care, I'm giving her an RM550x also!! =))
But yeah - surprised I can run a 7600k+1070 on this 200w MeanWell. This is way beyond what I thought I could get. @royalba94 is just wrapping up with the cable, so I should be able to button this up soon.
Also have to decide if I'm delidding or not.... But I haven't seen a temp higher than 70 open air. I'm also curious to see what happens when everything is put in the chassis...hopefully I'll be able to keep the same numbers on the GPU. Temps were always 65-70 on the GPU, but I'm more worried about the MeanWell getting even warmer due to the GPU being right next to it, therefore affecting its max output.
This 7600k just seems more...graceful than the other two chips. It's got more grunt yet seems to consume less power. I'm glad I tried that out!
Have a local buyer for 7350k, and I'll grab another 7600k while I'm out. I'll keep the more efficient one of the two I'll have, and give my sister the other one. (Hey - she doesn't care, I'm giving her an RM550x also!! =))
But yeah - surprised I can run a 7600k+1070 on this 200w MeanWell. This is way beyond what I thought I could get. @royalba94 is just wrapping up with the cable, so I should be able to button this up soon.
Also have to decide if I'm delidding or not.... But I haven't seen a temp higher than 70 open air. I'm also curious to see what happens when everything is put in the chassis...hopefully I'll be able to keep the same numbers on the GPU. Temps were always 65-70 on the GPU, but I'm more worried about the MeanWell getting even warmer due to the GPU being right next to it, therefore affecting its max output.
Last edited:
Quick update : sold 7350K and 1060 without much issue, and picked up another 7600K. This one was at first glance more efficient than the first (could hold stock clocks at 0.080v offset, vice 0.055v on the first) but in what seems to be another sensor vagary, vcore was about the same in both cases. Not sure if one processor could somehow pull more current than another identical one when using different voltage settings in the BIOS, but there you go.
Either way, decided to see if this was due to a reading error and see if I could push a bit. Got the new 7600K to a mild overclock of 4.3 on all cores with a 0.065v offset. No power cycles in D3 or on my stress tests of P95+Kombustor. I think it'll just stay right in the motherboard where it sits, and my sister gets her original one.
Cables have made their way into Canada but our national postal service has begun a series of rotating strikes. My general area isn't affected yet, but we'll see...should be a fairly straightforward assembly once cables come in.
Either way, decided to see if this was due to a reading error and see if I could push a bit. Got the new 7600K to a mild overclock of 4.3 on all cores with a 0.065v offset. No power cycles in D3 or on my stress tests of P95+Kombustor. I think it'll just stay right in the motherboard where it sits, and my sister gets her original one.
Cables have made their way into Canada but our national postal service has begun a series of rotating strikes. My general area isn't affected yet, but we'll see...should be a fairly straightforward assembly once cables come in.
Alright, all back together!
Bit of a preview of @royalba94 's splendorous work :
The part that plugs in the 360 was a bit different. Original 360 on left, @royalba94 's on the right. Ended up trimming the offending pieces with an Xacto knife. (Only one of the two - the other I didn't seem to need to.)
So then came the moment of Truth - and no, it didn't fit. Despite how low profile these cables were, they were still too high. So you'll see what I ended up doing - I trimmed the MeanWell internal shield a bit. There be sharp edges, yes - but nowhere near the cables.
No more pictures after that but you get the idea - it was standard S4M beyond that point.
Overall, this was easily the most challenging build I've ever done. I had to follow even more of a strict build order than a regular S4M build (had to take it almost completely apart), and had less room. .But it was worth it.
Temps seemed to be about 5 deg higher than open air but nothing too bad. GPU was now peaking at 75 instead of 70 or so, but all was well.
Biggest question now is long term dependability - although I don't think it'll happen, my biggest worry is one of the cables' insulation wearing over time.
So yeah - brickless 1070 is definitely possible, but it's work.. If you do end up going down that route and order from @royalba94 , I'd suggest telling him to make the cable ~1 inch longer than he made mine. It was BARELY long enough to connect.
Bit of a preview of @royalba94 's splendorous work :
The part that plugs in the 360 was a bit different. Original 360 on left, @royalba94 's on the right. Ended up trimming the offending pieces with an Xacto knife. (Only one of the two - the other I didn't seem to need to.)
So then came the moment of Truth - and no, it didn't fit. Despite how low profile these cables were, they were still too high. So you'll see what I ended up doing - I trimmed the MeanWell internal shield a bit. There be sharp edges, yes - but nowhere near the cables.
No more pictures after that but you get the idea - it was standard S4M beyond that point.
Overall, this was easily the most challenging build I've ever done. I had to follow even more of a strict build order than a regular S4M build (had to take it almost completely apart), and had less room.
.But it was worth it. Temps seemed to be about 5 deg higher than open air but nothing too bad. GPU was now peaking at 75 instead of 70 or so, but all was well.
Biggest question now is long term dependability - although I don't think it'll happen, my biggest worry is one of the cables' insulation wearing over time.
So yeah - brickless 1070 is definitely possible, but it's work.
. If you do end up going down that route and order from @royalba94 , I'd suggest telling him to make the cable ~1 inch longer than he made mine. It was BARELY long enough to connect.Lots of updates on this.
First off : Got not one, but two Samsung 860 EVO's at 500Gb at awesome Black Friday sales. Will be replacing my tired Seagate and adding a second drive.
Second : We decided to re-purpose the home theater room, where I had my projector - didn't feel right anymore for me to hog this room for myself, only for the occasional gaming session, while we could use it for the whole family, so we made into a playroom for the kids. I also decided to make the S4M build my music production station as well (had a 2012 MacBook Pro for this before), so my little studio setup is also my gaming desk now. What this means, for this project, is that I am now gaming on my shiny, new 1440p S2716DG Dell 27" monitor that I got on a wicked Black Friday deal. Good thing I got the 1070, I guess! One 860 will hold the OS and games, the other will hold Ableton Live and all of my synths/libraries.
Witcher 3 also finally came on sale on Steam a few weeks back, so I bought it. Have a good 20 hours on it since (a most amazing game if you've never played it), and even downclocking the 1070 to 75% power limit, I'm getting power cycles - mostly after hour-plus sessions. I attribute it to the heat - the MeanWell doesn't have a proper heatsink (the case is essentially becoming a bit of one, in effect - it gets very warm) and its capacity drops off pretty significantly past a certain temperature ceiling.
Given the fact I'm also going to use this for music production, stability is now a paramount concern. So, I think I'll drop the brickless requirement and accept the higher weight and size, but for the stability. Am totally happy with my current level of hardware, but just want to give it the power it needs. (Understand that the GPU is not going to be taxed during music production, but still.)
Thinking of using the MeanWell GST280A24. Got the idea from that spreadsheet on /sffpc Reddit, in the part that mentions @Thehack 's eventual Distro board. Basically half the price of the Dell 330W brick - the genuine ones, that is, in Canada anyway - so not worth spending double for 50W. (3-yr vs 1-yr warranty helps too.) That six-pin connector will obviously not fit my current setup though. Couple of questions for you electrical experts :
1. What is the purpose of the small cylindrical bit right behind the connector? Thinking I might have to cut behind it because there's really not much wire between it and the actual connector;
2. My understanding is that, seeing as I've basically committed to the C6 connector on my Mini, I can just route the 24V power through the C8 without issue. That correct? If so, I'd just fuse the output side of the GST280, with my existing C8 power cord, and call it a day. Thoughts?
In the meantime, I'll see how low I have to undervolt the 1070 for long-session stability, and how low I have to drop my Witcher 3 graphics settings...
First off : Got not one, but two Samsung 860 EVO's at 500Gb at awesome Black Friday sales. Will be replacing my tired Seagate and adding a second drive.
Second : We decided to re-purpose the home theater room, where I had my projector - didn't feel right anymore for me to hog this room for myself, only for the occasional gaming session, while we could use it for the whole family, so we made into a playroom for the kids. I also decided to make the S4M build my music production station as well (had a 2012 MacBook Pro for this before), so my little studio setup is also my gaming desk now. What this means, for this project, is that I am now gaming on my shiny, new 1440p S2716DG Dell 27" monitor that I got on a wicked Black Friday deal. Good thing I got the 1070, I guess! One 860 will hold the OS and games, the other will hold Ableton Live and all of my synths/libraries.
Witcher 3 also finally came on sale on Steam a few weeks back, so I bought it. Have a good 20 hours on it since (a most amazing game if you've never played it), and even downclocking the 1070 to 75% power limit, I'm getting power cycles - mostly after hour-plus sessions. I attribute it to the heat - the MeanWell doesn't have a proper heatsink (the case is essentially becoming a bit of one, in effect - it gets very warm) and its capacity drops off pretty significantly past a certain temperature ceiling.
Given the fact I'm also going to use this for music production, stability is now a paramount concern. So, I think I'll drop the brickless requirement and accept the higher weight and size, but for the stability. Am totally happy with my current level of hardware, but just want to give it the power it needs. (Understand that the GPU is not going to be taxed during music production, but still.)
Thinking of using the MeanWell GST280A24. Got the idea from that spreadsheet on /sffpc Reddit, in the part that mentions @Thehack 's eventual Distro board. Basically half the price of the Dell 330W brick - the genuine ones, that is, in Canada anyway - so not worth spending double for 50W. (3-yr vs 1-yr warranty helps too.) That six-pin connector will obviously not fit my current setup though. Couple of questions for you electrical experts :
1. What is the purpose of the small cylindrical bit right behind the connector? Thinking I might have to cut behind it because there's really not much wire between it and the actual connector;
2. My understanding is that, seeing as I've basically committed to the C6 connector on my Mini, I can just route the 24V power through the C8 without issue. That correct? If so, I'd just fuse the output side of the GST280, with my existing C8 power cord, and call it a day. Thoughts?
In the meantime, I'll see how low I have to undervolt the 1070 for long-session stability, and how low I have to drop my Witcher 3 graphics settings...
I'll attribute the power cycles to the fact that you're really pretty much running on >95% your psu's limit, both the brick and the distro. At that point you really should properly monitor and analyse your total power draw, and start thinking on cooling both your psus.
There's another workaround to this brickless setup, which is to use UHP-200-12 just for the gpu, and a smaller 100W acbels or meanwell's on the front of the case for the mobo/cpu. Hdplex 160, if only in stock, could also fit there.
Btw, it's not much of a build log without moar pics.... give peeps some incentives to read lol.
There's another workaround to this brickless setup, which is to use UHP-200-12 just for the gpu, and a smaller 100W acbels or meanwell's on the front of the case for the mobo/cpu. Hdplex 160, if only in stock, could also fit there.
Btw, it's not much of a build log without moar pics.... give peeps some incentives to read lol.
Aren't there a bunch of pics just a few responses above...? . I can post more, no worries, but right now, it's standard S4M from the outside.
I think the limiting factor is the MeanWell only - I do have the 360+160 from KMPKT for the dc-dc side so I think I'm good there. The problem with that though, is that although I think your idea of getting a second, smaller ac-dc for the board only is fantastic (hadn't thought of it myself - that's why I come here!), I'm not sure where exactly I'd fit that second ac-dc unit... There's virtually no room left in the case.
Oh, and @Choidebu : mind expanding on what "acbels" is...? Not sure I'm tracking. Cheers!
. I can post more, no worries, but right now, it's standard S4M from the outside. I think the limiting factor is the MeanWell only - I do have the 360+160 from KMPKT for the dc-dc side so I think I'm good there. The problem with that though, is that although I think your idea of getting a second, smaller ac-dc for the board only is fantastic (hadn't thought of it myself - that's why I come here!), I'm not sure where exactly I'd fit that second ac-dc unit... There's virtually no room left in the case.
Oh, and @Choidebu : mind expanding on what "acbels" is...? Not sure I'm tracking. Cheers!
Last edited:
The little cylinder is a ferrite core. It reduces AC frequency. Pretty much any conducting wire outside a case becomes an antenna and the ferrite core solves that.
