Yea. Using Wattman is good for that, I hear. Would you decrease the watts per each CPU state, or just the last speed state?
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Getting my 5,1 ready for Vive VR!
- Thread starter fendersrule
- Start date
- Sort by reaction score
I personally love to fine tune all states, but in general, only the last state matters.
Alrighty, finally starting to tweak with undervolting.
My previous Heaven score was 46FPS w/ +20 Power Limit.
I under-volted by 50mV across states 3 - 6 (1150 -> 1100). That netted me about 46.5 FPS. The reason is due to slightly less throttling.
I then proceeded to raise the GPU clock from 1340 to 1360 while keeping the mV at 1100. That needed me 47.1 FPS, highest FPS yet.
I then proceeded to raise the GPU clock from 1340 to 1400 which ended up in a crash (as expected). Sucks how Wattman throws everything away and you have to rebuild everything again from scratch.
I then proceeded to add 20mV to State 6 (1100 -> 1120) and try 1400Mhz again. That ended up in 46.9 FPS, no gain from 1360Mhz @ 1100mV. The reason for no gain was throttling began to occur, it seldomly hit 1400 which means it's power constrained.
So, there you have it. I feel this is the best it's going to get without pushing the Power Limit past 20%. For now, I feel like I have something right in the middle. It's undervolted by 50mV, and also overclocked by 20MHz and that's what provides the best performance (and safety) at the current +20 power limit.
unless h98 has any additional ideas (or gives me the go-ahead to push the power limit further), this is where it may sit (and I'm happy). 1400Mhz sounded like a really good number, but the only way to get higher than 1360 (and for it to STAY THERE CONSISTENTLY DURING GAMING) is to push the power limit.
185w x 20% = 222w
185w x 21% = 223.85w
I suppose I could up it to 21 %.
My previous Heaven score was 46FPS w/ +20 Power Limit.
I under-volted by 50mV across states 3 - 6 (1150 -> 1100). That netted me about 46.5 FPS. The reason is due to slightly less throttling.
I then proceeded to raise the GPU clock from 1340 to 1360 while keeping the mV at 1100. That needed me 47.1 FPS, highest FPS yet.
I then proceeded to raise the GPU clock from 1340 to 1400 which ended up in a crash (as expected). Sucks how Wattman throws everything away and you have to rebuild everything again from scratch.
I then proceeded to add 20mV to State 6 (1100 -> 1120) and try 1400Mhz again. That ended up in 46.9 FPS, no gain from 1360Mhz @ 1100mV. The reason for no gain was throttling began to occur, it seldomly hit 1400 which means it's power constrained.
So, there you have it. I feel this is the best it's going to get without pushing the Power Limit past 20%. For now, I feel like I have something right in the middle. It's undervolted by 50mV, and also overclocked by 20MHz and that's what provides the best performance (and safety) at the current +20 power limit.
unless h98 has any additional ideas (or gives me the go-ahead to push the power limit further), this is where it may sit (and I'm happy). 1400Mhz sounded like a really good number, but the only way to get higher than 1360 (and for it to STAY THERE CONSISTENTLY DURING GAMING) is to push the power limit.
185w x 20% = 222w
185w x 21% = 223.85w
I suppose I could up it to 21 %.
Last edited:
Alrighty, finally starting to tweak with undervolting.
My previous Heaven score was 46FPS w/ +20 Power Limit.
I under-volted by 50mV across states 3 - 6 (1150 -> 1100). That netted me about 46.5 FPS. The reason is due to slightly less throttling.
I then proceeded to raise the GPU clock from 1340 to 1360 while keeping the mV at 1100. That needed me 47.1 FPS, highest FPS yet.
I then proceeded to raise the GPU clock from 1340 to 1400 which ended up in a crash (as expected). Sucks how Wattman throws everything away and you have to rebuild everything again from scratch.
I then proceeded to add 20mV to State 6 (1100 -> 1120) and try 1400Mhz again. That ended up in 46.9 FPS, no gain from 1360Mhz @ 1100mV. The reason for no gain was throttling began to occur, it seldomly hit 1400 which means it's power constrained.
So, there you have it. I feel this is the best it's going to get without pushing the Power Limit past 20%. For now, I feel like I have something right in the middle. It's undervolted by 50mV, and also overclocked by 20MHz and that's what provides the best performance (and safety) at the current +20 power limit.
unless h98 has any additional ideas (or gives me the go-ahead to push the power limit further), this is where it may sit (and I'm happy). 1400Mhz sounded like a really good number, but the only way to get higher than 1360 (and for it to STAY THERE CONSISTENTLY DURING GAMING) is to push the power limit.
185w x 20% = 222w
185w x 21% = 223.85w
I suppose I could up it to 21 %.
IMO, just use your current "optimised" setting is fine. There is no point to go too far for possible extra 1-2 FPS.
-50mV is actually pretty good.
Wattman must reset everything to default after crash, otherwise your computer may crash straight away after driver reload / reboot. No time for you to perform corrective action.
I won't recommend anyone to go beyond +20 with this card on a cMP. It's close enough to 225W.
Since your card only has one ROM, I won't recommend you to do any BIOS editing. For gaming / VR, you should use Windows anyway, so Wattman is good enough.
Alright! I've reached the end of Phase 2 of undervolting the RX 580. Amazing results so far. I feel like I have some headroom left, but I thought I'd recap:
Regarding Heaven 4.0 @ 1440p:
Out of box = 44.9 FPS.
Added +20 Power Limit = 45.4 FPS
Reduced voltages by 50mV = 46.6 FPS.
Added 20MHz to the core clock = 47.1 FPS.
Took another 10mV off = 47.4 FPS.
That's a gain of 3 FPS right out of the box with optimizations @ 1090mV.
Regarding Steam VR, I went from a 6.8 to a 6.9. As expected.
Good ol' Super Position @ 1440p
Before:
After:
Regarding Superposition VR Optimum...some REALLY interesting results here. Screw the final number, look at the huge differences on the numbers that count.
Before:
After:
Regarding Heaven 4.0 @ 1440p:
Out of box = 44.9 FPS.
Added +20 Power Limit = 45.4 FPS
Reduced voltages by 50mV = 46.6 FPS.
Added 20MHz to the core clock = 47.1 FPS.
Took another 10mV off = 47.4 FPS.
That's a gain of 3 FPS right out of the box with optimizations @ 1090mV.
Regarding Steam VR, I went from a 6.8 to a 6.9. As expected.
Good ol' Super Position @ 1440p
Before:
After:
Regarding Superposition VR Optimum...some REALLY interesting results here. Screw the final number, look at the huge differences on the numbers that count.
Before:
After:
Last edited:
The superposition VR score is interesting. You have higher min and max FPS, but lower average. Anyway, the real single most important to define the gaming experience usually is the min FPS, not the average. So, that's a 18% improvement, looks very good to me.
Final mV set at 1060. I didn't even hit the floor limit either, but I finally hit a point where throttling was just about non-existent. FPS gains were starting to become very minuscule. No point to go further. Check out these bad ass numbers:
These gains are significant IMO. We're talking up to 10% faster overall performance, nearly 30w in less power draw, cooler GPU, and quieter fan speeds.
RX 580 @ 1060mV:
RX 580 (1150mV):
[doublepost=1506658054][/doublepost]RX 580 @ 1060mV:
RX 580 (1150mV)--THIS WAS A LUCKY-ASS BENCH. Usually it's 41ish:
[doublepost=1506658857][/doublepost]RX 580 @ 1060mV:
RX 580 (1150mV)--THIS IS THE HIGHEST I've seen it go. Usually it's 6.6-6.7:
[doublepost=1506659750][/doublepost]RX 580 @ 1060mV:
RX 580 (1150mV):
These gains are significant IMO. We're talking up to 10% faster overall performance, nearly 30w in less power draw, cooler GPU, and quieter fan speeds.
RX 580 @ 1060mV:
RX 580 (1150mV):
[doublepost=1506658054][/doublepost]RX 580 @ 1060mV:
RX 580 (1150mV)--THIS WAS A LUCKY-ASS BENCH. Usually it's 41ish:
[doublepost=1506658857][/doublepost]RX 580 @ 1060mV:
RX 580 (1150mV)--THIS IS THE HIGHEST I've seen it go. Usually it's 6.6-6.7:
[doublepost=1506659750][/doublepost]RX 580 @ 1060mV:
RX 580 (1150mV):
Last edited:
HTC has a superb bundle deal with the Vive that includes the GTX 1070 (and FO4 VR) and I jumped on it. I was thinking to simply just sell the GTX 1070, but I think I will eat my money and try it out on the cMP.
Stay tuned for more benchmarks, as well as some feedback about the nVidia web drivers. I expect the GTX 1070 to blow the RX580 away, although, the RX580 is a great card.
I've already passed phase 1 with flying colors. Last night I installed the 10.12.6 web driver on a RX580 without any hitches. So I learned that the web driver installation does not need the nVidia card to be installed, which makes life easier. Neat little menu-bar App that allows me to switch the web driver off and on.
So far, smooth as silk as I begin switching over to nVidia.
I may use the RX580 for awhile with the Vive before switching over to the GTX 1070. So it may be a little while before I update this thread again.
Stay tuned for more benchmarks, as well as some feedback about the nVidia web drivers. I expect the GTX 1070 to blow the RX580 away, although, the RX580 is a great card.
I've already passed phase 1 with flying colors. Last night I installed the 10.12.6 web driver on a RX580 without any hitches. So I learned that the web driver installation does not need the nVidia card to be installed, which makes life easier. Neat little menu-bar App that allows me to switch the web driver off and on.
So far, smooth as silk as I begin switching over to nVidia.
I may use the RX580 for awhile with the Vive before switching over to the GTX 1070. So it may be a little while before I update this thread again.
Really interested to hear how the Vive goes on the cMP - started looking at VR Ready gaming PCs recently to see what the world was like (ugly and overly inspired by Lamborghini's design language) over there. Then discovered Dell doesn't let you build to order any more - wtf when did that happen?
Will have benches posted this week.
The 5,1 with a 580 or a 1070 runs VR flawlessly. Tested both.
1070 allows to crank up some super sampling. Rx580 I never pushed it, but it's certainly capable for VR.
Sneak preview: 1070 is 50% faster over a rx580 on a 5,1 in super position VR and Steam VR bench.
Im converted to Team Green.
The 5,1 with a 580 or a 1070 runs VR flawlessly. Tested both.
1070 allows to crank up some super sampling. Rx580 I never pushed it, but it's certainly capable for VR.
Sneak preview: 1070 is 50% faster over a rx580 on a 5,1 in super position VR and Steam VR bench.
Im converted to Team Green.
Now, we just need to get Tim Cook and the Amigos to at least support Team Green. Apple's open hostility to Team Green isn't good for many of their customers.
I have a 5,1 12-core with 64 gigs of memory. I ran the Vive on it under Bootcamp before my MVC GTX 970 died. It was an "OK" experience, but vive runs so much better on a dedicated Windows box with an i7 processor, 16 GB of memory, and the GTX 1070. I paid $1200 USD for it about a year ago. I still hate Windows (Win10), but the Vive is just a significantly better experience on a Windows box than it is under Bootcamp.
That's due to the 970, not the 5,1. The 970 is scraping on the barest minimum possible. Vive is awesome on a 5,1 in my experience and has plenty of horsepower. And I supersample to 1.3x. Headset hasn't notified me of any dropped frames yet.
Surprised that the westmere can still be relevant today.
Not sure how it's "significantly" better when the experience of a 5,1 and a 1070 is butter. And I was impressed with how the rx580 fared. VR is GPU intensive.
Surprised that the westmere can still be relevant today.
Not sure how it's "significantly" better when the experience of a 5,1 and a 1070 is butter. And I was impressed with how the rx580 fared. VR is GPU intensive.
Last edited:
It was not the GPU that I didn't like. It was the Vive under Bootcamp that was unsatisfactory. I've never been happy with Bootcamp, but I know a lot of people who have been happy with it.
My friend, you really make me seriously deciding how to switch to green team.
Since I don't want to do any mod or use external PSU, I am now having 3 plans.
1) Get a 250W 1080Ti, target is the EVGA SC2 with icx
2) Get a factory overclocked 1080, the target is Gigabyte AORUS Xtreme (review shows max TDP ~250W)
3) Get 2x 150W TDP 1070, target EVGA gaming ACX 3.0
Option 1. Even though it's a 250W card (lowest in 1080Ti), but few guys did run into the shutdown issue even with the FE. And quite a few reviews pointed out that 1080Ti can pull up to 280W indeed. It’s the best performance single card solution, but I don't want to be forced to perform the pixlas mod (if anything goes wrong).
Option 2. Clearly the cMP can drive a 1080 without any issue. So, I pick this highly factory overclocked 1080 as my 2nd single card option. I am pretty sure it won't cause any power issue. This should be the safest high performance single card solution.
Option 3. Dual 1070, self create side by side split frame rendering on my 32:9 monitor. I am sure each single mini 6pin can power a 970 without issue. But not seeing may talk about power a 1070 with single mini 6pin. Calculation and review looks good to me (by using mini 6pin can actual deliver 120W, and the target 1070 should really stop pulling at around 150W). This should give me best performance in most case, however, dual card also means more potential trouble.
In any case. I expect the card will pull about 50W from the slot. Therefore all 1070, 1080 and 1080Ti will pull 100-110W from each mini 6pin. Of course I will balance the load, but seems 1070 and 1080 can stay very close to 100W, and 1080Ti is really right at the edge (110W). Any suddenly extra demand can trigger the shut down protection.
Last edited:
My 1080 FE runs beautifully... in BootCamp.
I look forward to your results in macOS! I found the 1070 FE with the WebDriver to be too buggy for my liking.
I very much trust your postings so I am excited to hear how you get on with NVIDIA.
My 1080 FE runs beautifully... in BootCamp.
I look forward to your results in macOS! I found the 1070 FE with the WebDriver to be too buggy for my liking.
I very much trust your postings so I am excited to hear how you get on with NVIDIA.
I've yet to see any bugs with the 1070 in MacOS.
The Pixlas mod is a bit intimidating
I have confidence to perform the work. But I am not sure which cable have to tap on. That's the problem. And I didn't use voltmeter for at least 20 years already, that make me even harder to confirm if I perform the right mod.
This is why I want to avoid it at this moment.
Anyway, I am now still prefer the 1080Ti more. I prefer single power card solution. At this moment. I found that three 1080Ti may able to work with just the mini 6pins.
1) The FE
2) EVGA dual fan model P/N: 11G-P4-6391-KR
3) Galax EXOC
But I really want to avoid the FE, from what I studied, it's the slowest 1080Ti, but not necessary the lowest power consumption one. So, no point to go for the worst cooling solution.
EVGA looks good to me. It's a 250W TDP card, however, I can't found a reliable review for this exact card yet. A review is about it's similar model, which seems target at TDP 270W in real world. So, not sure if this is a good one yet. But technically, This 6391 model should work.
EXOC, just know this card an hour ago. At this moment, it seems the best candidate in terms of power consumption. I found this graph in Tom's Hardware review.
Its BIOS set the target at 250W but not higher. Also, the momentary peak is just 300W (some other card can go well above 400W, this may be the reason why some 1080Ti do trigger the cMP's shutdown protection).
I will keep studying, and found out if there is any higher success rate solution.
[doublepost=1509981648][/doublepost]BTW, anyone know if those card's back plate will be an issue?
Last edited:
IDK, but I don't think there's a such thing as a slow 1080Ti. Any of them is still the fastest card in the world, which 99.9% of the population doesn't even need or should care about because they don't have a use case that requires a 1080Ti.
The 1080 already is overkill for most people.
The 1080 already is overkill for most people.
I know the 1080 is good enough for gaming. I just try to discover the possibility to use this almost the fastest card to push my monitor to limit. I think even the 1080Ti, it's impossible to push all 3840x1080 pixels in max setting but still deliver 144FPS.
In fact, I already have the plan in my mind about how to handle this power hunger card.
1) Get the lowest TDP 1080Ti, and looking for best cooling solution (hot chip draw more power).
2) Perform the heavy load benchmark test.
3) If it really shutdown my Mac. It should be very easy to lower the TDP in Windows by just running Afterburner (or any software that specifically design for the green team). As you said, there is no slow 1080Ti. Lower the TDP may lower the performance, but still good enough, and I can enjoy a cooler card as well.
In MacOS, my highest work load nowadays is just FCPX. Not heavy at all. So, as long as I don't touch Furmark and CUDA-Z heavy test. Everything should be fine. Since my joy is not running benchmark, so, it's not big deal for me.
My friend, you really make me think, and make me step out my comfort zone (using non OOTB card).
[doublepost=1509985582][/doublepost]
Thanks, in this case, I almost certain will go for the Galax 1080Ti. In fact, the exact model I am looking for is available in Hong Kong (EVGA is not). Just need to confirm if it's in stock, and I am good to go.
If end up doesn't work (either super glitch in software or hardware), I am sure I can sell it to some of my friend with reasonable price. In fact, I am not too worry about the power draw. What worry me a bit more is the driver support. The feedback from other really vary a lot. From flawless to totally unusable. From memory, I never see any other hardware has this kind of inconsistence user experience.
Anyway, once I confirm there is a card in stock, I will get one and test it myself.