Jump to content

[BENCHMARKS] The OFFICIAL Thread


Brian

Recommended Posts

So I switched 7 of the 980's memory chips over to the 980(m?). I broke the 8th. ~200MHz memory clock gain.

 

1502/7160 firestrike:

 

http://www.3dmark.com/3dm/13957945

 

My gpu scores are just not in line with the normal 980 mxm cards. Extrapolating my score for their clocks I'd get ~16800, while they get 17100. Maybe I'm getting a little optimus penalty? Its strange to be getting it at such low fps though. Usually it only shows up for 200fps+.

 

1495/7160 3dm11: (needed to drop voltage to avoid AC cutoff)

 

http://www.3dmark.com/3dm11/11477667

 

Meanwhile 3dm11 does very well. 21748 is right where a 1495/7160 980 should be.

 

I am eyeing a dead 970 with Samsung memory on ebay now for 8 more 7GHz memory chips.

 

 

UPDATE:

 

I reran firestrike and scored much higher. It looks like something went wrong on that last run. Same clocks as before, but over 16k gpu:

 

http://www.3dmark.com/3dm/13958342

 

firestrike loves memory clocks.

Edited by Khenglish
  • Thumbs Up 5
Link to comment
Share on other sites

2 hours ago, Khenglish said:

So I switched 7 of the 980's memory chips over to the 980(m?). I broke the 8th. ~200MHz memory clock gain.

 

1502/7160 firestrike:

 

http://www.3dmark.com/3dm/13957945

 

My gpu scores are just not in line with the normal 980 mxm cards. Extrapolating my score for their clocks I'd get ~16800, while they get 17100. Maybe I'm getting a little optimus penalty? Its strange to be getting it at such low fps though. Usually it only shows up for 200fps+.

 

1495/7160 3dm11: (needed to drop voltage to avoid AC cutoff)

 

http://www.3dmark.com/3dm11/11477667

 

Meanwhile 3dm11 does very well. 21748 is right where a 1495/7160 980 should be.

 

I am eyeing a dead 970 with Samsung memory on ebay now for 8 more 7GHz memory chips.

 

 

UPDATE:

 

I reran firestrike and scored much higher. It looks like something went wrong on that last run. Same clocks as before, but over 16k gpu:

 

http://www.3dmark.com/3dm/13958342

 

firestrike loves memory clocks.

This is fantastic, how did you do that?

  • Thumbs Up 1
Link to comment
Share on other sites

On 8/8/2016 at 11:41 AM, Khenglish said:

...

 

http://www.3dmark.com/3dm11/11477667

 

Meanwhile 3dm11 does very well. 21748 is right where a 1495/7160 980 should be.

 

...

 

 

 

 

Awesome Khenglish! :thumbsup:

 

Looks like you may want to re-run the 3DM11 bench, too! Your old physics and overall score where a bit higher...or maybe the CPU throttled?

 

http://www.3dmark.com/3dm11/11445450

Link to comment
Share on other sites

I already have to do bench runs with the screen off to avoid AC cutout. That little extra power draw from the CPU when overvolting takes a notch off the GPU voltage. With only 4 older Ivy Bridge cores I'm never winning anything on physics, so I just remove the additional turbo voltage and clock the CPU down 250MHz.

 

I have no idea why I am losing AC power. I found the 2 AC current monitoring resistors on the motherboard and added a 3rd in parallel, so I should have a 50% higher power limit now. Instead I only got a 2W efficiency improvement due to the 2 AC power FETs (improving current delivery capability seemed like a good idea in addition to just raising the current limit) I added and the 3rd parallel monitoring resistor. I cannot find any other circuits which could be monitoring AC power anywhere in the schematics.

 

The motherboard does have an AC undervoltage check circuit. Maybe that is triggering, but I'd be very surprised if it was as the 330W PSU I am using has a .5V higher nominal voltage than the stock 180W PSU. I'd have to be dropping a lot of voltage across the PSU cable and motherboard connector for that to happen. Its not hard to check though so I suppose I will.

 

If it is AC undervoltage due to the cable and connector there is no cheap fix. The motherboard does have the mounting points for the high power PSU connector and I could probably get it from rj-tech. The PSU cable for the correct plug is another story, and I don't see how I can get it without buying a $165 new PSU that is identical to the one I already have. So it looks like ~$200 on just cables and connectors to fix the power problem if it is AC undervoltage.

 

I have no idea why Clevo used a weaker PSU & PSU connector on the P150 series than the P170 series when both systems had identical hardware. As the PSU connector is the only difference between the P170 and P150 motherbaords all it did was require them to have a separate production line for the P150 boards and likely increased their production costs.

  • Thumbs Up 1
Link to comment
Share on other sites

So it turns out I am getting PSU shutdowns. I measured the voltage and AC voltage in cuts to zero on the power loss. I suppose the PSU cuts itself out when it goes a little over the limit, and only fully shuts down if it goes way over.

 

Shutdown occurs at combined 67W CPU + 233W GPU. Add in power losses and other components like memory for another 50W and I lose power around 20W over the 330W PSU rating. I heard these things run up to 450W despite the rating, but I suppose not.

 

I've already doubled up both GPU power supply FETs, and 2 of the 3 AC-in FETs. I am dropping around .2V between the motherboard AC-in and the GPU via board resistance. I could add wires to partially bypass the MXM slot and motherboard, but I can't see that helping by more than .1V, while ruining the ability to measure GPU power draw.

 

I'm getting .1V across the 4 doubled up GPU supply FETs. I could pull the 2 I added and replace them with 2 spare stronger FETs, but that would only help me by .5W.

 

So by only adding some wires to keep GPU power reading and replacing the 2 FETs, I can improve efficiency by a full watt... that's not going to help much. The cable seems to be another 1-2W.

 

I think most of my inefficiency is due to using the 5.5mm x 2.5mm standard motherboard jack on the P150EM. My measurements and math are pointing to that dropping me around .4V, or a little over 7W.

 

So 2-5W improvement by replacing the motherboard and PSU connectors is starting to get significant, but still not an earthshattering improvement. Summing everything up only saves me around 7W.

 

Honestly it looks like the thing to do is just open up the PSU and raise its power limit. This is only for benching anyway, so it shouldn't overheat and die.

Edited by Khenglish
  • Thumbs Up 2
Link to comment
Share on other sites

I would also seize the chance of opening the PSU and changing the power cord, its just tin plated steel/iron in almost all of the power bricks so you can indeed have a 1v or more Vdrop across the cable with such loads.

330Watts at 19v are 17.3A.

If you guess a 1v Vdrop that means that the cable as a resistance of 0.057Ohms, that might be indeed what is happening.

  • Thumbs Up 1
Link to comment
Share on other sites

2 hours ago, senso said:

I would also seize the chance of opening the PSU and changing the power cord, its just tin plated steel/iron in almost all of the power bricks so you can indeed have a 1v or more Vdrop across the cable with such loads.

330Watts at 19v are 17.3A.

If you guess a 1v Vdrop that means that the cable as a resistance of 0.057Ohms, that might be indeed what is happening.

 

Its not aluminum? I am getting 18.7V after the mobo jack. I xan't measure the voltage aftet the cable but before the jack without cutting up the cable. I did do some measurements at the brick itself a while back and it has very good load line regulation, putting out over 19.5V at around 250W of power draw, and less than 19.7V at no load.

Link to comment
Share on other sites

On 8/10/2016 at 8:37 PM, Khenglish said:

So it turns out I am getting PSU shutdowns. I measured the voltage and AC voltage in cuts to zero on the power loss. I suppose the PSU cuts itself out when it goes a little over the limit, and only fully shuts down if it goes way over.

 

Shutdown occurs at combined 67W CPU + 233W GPU. Add in power losses and other components like memory for another 50W and I lose power around 20W over the 330W PSU rating. I heard these things run up to 450W despite the rating, but I suppose not.

 

I've already doubled up both GPU power supply FETs, and 2 of the 3 AC-in FETs. I am dropping around .2V between the motherboard AC-in and the GPU via board resistance. I could add wires to partially bypass the MXM slot and motherboard, but I can't see that helping by more than .1V, while ruining the ability to measure GPU power draw.

 

I'm getting .1V across the 4 doubled up GPU supply FETs. I could pull the 2 I added and replace them with 2 spare stronger FETs, but that would only help me by .5W.

 

So by only adding some wires to keep GPU power reading and replacing the 2 FETs, I can improve efficiency by a full watt... that's not going to help much. The cable seems to be another 1-2W.

 

I think most of my inefficiency is due to using the 5.5mm x 2.5mm standard motherboard jack on the P150EM. My measurements and math are pointing to that dropping me around .4V, or a little over 7W.

 

So 2-5W improvement by replacing the motherboard and PSU connectors is starting to get significant, but still not an earthshattering improvement. Summing everything up only saves me around 7W.

 

Honestly it looks like the thing to do is just open up the PSU and raise its power limit. This is only for benching anyway, so it shouldn't overheat and die.

Despite what is going on with you and the PSU, it's max shut off is 442W to 452W. I get this number by using 5 PSU's 1 at a time. And that is where they cut out at. They are rated at 330W RMS from my understanding. If it was max watts then it would shut off a little before 330W or right at 330W. And from what I was told. They are rated at 330W because that is what the power jack to the board can handle without being a threat to itself. (So I was told)

  • Thumbs Up 1
Link to comment
Share on other sites

So I replaced the PSU cable with one that's twice as thick. This got me an extra .15V to .2V, so a savings of around 3W. So yeah the cable is not aluminum. I do not understand why they would not use aluminum. Aluminum would drop around 1/100th the voltage of this steel cable. I guess a thick cable would cost too much, and a thinner one would break over time?

 

3W is nice, but very hard to notice.

 

I replaced one of the two 8mOhm current reading resistors on the PSU with a 5mOhm and no change in power limit. Maybe I only got one power rail or something and need to find another set.

  • Thumbs Up 2
Link to comment
Share on other sites

You can give even 16mm square electrical wire and further will not result. the weakest link in P150xM is plug/connector. If you do not replace it forget to the high OC. All my P150 are replaced connector/mobo from P170. BTW, My QS K5000M (9300 points in 3DMark) had poor ASIC but thicker PCB than standard OEM card, so better contact.

Edited by Clyde
  • Thumbs Up 2
Link to comment
Share on other sites

1 hour ago, Clyde said:

You can give even 16mm square electrical wire and further will not result. the weakest link in P150xM is plug/connector. If you do not replace it forget to the high OC. All my P150 are replaced connector/mobo from P170. BTW, My QS K5000M (9300 points in 3DMark) had poor ASIC but thicker PCB than standard OEM card, so better contact.

 

When using the standard p150em psu cable on a 330w psu I was getting .4v dropped by the cable, and .2v dropped by the connector. The 330w psu cable is twice as thick as the 180w cable. When I switched the 180w cable with the 330w cable while keeping the 180w connector I gained about .2v. You likely gained another .1v by also switching the connector.

 

I miss the thicker pcb on the Clevo 7970m. The pcb and components on that card were the highest quality I have ever seen on an mxm card, and it outclocked its desktop equivalent, the 7870. Too bad the maxwell cards did not have the higher end pcb.

  • Thumbs Up 2
Link to comment
Share on other sites

  • 3 weeks later...

http://valid.x86.fr/5d4sj0
[IMG]

 

Here's a single 980M world record in Aquamark @ 5.0GHz and GPU at 1500/1800

http://hwbot.org/submission/3309866_
[IMG]

 

wPrime 32M passed at 50x4
[IMG]

I cannot get Fire Strike to pass at 50x4, but here is a 49x4, second place on HWBOT.org behind single 980M on Brother @Johnksss@iBUYPOWER and his Xeon powered P570WM.

http://www.3dmark.com/fs/10096081
[IMG]

Link to comment
Share on other sites

17 minutes ago, Mr. Fox said:

Around 62°C.

Mmm... you needing more voltage goes against what I know about Maxwell, though. Think you can try getting the card down to the mid 40s? Kill the CPU overclock (or lower it to a lesser number just for the testing), turn off the machine, remove the back cover, shove a 15°c A/C vent into the heatsink portion for about 15 minutes until it's COLD, boot it, instantly apply the OC and run the benchmark with maximum fans? See if you manage to get the card rolling with the lesser voltage?

  • Thumbs Up 1
Link to comment
Share on other sites

19 minutes ago, D2ultima said:

Mmm... you needing more voltage goes against what I know about Maxwell, though. Think you can try getting the card down to the mid 40s? Kill the CPU overclock (or lower it to a lesser number just for the testing), turn off the machine, remove the back cover, shove a 15°c A/C vent into the heatsink portion for about 15 minutes until it's COLD, boot it, instantly apply the OC and run the benchmark with maximum fans? See if you manage to get the card rolling with the lesser voltage?

No, that won't do it. It needs more voltage to go past 1500ish on core. Around 1535 or so the driver crashes (TDR) about 5 to 10 seconds into a benchmark due to lack of core voltage when the GPU is still around 30°C. It's not having black-screen issues, artifacts or lockups or anything like that. This 980M has been hard modded (added MOSFETs and memory voltage mod).

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...

Important Information

By using this site, you agree to our Terms of Use. We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue.