S-XPS 1645 AC Power Throttle Issue Investigation

[max420]

okay, ill have to do that ina few hours, seems i have to go and fix something. lol

i was satisfied after my gaimng problems went away, but ive never really checked to see if that fixed my throttling....if not, i shouldve just bought the 720qm.

[Mitchell2.24v]

I've been reading (most of) this thread with quite some interest as I'm about to order a 1645 with 720/256SSD/RGBLED/4G, and therefore will probably run into this problem while gaming.

What I'm wondering is this, and please excuse me if this has been discussed before, I haven't found it.

1. Had anyone tried the test without a battery installed? Could the charging of the battery cause a problem?

2. If I understand correctly, the i7's determine their max overclock partially based on temperature. Could the reverse also happen and therefore downclock the i7? Maybe the cooling system is part of the problem? I wouldn't know why using a 130W brick would help in this case, but then again I'm no expert in this field

3. Has anyone noticed the throttling during 'regular' work? I'm buying the machine to do development work on, and plan to run a VM next to Visual Studio and SQL server. MY guess is that it will do just fine, and I will just have some trouble while gaming.

[2by4]

Originally Posted by daraj

TriangleLaptops doesn't seem to have the 130W anymore. Any other options? I don't want to get rapped by ordering from Dell.

Just Google JU012 adapter and you will get many places to buy from. You should try to have Dell ship one to you for free.

Note: Make sure that it is the New Slim Model

[eblock12]

Originally Posted by Mitchell2.24v

1. Had anyone tried the test without a battery installed? Could the charging of the battery cause a problem?

Yes, this doesn't change the behavior.

Originally Posted by Mitchell2.24v

2. If I understand correctly, the i7's determine their max overclock partially based on temperature. Could the reverse also happen and therefore downclock the i7? Maybe the cooling system is part of the problem? I wouldn't know why using a 130W brick would help in this case, but then again I'm no expert in this field

Turbo boost lets it "overclock" one or more cores when there is enough temperture headroom and there's the right workload. But the CPU should atleast be able to hit the 12x multiplier (on the 720) even without turbo boost. Also the tempertures don't seem to be an issue on this machine.

Originally Posted by Mitchell2.24v

3. Has anyone noticed the throttling during 'regular' work? I'm buying the machine to do development work on, and plan to run a VM next to Visual Studio and SQL server. MY guess is that it will do just fine, and I will just have some trouble while gaming.

It's really hard to measure because under normal workloads most of the CPU cores enter idle states and it throttles constantly under normal operation. So unless you deliberately load the CPU with some benchmark/game/test its not possible to determine if its underperforming or not.

[crazycrab]

The new BIOS update A03 is out. Has anyone tried it out yet? Does it solve the problem? is it any better?

[Mitchell2.24v]

So much for my contribution

Originally Posted by eblock12

Turbo boost lets it "overclock" one or more cores when there is enough temperture headroom and there's the right workload. But the CPU should atleast be able to hit the 12x multiplier (on the 720) even without turbo boost. Also the tempertures don't seem to be an issue on this machine.

I have an i7 860 in my desktop (which is why I now also want an i7 in my laptop ), so I did some reading into the i7. What you say makes sense, and if there is no temperature issue, there goes my theory.

Originally Posted by eblock12

It's really hard to measure because under normal workloads most of the CPU cores enter idle states and it throttles constantly under normal operation. So unless you deliberately load the CPU with some benchmark/game/test its not possible to determine if its underperforming or not.

I'm using this sidebar gadget to monitor the cores (can't remember its name) on my desktop, and during regular work a few cores sometimes peak out, but hardly ever all four (or eight threads). So I assumed for work I'd be fine with the 1645.

Thanks for the info.

Mine is scheduled for january 25th, so maybe they are working on something and delaying deliveries for a while... Once it gets here I will let you all know what's up (BIOS, powerbrick, etc). And I will keep my eye on this thread...

Can anyone compare an 720QM to an 860 from experience? I know my current laptop (P8700 @2.53Ghz) is terribly slow compared to the 860, and the 720 won't work wonders, but I'm hoping the extra threads wil help a bit, and hopefully the SSD will do some good as well.

[atlstang]

Originally Posted by max420

okay, ill have to do that ina few hours, seems i have to go and fix something. lol

i was satisfied after my gaimng problems went away, but ive never really checked to see if that fixed my throttling....if not, i shouldve just bought the 720qm.

look forward to your results.

looks like dell really does want to get my laptop, which i reluctantly agreed too as long as theres no interruption etc. I was thinking this is probably a good opportunity to put all our testing results / utilities etc, on the desktop so that they have the info right in front of them. Hate to do the work for them, but we also want this problem fixed.

anyone with results and specs ill put in a document for the "engineers" to see hopefully.

[PsYcNeT]

So I'm trying out a few things while bored at work today with the brightness part of the problem. You guys all like Prime95 and Intel Burn Test, but I'm more partial to Cinebench 10 64-bit mulitcore test for pure CPU pushing.

Cinebench10 64 makes all C0 states stay at 100 for the test duration, so I decided to try it in conjunction with Furmark for some AC tests.

Setup: 720qm, WLED, 90w IN, restart after every test, battery NOT fully charged.

The 1645 has 15 levels of brightness, which work out to about 7% per, give or take.

Test 1: CB10 alone
All brightness levels have no effect, multiplier runs solid ~12.1x for entire test, with all C0 levels at 99.9-100.

Test 2: CB10 w/ Furmark @ 800x600 (window)
I start Furmark with the render toggled 'off' (spacebar) which means the GPU is doing no work. ~12.1x multi and ~100 C0. When Furmark is toggled 'on', the multiplier immediately drops to 7.0, and the C0 level is halved. This is at '0%' brightness.

Test 3: Furmark alone, 800x600 window, incrementally increasing brightness every 20 seconds
Every time there is a brightness increase, the multiplier dips to 8.0-10.0x, but soon rises back to 9.5-12x about 1-2 seconds later. Once you pass brightness level 12, it begins to dip to 7.0x before rising back up, but starts to fluctuate between 7.0-9.0x. At level 14 (1 below max), it locks down to 7.0x, never to rise again. C0 levels averaged ~14-15% the entire test.

Test 4: Furmark alone, 1024x768 window, incremental brightness increase again
More or less the same deal as above, except this time it bombed down to the 7.0-9.0x multiplier at level 9, and settled on level 13 at 7.0x. Interestingly, at levels 11/12, it would drop to 7.0x and hang, but rise to 8.0x at times. The C0 level was anywhere from 11-20%.

Test 5: Furmark alone, fullscreen 1600x900, incremental
Level 0 runs at 10-11x (no higher). Level 4 shows 8-11x. Hits 7.0-9.0x at level 7. Basements (hits 7.0x permenantly) at level 12.

In summation (TLDR): The less intensive the GPU use, the higher the brightness can be before the CPU basements. Fullscreen gaming is best done at levels 0 through 3, but doesn't stutter and drag-lag till ~12. I assume the numbers are probably even lower in more intensive games.

[theoak]

Originally Posted by PsYcNeT

So I'm trying out a few things while bored at work today with the brightness part of the problem. You guys all like Prime95 and Intel Burn Test, but I'm more partial to Cinebench 10 64-bit mulitcore test for pure CPU pushing.

Cinebench10 64 makes all C0 states stay at 100 for the test duration, so I decided to try it in conjunction with Furmark for some AC tests.

Setup: 720qm, WLED, 90w IN, restart after every test, battery NOT fully charged.

The 1645 has 15 levels of brightness, which work out to about 7% per, give or take.

Test 1: CB10 alone
All brightness levels have no effect, multiplier runs solid ~12.1x for entire test, with all C0 levels at 99.9-100.

Test 2: CB10 w/ Furmark @ 800x600 (window)
I start Furmark with the render toggled 'off' (spacebar) which means the GPU is doing no work. ~12.1x multi and ~100 C0. When Furmark is toggled 'on', the multiplier immediately drops to 7.0, and the C0 level is halved. This is at '0%' brightness.

Test 3: Furmark alone, 800x600 window, incrementally increasing brightness every 20 seconds
Every time there is a brightness increase, the multiplier dips to 8.0-10.0x, but soon rises back to 9.5-12x about 1-2 seconds later. Once you pass brightness level 12, it begins to dip to 7.0x before rising back up, but starts to fluctuate between 7.0-9.0x. At level 14 (1 below max), it locks down to 7.0x, never to rise again. C0 levels averaged ~14-15% the entire test.

Test 4: Furmark alone, 1024x768 window, incremental brightness increase again
More or less the same deal as above, except this time it bombed down to the 7.0-9.0x multiplier at level 9, and settled on level 13 at 7.0x. Interestingly, at levels 11/12, it would drop to 7.0x and hang, but rise to 8.0x at times. The C0 level was anywhere from 11-20%.

Test 5: Furmark alone, fullscreen 1600x900, incremental
Level 0 runs at 10-11x (no higher). Level 4 shows 8-11x. Hits 7.0-9.0x at level 7. Basements (hits 7.0x permenantly) at level 12.

In summation (TLDR): The less intensive the GPU use, the higher the brightness can be before the CPU basements. Fullscreen gaming is best done at levels 0 through 3, but doesn't stutter and drag-lag till ~12. I assume the numbers are probably even lower in more intensive games.

What version of the BIOS do you have?

[atlstang]

post your full specs if ya could. thats cool your cpu doesnt throttle under the intel test, is that after the fans etc start going? i have the 1080p screen maybe that takes more power? as the intel test seems to put it right on the very edge of throttling for me anyway. ill start saving these.

it does seem to be a balance between gpu load and cpu load. furmark loads the gpu 100%. I would assume that if you just load 1 core with prime95 or whatever, and start furmark you would get the same throttling. Much like i see on my own machine.