Overclocking and undervolting my Llano based laptop

[SlowSpyder]

A couple of years ago I bought a $450 Toshiba L755D-S5218 laptop with an AMD A6-3400m (1.4GHz/2.3GHz 1.35v/1.075v base/turbo, 320 VLIW5 shader GPU @ 400MHz). I stuck an SSD in it and doubled the memory to 8GB (4x2) 1333MHz.

I used K10stat to tweak the APU. So far I've been able to overclock it to 2.867GHz/1.650GHz turbo, all while undervolting it! I have the turbo voltage at 1.265, base voltage is 1.0125. I started by undervolting it figuring I'd save a little power for free, run cooler to get turbo clocks more often and longer, but I kept going and it kept passing stability tests, so I decided to add some clock to see where it would end up. I have to think because turbo only kicks in for a short-ish time it makes the higher clock easier to achieve than if I was trying to run that clock full time.

Also, I used a Kill-o-Watt and am using an honest ~3-3.5 watts less power use at the wall socket when loading all the cores, so I have to imagine I'll see a little extra battery life. I figured I could go for high clocks since power use goes up/down with frequency a lot less than it goes up/down with voltage. I figured if the Kill-o-Watt stayed under the factory settings (pulling ~42.x watts max) while I overclocked, than the heatsink/fan should be able to cope... so in other words, if the CPU was willing, keep going even if it seems high for a laptop.

Anyway, enough blabbing, I have a few benchmarks and power use numbers to share. I should add, the laptop uses ~17 watts according to the Kill-o-Watt when completely idle with the display at max. When the display auto turns off, the laptop uses ~9.5 watts while idle, and of course no display.


CPU Mark 99 (yea... old, but I had it downloaded, so why not )

Factory clocks (uses 27-28 watts while testing)
237

Overclock (uses 25-26 watts while testing)
271


Cinebench 11.5

Factory clocks (Kill-o-Watt shows 40-42.5 watts while testing)
1.77

Overclock (Kill-o-Watt 37-39 watts while testing)
2.01


HyperPi 1 Million / Normal Priority

Factory clocks
1 CPU (Kill-o-Watt shows 29-30 watts while testing)
38.602 seconds

4 CPU (Kill-o-Watt shows 38.5-40 watts while testing)
Best 46.930
Worst 48.460
Avg. 47.524

Overclock
1 CPU (Kill-o-Watt shows 27.5-28 watts while testing)
34.055 seconds

4 CPU (Kill-o-Watt shows 37.5-38.5 watts while testing)
Best 43.835
Worst 44.850
Avg. 44.098


I'm going to run some more benchmarks later, but getting late tonight. I'll grab Cinebench 15, but what else?

I also bought an A8-3500m (2.4GHz/1.5GHz 400 VLIW5 shaders @ 444MHz That's 25% more shaders and 11% more clock speed) for dirt cheap on Ebay a while back. Both have the same 35 watt TDP, the A8-3500m being the highest end 35 watt part for this socket. It'll be fun to overclock this APU in this laptop, plus it gets the added advantage of Noctura NT-H1. I'll do some graphics benches too. If you have an APU laptop, check out K10stat....!

 

[Enigmoid]

Interesting. Could you throw some 3dmark on there?

This appears to be something AMD has room to improve on (intel as well but they are not quite as bad). A lot of their chips on stock run excessively high voltages for their operating clockspeed.

How does turbo work? A lot of llano reviews I read a while back said the turbo implementation was poor.

 

[SAAA]

So... 20% more performance all while using less power. Anyone has an idea on why AMD didn't do this by themselves? Maybe not all those APUs can overclock this much?
Anyway all the better for the end user, I wish I could do the same on my laptop...

 

[SPBHM]

I think 2.01 is more or less the same as the Athlon 5350? but llano IGP should be clearly faster, specially with dual channel ram...

but nice OC, a shame you can't OC the GPU

 

[Soulkeeper]

The stars cores in the llano are good for 3.2 to 3.4GHz on the desktop side, usually with lower than stock voltage.
I own a couple of them and they all top out around there.
I think AMD increased the stock voltages for profit reasons, ie: a few more % would pass validation.

Nice to see you were able to get some more out of a laptop.

 

[SlowSpyder]

Interesting. Could you throw some 3dmark on there?

This appears to be something AMD has room to improve on (intel as well but they are not quite as bad). A lot of their chips on stock run excessively high voltages for their operating clockspeed.

How does turbo work? A lot of llano reviews I read a while back said the turbo implementation was poor.

Sure, I'll download and run tonight.

The turbo is pretty poor in my opinion, but I believe it is improved in newer model APU's. While watching CPU-Z while benching the majority of the time, even with single threaded benches, the cores run at the highest base clock for all cores, in my case 1.65GHz (1.4GHz factory). You'll see it jump up to 2.86GHz for a few seconds here and there. It would jump up to 2.3GHz just the same before I touched the clocks. But, it never stays there than for more than a few seconds at a time. That's what started this whole thing, I wanted to undervolt to see if I could get more turbo more often, but when my scores weren't budging I went for clock speed.

I think AMD may not monitor power/heat in real time, but uses an algorithm to ramp clocks based on load and what would be power use at 1.35 volts. So that would explain why as I lowered the power use I continued to get the exact same Cinebench score through three or four voltage drops. Or, like I said, I could be wrong and the few watts I have saved just isn't enough to have an effect on the clocks?

 

[SlowSpyder]

I think 2.01 is more or less the same as the Athlon 5350? but llano IGP should be clearly faster, specially with dual channel ram...

but nice OC, a shame you can't OC the GPU

Yea, I thought the same thing, this puts me in 5350 territory. But, if I load up my GPU I should still be closer to 35 watts (though I'm sure I could tweak clocks and voltages to get close to 25 watts and probably not be far behind the 5350 on the CPU side, and faster on the GPU). Also, to be fair the 5350 can overclock too. But for how old this core is, I'm happy with the free level of extra performance I am able to get.

 

[TrulyUncouth]

Thanks for posting this, very interesting- even made me look online for prices on similar models just to play around with. How would you describe the day-to-day performance? Does it run netflix and youtube well enough at those clocks? Any specific games you play and how do they perform?

 

[SlowSpyder]

Thanks for posting this, very interesting- even made me look online for prices on similar models just to play around with. How would you describe the day-to-day performance? Does it run netflix and youtube well enough at those clocks? Any specific games you play and how do they perform?

Games it sees are usually lighter weight. Games like Fate/Torchlight/Torchlight II, Minecraft, Titan Quest, Divine Divinity (2D) etc. I've even played a bit of Skyrim (pre overclock) on it. I will definitely add a few game benches/impressions if I can, both at low res and what I'd actually play the game at on this machine. One old game that it was too slow to play before was Beyond Divinity, a semi 3D older hack & slash RPG, I'll see how it is with the CPU bump tonight.

With the factory clocks the machine was a bit 'clunky' to use in Windows, now it is very smooth and responsive. I never had a problem with Youtube/Netflix at the factory clocks, so I would expect no issues with even faster clocks.

Last night I got it up to 2.967GHz @ 1.25v, but got a BSOD while benching. So, it would seem ~3GHz turbo is achievable while still using less wattage than factory if I was willing to bump the voltage. I'll probably leave it at 2.867GHz and be happy there. But I think I want to work more on the base clock since the APU seems to spend most of it's time there, that is the clock speed that seems to affect my scores the most.

 

[Instan00dles]

would love to try this with my laptop, not too concerened with it being fast but I would love to get some extra battery life out of the thing.

 

[SlowSpyder]

Ran 3DMark11 @ 1650MHz/2867MHz (doubt it really ever went into turbo with the GPU maxed).

471 - Firestrike 1.1 (graphics 470 physics 2413 combined 215)
3186 - Cloudgate 1.1 (graphics 4171 physics 1745)
35561 - Icestorm 1.2 (graphics 46137 physics 19732)

My score is better than 7% of all results. ;p

Guess I won't be joining the hall of fame. More of a GPU test than a CPU test. But, it'll be nice to revisit this bench once I have that A8-3500m in. Downloading Cinebench 15 now, will update soon.

Anyone know of a way to overclock the GPU? I'm not sure if you can with Llano APU's.

 

[SlowSpyder]

169 in Cinebench R15. Just found my Kill-o-Watt, going to rerun this test at factory clocks and again at 1650MHz/2867MHz.

I have my Vishera up to 5.3GHz and counting in my desktop, just lapped that thing all the way down to the copper. Splitting my time between the heavy iron and the light and efficient laptop. Can't... stop... benching...

 

[TrulyUncouth]

169 in Cinebench R15. Just found my Kill-o-Watt, going to rerun this test at factory clocks and again at 1650MHz/2867MHz.

I have my Vishera up to 5.3GHz and counting in my desktop, just lapped that thing all the way down to the copper. Splitting my time between the heavy iron and the light and efficient laptop. Can't... stop... benching...

Thanks for running the benches. I'd be curious how DOTA 2 runs if you wanna give that a whirl. Keep up the interesting work.

 

[SlowSpyder]

Sorry, what is DOTA 2?

Stock scored 151 in Cinebench 15. I'll get power use numbers later, too.

 

[TrulyUncouth]

It's a free to play game on Steam, made by Valve. It is the most popular game by far on their entire service. Right now a >6 million dollar tournament is going on for the game. If you wanted to try it to benchmark it only takes a few minutes to make an account and the download for the game is ~4GB. If you don't want to mess with it, don't worry- was just a suggestion.

 

[SlowSpyder]

No problem, I'll take a look this weekend and give it a go at factory/oc settings.

 

[tential]

Surprising you haven't heard of Dota lol
Only one of the biggest games to have come out I'd say alongside Counter Strike and now League of Legends.

Steam stats for the day....

Top games by current player count
Current Players Peak Today Game

795,648 796,342 Dota 2
114,278 138,380 Counter-Strike: Global Offensive
56,392 57,277 Team Fortress 2
37,170 56,395 Football Manager 2014
28,280 36,257 Sid Meier's Civilization V

 

[SlowSpyder]

Maybe that's why... I almost never play first person shooters. I rarely wander away from the RPG / strategy genres these days.

*edit - I guess it isn't a shooter, just looked at some screen shots. Maybe I'll have a new game to play, thanks.

 

[SlowSpyder]

It's a free to play game on Steam, made by Valve. It is the most popular game by far on their entire service. Right now a >6 million dollar tournament is going on for the game. If you wanted to try it to benchmark it only takes a few minutes to make an account and the download for the game is ~4GB. If you don't want to mess with it, don't worry- was just a suggestion.

Downloaded, am messing around with it. I don't see a benchmark, but haven't created an account. I ran the tutorial with 1.7GHz/2.8GHz settings, it seemed ok to me. I had FRAPS running, it dipped to 22FPS at one part, but mostly stayed 28-30FPS, highest I saw was 32FPS. I'll try some different clockspeeds, not sure if I could be GPU limited here. Again, curious to see if the A8-3500m graphics makes a big difference... if only fast laptop ram was cheap.

Helping a buddy work on his car, but I'll try and get you the benchmarks when I have more time to look later.