Jump to content

OFFICIAL: M18x Benchmarking Thread


StamatisX

Recommended Posts

Clevo cut back on voltage regulation with their haswell laptops (CPU regulation cut from 3 phase to 2). Maybe AW did the same? AW did cut the battery from 12 cell to 8 right? The new AW18 looks inferior to the old m18x R2 to me.

Also could you tell more about your fan replacement?

I'm not sure if they did that or not. I should look. Is the way you can tell by counting MOSFET (2 MOSFET instead of 3?) or some other way to tell? If they did, that would be very stupid. I wonder Clevo did that?

The fan replacement is really simple. I recently discovered there was an 11.7 CFM CPU fan available to replace my original wimpy 4 CFM so I got one. It was a direct replacement with any kind of mods.

It helped a lot (roughly 5°C temperature decrease under overclocked load). Although it moves a bunch more air, I don't notice any difference in noise.

16nRywHl.jpg

Link to comment
Share on other sites

I'm not sure if they did that or not. I should look. Is the way you can tell by counting MOSFET (2 MOSFET instead of 3?) or some other way to tell? If they did, that would be very stupid. I wonder Clevo did that?

The fan replacement is really simple. I recently discovered there was an 11.7 CFM CPU fan available to replace my original wimpy 4 CFM so I got one. It was a direct replacement with any kind of mods.

It helped a lot (roughly 5°C temperature decrease under overclocked load). Although it moves a bunch more air, I don't notice any difference in noise.

16nRywHl.jpg

For overall power usage capability the limit is going to be a few MOSFETs near where power comes into the laptop. These FETs allow switching between plugged in and battery power. You should be able to add a resistor to the current sensing circuit to raise your power limit. For example if a .01Ohm resistor is used to sense current, adding an additional .01Ohm resistor in parallel will double your power limit since the current sensing chip will only read half the voltage drop and thus think current is half of what it really is.

As for CPU voltage regulation you can just check the number of big inductors near the CPU socket. The number of inductors you have is the number of phases. Fewer phases is almost always bad since although you can have more FETs per phase to make up for a phase loss, there will be fewer gate drivers to control the FETs to get quick switching times for accurate regulation. (the way regulation works is you have a set of pull-up and pull-down FETs. These FETs turn on and off rapidly to create your CPU voltage. If voltage is low the pull-up FETs turn on. If it is high the pull-down FETs turn on)

Link to comment
Share on other sites

Well... one nice thing about Haswell... with winter almost upon us, it probably makes a really great windshield defroster/de-icer.

Ivy Bridge vs. Haswell - Back-to-Back Comparison, 3 feet apart in the same room, same ambient temps.

CPU Clock Speed: 4.3GHz
GPU Clock Speeds: 975/1400 (+125/+300)
Display Driver: GeForce 331.65 WHQL
Operating System: Windows 7 Ultimate x64
3DMark11 Results Detailed Comparison: Results - Alienware 18 versus Alienware M18xR2
Alienware 18 - 4930MX | GTX 780M SLI -CPU Max Temp: 105°C Alienware M18xR2 - 3920XM | GTX 780M SLI - CPU Max Temp: 94°C
Link to comment
Share on other sites

@Mr. Fox HUGE memory frequency advantage on the IVB system. That explains the physics/combined performance loss, but I don't know what is causing the big GPU performance loss.

And you have 4 780ms... makes me jealous.

I'm sure it helps, but it's not going to put that great a distance between the Physics scores. Haswell is just a sucky product for laptops or desktops. At lower clocks it is better than Ivy Bridge, but it does not handle extreme overclocking in most cases... unless you get really lucky with an unusually good CPU and have an extreme cooling solution.

I think the extra power draw from Haswell may be adversely affecting GPU performance, but yeah... a big disparity there on the GPU side as well as CPU. It takes a ton of power to prevent Haswell from clocking down under load.

I think this video sums it up pretty well...

Link to comment
Share on other sites

I'm sure it helps, but it's not going to put that great a distance between the Physics scores. Haswell is just a sucky product for laptops or desktops. At lower clocks it is better than Ivy Bridge, but it does not handle extreme overclocking in most cases... unless you get really lucky with an unusually good CPU and have an extreme cooling solution.

I think the extra power draw from Haswell may be adversely affecting GPU performance, but yeah... a big disparity there on the GPU side as well as CPU. It takes a ton of power to prevent Haswell from clocking down under load.

I think this video sums it up pretty well...

Yeah Haswell is not worth it. I'm just trying to find a cheap E1 3920xm/3940xm on ebay

Link to comment
Share on other sites

  • 2 weeks later...

did some winter benchmark in front of the cold swiss mountain air ;-))) cpu @4.2ghz with xtu and throttlestop active for the temp spikes, gpus +180/+320 still standard voltage. was tired so i stopped, will try further next time with raised voltage :) ok for today hehe

NVIDIA GeForce GTX 780M video card benchmark result - Intel Core i7-4930MX,Alienware 01W2J2

  • Thumbs Up 2
Link to comment
Share on other sites

  • 2 weeks later...

Since I`m snatching one of SVL`s Vbios, its only fair I share my results here. This is my Alienware 18, with 4800MQ and GTX 770M SLI on stock Vbios. Aka Dell Vbios.

Captain Cloudfire`s overclock log, November 20th 2013

Attempt #1:

Stock clocks.

NVIDIA GeForce GTX 770M video card benchmark result - Intel Core i7-4800MQ,Alienware 01W2J2

Attempt #2:

+50/+400 OC

NVIDIA GeForce GTX 770M video card benchmark result - Intel Core i7-4800MQ,Alienware 01W2J2

Attempt #3:

+100/+400 OC

NVIDIA GeForce GTX 770M video card benchmark result - Intel Core i7-4800MQ,Alienware 01W2J2

Attempt #4:

+135/+700 OC:

Failed. Froze in test 4. Had to reboot on black screen.

Attempt #5:

+135/+600 OC

Failed. Froze in test 4. Had to reboot on black screen.

Attempt #6:

+135/+500 OC.

Failed. Froze in test 6 (the last one). Had to reboot on a black screen. It must be the memory clock.

Attempt #7:

+135/+450 OC.

Failed. Again. Sigh. %¤#"!¤"!. In test 6. Had to reboot on a black screen. Could it be the core clock?

Attempt #8:

+135/+400 OC.

Fail fail fail fail. Froze in test 1 this time (?!!1?). Had to reboot on a black screen. Same memory OC as the first 3 tests that worked. Fishy. Very fishy.

Attempt #9:

Not shure where to go next to be honest. hmmmm

+120/+400 OC

May the Devine Cow God be gracious upon this machine. As well as looking out for the sanity of the guy trying to recover from all the failed attempts.

HAPPY DANCE ^_^

NVIDIA GeForce GTX 770M video card benchmark result - Intel Core i7-4800MQ,Alienware 01W2J2

Attempt #10:

Higher memory clocks, same core clock. For science!

+120/+600 OC

Failed. Looks like science thought me there are several blockages here.

Attempt #11:

+135/+400 OC

Nope, still doesnt work.

And that is enough of today`s adventures.

Final thoughts on GTX 770M SLI

GTX 770M apparantly do not have the same high voltage memory you find on GTX 780M. It have low voltage memory like you find from the GTX 600M series.

Meaning around 1200MHz is max memory clock.

Core clock goes to max 917MHz (ish) on stock vbios. It could go higher with a better vbios but only way to find out is trying a modded vbios. Perhaps later.

Graphic score went from 9591 (stock) to 10822 (Max OC on stock vbios).

With max OC the GPU temp maxed out at 70C. So if I use a good vbios I know I can squeeze out a lot more out of these GPUs.

Stock clock results

post-20082-14494996459781_thumb.jpg

post-20082-1449499646007_thumb.jpg

Max Overclock results

post-20082-14494996460225_thumb.jpg

post-20082-14494996460499_thumb.jpg

Link to comment
Share on other sites

@Cloudfire. That's interesting, looks like you've worked out your maximum overclock at stock voltage quite sucessfully. Your 770M's overcock pretty much like my 670MX, which I guess isn't that surprising considering they have an almost identical layout apart from the stock frequencies. You managed 100Mhz more on your memory than me, but I managed to get a little more on the core, 928Mhz on the core. Is your core running at 0.925V? (That would be the same as my 670MX). Should handle more voltage quite easily if your cooling is good? Ah, just seen that your temps are 70 degC, that's not bad, looks like it will handle some more voltage & OC quite well.

Link to comment
Share on other sites

I'm sure it helps, but it's not going to put that great a distance between the Physics scores. Haswell is just a sucky product for laptops or desktops. At lower clocks it is better than Ivy Bridge, but it does not handle extreme overclocking in most cases... unless you get really lucky with an unusually good CPU and have an extreme cooling solution.

I think the extra power draw from Haswell may be adversely affecting GPU performance, but yeah... a big disparity there on the GPU side as well as CPU. It takes a ton of power to prevent Haswell from clocking down under load.

I think this video sums it up pretty well...

Off the topic mate, but how did you get your Alienware M18XR2 overclocked to 4.9GHz?

Link to comment
Share on other sites

I benched this some time ago. Haven't been doing much of it lately, until Dell fixes the BIOS. The 4930MX is hot, which makes it somewhat hard to work with. I have tried repasting many times, perhaps now I need to consider lapping the copper heatplate on the CPU. Or I have to get some 2133MHz 1.35V SODIMM. Those are rather expensive though.

post-12798-14494996472015_thumb.jpg

NVIDIA GeForce GTX 780M video card benchmark result - Intel Core i7-4930MX,Alienware unknown

Link to comment
Share on other sites

Well I had a go today with my GTX 770Ms.

This time with the famous SVL modified Vbios. Thanks for making it for me. This is probably not the highest performance I can get, since I did not dare go any higher than +0.05V (1.05V) on my GPUs.

#1: Stock clocks on stock Vbios.

3DMark 11

post-20082-14494996483611_thumb.jpg

post-20082-14494996484047_thumb.jpg

Unigine Heaven

post-20082-14494996484361_thumb.jpg

Unigine Valley

post-20082-14494996484619_thumb.jpg

When looking at the pictures, you will notice that the 770M actually go up to 863MHz thanks to GPU Boost 2.0.

#2: Max overclock with stock Vbios

I previously thought it only ran at 915MHz but its actually almost 1000Mhz thanks to GPU Boost.

3DMark11

post-20082-14494996484889_thumb.jpg

post-20082-14494996485195_thumb.jpg

Unigine Heaven

post-20082-14494996485528_thumb.jpg

Unigine Valley

post-20082-14494996485824_thumb.jpg

This was the highest I could go. Tried any higher core clock or memory clock and the screen would black out. So I contacted SVL and asked if he could modify a new vbios for me. He agreed, and I decided to test it out to see if its any good to have anyone modify vbios for you.

#3. SVL Vbios

This is how GPU-Z report his vbios. Compared to the stock vbios, you now have GPU boost disabled which you can see from the 863MHz clock. It also have a function that monitor the TDP plus the voltage is 1.00V (0.963V originally).

post-20082-14494996486089_thumb.jpg

My first try was to figure out if it could go higher than the 980MHz block you have on the stock vbios. So I overclocked the GPU to 1000MHz which it ran without any hiccups. GPU-Z doesn`t report the correct clock since Valley and other benchmarks report it as 1000MHz.

post-20082-14494996486322_thumb.jpg

Worked out perfectly. :thumbsup:

This was looking promising.

Time to turn the clock up. 1050/1200 next.

3DMark11

post-20082-14494996486626_thumb.jpg

post-20082-1449499648694_thumb.jpg

Unigine Heaven

post-20082-14494996487309_thumb.jpg

post-20082-14494996487587_thumb.jpg

#4: Max overclock with SVL Vbios

Here is the max overclock I managed with SVL vbios. Note that this involved overvolting the GPUs from 1.00V up to 1.05V. The highest clocks I could do with my GTX 770Ms was 1100MHz on the GPU and 1200MHz on the VRAM.

3DMark11

post-20082-14494996486626_thumb.jpg

post-20082-14494996487856_thumb.jpg

Unigine Heaven

post-20082-14494996488184_thumb.jpg

Unigine Valley

post-20082-14494996488466_thumb.jpg

Conclusion:

New vbios from SVL worked out pretty well. I could increase the clocks additional +120MHz thanks to unlocked voltage, not to mention that the overclock slider is unlocked by SVL. Without that, you are stuck on 1000MHz anyway.

I could have probably gotten some more out of it considering the GPUs only maxed out around 78C but I did not dare increase the voltage more since the screen blacked out higher than 1100MHz. Maybe later.

3DMark11 Performance:

Stock: 9605

Max OC: 11740

22.2% better performance

3DMark11 Extreme:

Stock: 3080

Max OC: 3829

24.3% better performance

Unigine Heaven:

Stock: 32.0FPS

Max OC: 41FPS

28% better performance

Unigine Valley:

Stock: 40.1 FPS

Max OC: 51.8FPS

29.2% better performance

  • Thumbs Up 1
Link to comment
Share on other sites

@Cloudfire, you got some good results there! I wouldn't go any higher than the 1.05V either, even if your temperatures were lower than they are. You said in your post that GPUz doesn't report the correct GPU core frequency, actually it is GPUz that is reporting the correct frequency, and the other software that's not reporting the correct frequency. When you set your core at 1000Mhz you got 993Mhz reported in the Sensor Graphs in GPUz, the 993Mhz is the actual frequency they were running at. The GPU frequencies go up in 'steps', I think they are 13Mhz steps, so if you set 1000Mhz, it will just set the GPU to the next lowest step (which happens to be the 993Mhz). But, that's not really that important, you're getting some good results there, looks like you've reached the limit!

Link to comment
Share on other sites

Oh I see. Thanks, I thought Valley was pretty reliable. Good to know :)

I`m pretty satisfied. One thing is breaking the GHz barrier. Another one is increasing it by a further 100MHz. 1100MHz is pretty high clocks no matter what GPU you are looking at.

I was just comparing against desktop GPUs and it seems that my SLI notebook is somewhere between GTX 680 and GTX 780 in performance. Roughly a little bit faster than R9 280X. Not too bad for a notebook that only runs 1080p ey?

post-20082-14494996499982_thumb.png

Also during my testing I found out that your 1100MHz clock on the VRAM is more than enough for these GPUs. With the GPU running @ 1100MHz in the Valley test, with everything at Ultra and even 8xAA, I only taxed the memory controller by 73%. So if I find a game I can`t play on stock clocks, there is very little use pushing the memory to 1200MHz.

:)

Link to comment
Share on other sites

Yes, the GHz barrier that you broke is quite impressive, kind of like desktop style frequencies! Yep, you're right about saying your performance is between a 770 & a 780, at least in benchmarks, in gaming I would think it would be there or thereabouts too - that's a lot of performance for a notebook!

So you saw that the core overclocking was more useful than the memory overclocking, that's what I've found too. You could run some game benchmarks at 1100Mhz vs 1200Mhz on the VRAM and see if there's any difference. If there's no difference, then you may as well run it at 1100Mhz I suppose, that way it's kinder on the GPU.

Link to comment
Share on other sites

  • 2 weeks later...

Good scores Mr Fox, although lower on the GPU than last time, but I noticed you've taken the CPU higher. 1.5V on the CPU! Don't Intel Chips start to get damaged around about 1.4V? (I've not looked into it for a while, but I've got 1.4V somewhere in my mind from a Guru3D review a while back).

Link to comment
Share on other sites

Nope, I make the power available and the CPU decides how much it needs to use. Heat kills and too much static voltage can, too. But, that's not too much for 47x4 or 48x4... There's more available and the CPU is choosing not to use it. Have to be careful about believing some of the stuff you read. Sometimes it's not true. Same settings worked nice for the 2920XM as well.

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.