M17x Performance - GTX 280M SLI vs. Mobility 4870 CF

I'd like to share my experience with these video cards in M17X-R1.
Had 280s for almost 5 months and yesterday got them replaced with 4870s.
I don't bench aggressively and only occasionally game so take this with a grain of salt since I compare them NOT OC'ed, just stock, anyway:

Performance -

There's been a lot of testing and benching of both GPU types but what I found is that both have pretty much equal performance in games, I'd give ATI cards a little bonus for better drivers (only tried 3 versions and all worked great, but the best performance is possible so far with the latest Catalyst 9.12 + Mobility Modder), whereas with Nvidia cards tried all of the available drivers and was having issues with ALL of them (the most stable and performing were the 186.82 drivers but got higher MCP/CPU temps every time I tried them).
My stock GPU Vantage score with Nvidia GPUs was around 8k while the ATIs pushed it straight to 11k. (again, no OC'ing and no tweaking).
I only compared 2 titles so far (Dragon Age and Kings Bounty The Armored Princess). Both GPU types showed similar FPS, however the game play was more smooth with ATIs (probably due to the fact that Nvidia cards contributed to stuttering). Again, there's a lot of benching info on this forum and I'm not going to enter that section here, this is just a SUBJECTIVE IMPRESSION of a single person.



Powermizer vs PowerPlay

Nvidia clearly loses this one IMHO, since the cards seem to always dynamically clock themselves to the wrong direction with all of the latest drivers (I had one of the GPUs being stuck at max clocks till reboot with 195.6X drivers, both cards up-clocking without need to their max clocks during simple web-browsing using the 195.8X drivers etc.), not to mention that every change of clocks on Nvidia GPUs resulted in high DPC spikes and stuttering.
Some people tried disabling the PowerMizer but this solution creates unnecessary constant stress on GPUs/fans which I'm not ok with, especially when doing my work on the machine and not gaming.
PowerPlay on the other hand works flawlessly so far; the cards are only up-clocked when gaming/benching and quickly brought back when the load subsides.


Temperatures!

Here is a strange and tricky part...
Nvidia cards seem to run less warm than ATIs when idle (280s ~ 40c, 4870 ~ 50c), but when @load both GPU types equalize around 80c. However, and this is, again just my assumption, it looks like Nvidia cards are calibrated to show lower temps than the actual ones while the ATI cards show higher temps than the real ones. to come to this conclusion I made a simple test - ran Dragon Age . I know exactly (spent many hours on it and Crysis) how hot your keyboard (QWER keys especially) gets when the 280s are warmed up to 75c+,...it gets hot! And guess what, when ATI GPUs hit 80c and hovered around 75-78c for quite some time, the keyboard remained cool. I started paying more attention to the issue and found another strange thing - With 280s the fans kick to 60% at around 60c and to almost a full blast around 70-75c, while with 4870s they kick to 60% around 70c and to a full blast near 80c. This is yet to be investigated, but to me it seems that probably the fans should kick in at the same temp regardless of the GPU type, they should have independent sensors, and if it's true, then one of the two (maybe both) are calibrated to the wrong temps and thus we get different readings.


Noise

Both are noisy, but Nvidia cards are clearly more loud, they have many more melodies to irritate/satisfy your ears


DPC Latency and Stuttering

So why do the ATI cards when put into Nvidia motherboards reduce the latency and stuttering??? Many have been asking that question here and @Dell tech support, but we didn't get the answer (just some strange mumbling about some driver problems, BIOS issues etc.)
I did my own digging on the subject. And again, this is strictly subjective and correct me if I'm wrong but...:
It seems that Nvidia cards rely more on software acceleration rather than the hardware one, thus creating a lot of stress on MCP/chipset and resulting in much more stuttering (now add to this the current consumption of 150Watt for two cards and happy lagging!)
ATI GPUs on the other hand rely more on the hardware acceleration putting less pressure on the mobo and avoiding the stuttering. To support this thought also found that with ATI cards my MCP and CPU temps are always lower by 3-4c, and during gaming they are lower by up to 10c!!! Also the ATI cards cool down WAY faster than the Nvidia ones.
Another thing is that Dell stopped offering the Nvidia cards shortly after the issue was discovered. This might be for a totally different reason (political) but it looks like the real deal...

Anyway, I really liked the Nvidia cards, but after switching to the RED team, can only be happier
Hope this gives some mind food for you guys&gals.

 

Interesting observations. Thanks for putting this together. Going to leave this in the main AW forum as its relevant to issues at hand.

 

Wow! Very nice post. A +++

 

good eye

finally some die hard nvidia users seeing that well the green camp isn't so green

 

Nice observations Aikimox, I would like to add two comments about the powermizer and the latency. This is an article that I read in AnandTech concerning the Power Management. http://it.anandtech.com/IT/showdoc.aspx?i=3722&p=2 According to the article:

"The most important information can be found in the ACPI tables, which describe the capabilities of the different devices of the platform. Once the kernel has read and interpreted them, the role of the BIOS is over. "

"The boss of the ACPI based power management is the power management component of the kernel. The kernel power manager handles the devices' power policy, calculates and commands the required processor power state transitions, and so on. Of course, a kernel component does not have to know every specific detail of each different device."

In other words a poorly written BIOS affects the way the OS handles the devices like GPUs in our case. The powermizer would work more efficiently if our BIOS could provide more accurate information to the OS. The result is the latency that occurs from the ACPI and causes the stuttering we all know. Poorly written drivers play their role too.

I hope I helped a little to make things more clear.

 

Good info!
But since PowerMizer is the equivalent of PowerPLay in what it does why doesn't PowerPlay result in latency issues when the cards change their clocks? The ATI cards were working fine even with the A02 BIOS.
I think we all know that the real culprit of the problem is the weak mobo and ATI GPUs can only minimize stuttering by stressing it less than Nvidia cards.
It was confirmed to me by phone by the agent who spoke to someone from the engineering team, that the issue is related to Nvidia strictly...I now can guess he meant the Nvidia chipset, not GPUs.
Anyway, I'm not trying to find the latency cause here, but to compare the GPUs in this specific system.

 

Thank you for this information, I'm still trying to get the r2 but if that doesn't happen and that looks likely i will receive the same set up you have.

previously with the 260sli so very intrested and pleased to read your post

 

Thanks for the review, Aikimox. I especially appreciated your diligence to retain objectivity, and of course, your eye for detail. Well done.

Now regarding your theory on extra CPU load, I too would like to add a comment:
I believe this is linked to the PhysX drivers that go with the Nvidia drivers. PhysX is actually not all GPU, but also CPU. The setup allows Nvidia cards to offload CPU-specific tasks-which might be better done on CPU-to the CPU while managing the rest on the GPU hardware. You can have a look on the net to verify this, but that is the deal on PhysX.

 

well nvidia has 4 different clocks.
xxx
2d clocks
3d clocks
extra

this is alittle tricky and the driver has to be very optimized o do it right.
2d for battery very light gaming like mario or sonic.
thorttle for something
3d for watchng movies etc
extra for gamnig video encoding. etc

where as ati has only 2 clocks
2d and 3d.
basically high and low.
so gaming is 3d which is high. and when it's not gaming it just goes to 2d which is low.


a couple months ago when my 9800m gtx i got the idea from sunrise where i change al my clocks (2d, throttle, and 3d) all to the same clocks 275/550/301
so when i am watching videos doing something light. the video wont stutter because of low clocks. and i left extra(the gaming one) well for gaming. to make it easier for the card/powermizer to change.

of course it still thought it was going from 2d/3d/throt. but the clocks where the same.

 

Sounds like a similar idea to the one I had when I was looking at Nhibitor.
Didn't really want to tamper with the Card VBIOS though in the end (I was going to either delete the other clocks, or just set them all to maximum clocks, as thats basically what Im doing now anyway)

This thread does remind me why I loved my 4850 CF equipped M17 though. No stuttering, no massive driver faffing to find a decent set, just mob-mod the latest, and it really was not that much slower than the GTX280ms at all at stock speeds. The ATI cards gave massive love when I upgraded to the X9100 as well.

 

Thanks for the feedback guys!
Regarding Physx :
This might be the reason for the additional CPU/MCP load, could you try disabling it and comparing the temps. Don't remember any difference when I did with 280s but maybe I wasn't paying too much attention.

As for the clocks, it's a very good idea to make just two levels for the cards.
Also for the records: when Nvidia GPUs change their clocks the latency spikes are often 60k+, while ATI cards only bring 4k ones (those do not contribute to audio/video/gaming stuttering).

 

First off, afraid you are wrong Don about the number of clock settings, the ATI's actually have more lol


Secondly, performance differences! Here is the stock scores for GTX 280m SLI and 4870 CF that I got for 3DMark06 and 3DMark Vantage

GTX280m SLI:
3Dmark06: 13488
3DMark Vantage: P10416

4870 CF:
3DMark06: 14521
3DMark Vantage: P11000

So as you can see, at stock, the 4870s are a huge boost in synthetics at least

 

Scook, do you still see huge latency spikes ?
I have a few around 5k (rarely), and yesterday for the whole day of testing got 1 or 2 37k ones.

 

I left it running while doing a lot of stuff and at idle, all low green

During regular system use I think my MAX was like 4500 which is NOT BAD compared to what I had lol

 

1 or 2 37k the whole day is more than perfect

On my M17x R1 I used to have 65k lol

 

You make a point. But even so, CUDA processors generally put more burden on the CPU when compared to Stream Processors. I do know that Mirror's Edge runs about 6 degrees less on CPU when I disable PhysX. But you know, I don't mind at all, because CPU temps have been great. It's just that CUDA and PhysX do rely more on CPU than some would believe.

If you're interested:
http://www.pcper.com/article.php?aid=745&type=expert

It's a bit of a long read, but if one were to skim through the piece, one would find that the CUDA setup puts more stress on CPU when compared to ATi Stream.

 

Well I game a lot - my steam score was 10 with 140 hours over a two week span in December (best 3 weeks off ever!). I really love my Nvidia based system and would prefer a software fix.

With a clean Vista x64 install (respawn) at the start of the December and the latest verde driver I had zero problems in any of the 30 games I bought during Steam sales.

It wasn't until I dusted off my Win7 upgrade and actually used it that I experienced noticeable stutter problems.

I have a hard time believing some of the crazy stuff like temps and reverse powermizer. You are doing something wrong or changing a registry setting somewhere because software does not just spontaneously write new instructions. I was like one of those crazy Korean Starcraft players for three weeks straight and experienced no problems with my M17x - the cooling in particular was god like.

The 4870 fix seems dubious to me, I am not going to read through 1500 posts to see if any AMD users suffered from stutters. I see no point in bothering for a 10% benchmark gain unless Dell says it is a 100% certain fix.

I have never read anyone else say that AMD's mobile driver program is good. You are the first one. These cards have their own well-known stuttering issues too (google microstutter).

 

Lewdvig, I am kinda' in the same boast as you - just looking for real answers, and critically analysing what is already present.

Issue with microstutter. It is present in both SLi and Crossfire solutions; it's an inherent issue with dual GPU solutions, but hardly shows when frame-rates are stable. This is why most people don't notice the microstutter, but all Xfire and SLi rigs are prone this this problem if bandwidth or performance are hampered at any point.

 

@Lewdvig

He's not actually commenting on thier mobile driver programme, which you're right, isn't great, hes using mobility modded desktop drivers, which generally work perfectly.
ATI have also recently gone on record as trying to get mobility support into the standard drivers, however it depends on support from the laptop manufacturers, indeed in the recently leaked 10.1 drivers for desktop cards, some mobility support is ACTUALLY INCLUDED.

ATI drivers in recent years have been getting better and better, whilst as can be seen from the amount of general anger towards the 19x. series drivers from Nvidia [and to a lesser degree the 18x drivers], the same cannot be said there, again something I have even seen coming from previously-known Nvidia fanboys on other forums.

Also regards microstutters, it was a well noted fact for desktops as well as laptops that the 4 series made a massive improvement in the field of microstutter for owners, I've read multiple reports from 3 series crossfire owners reporting that microstutter virtually dissapeared when going to the 4 series, my experience with the 4850 mobile crossfires was also slightly more fluid than mine with the GTX280m setup, however I try not to pay attention to it. Neither side are perfect however, it is an intrinsic problem with dual-GPU. Nvidia's SLI setups also apparently suffer less in the GT2** based revisions (and likely Fermi), however as the mobile cards are still G92 based at the top end, we have not seen further hardware optimisation for this yet, whereas ATI cards deriving from newer designs and silicon have an advantage in attempting to minimise the appearance of said stutters.

Suggestions like reverse powermizer are not crazy assumptions either, for many people the DPC latency spikes can be directly, and obviously traced to when the cards change clock speeds, for some people disabling powermizer fixes the issue (or massively reduces it), whilst other people do not note an improvement, partially as the powermizer functionality is very well ingrained into the drivers, and in many cases whilst one card will stop clock speed changes, the other continues.

Disabling powermizer is also not crazy on registry terms, as you actually need to write registry keys to do so, these are keys that used to be installed with the Nvidia drivers in many cases, and still are in some cases. Whilst a lot of drivers no longer create these entries on installation and use default functionality, they will still accept instructions from the registry keys if present.

Cooling is not generally an issue on the M17x or M980NU, however subsets of both laptop's users both suffer the same issue, whilst using the same graphics cards and chipsets, suggesting a related or shared root problem/cause.

Regards ATI cards, then again this is quite possible a fix in this case. For owners suffering issues with powermizer related spikes, UNINSTALLING the Nvidia GPU drivers COMPLETELY removes the issue...no more spikes, no more stutters, but obviously we're then running standard VGA, in that sense the fault is directly traced to Nvidia, either due to interaction with the system/BIOS, or issues with either the VGA drivers and associated functionality/VBIOS functionality.

Sorry if this seemed like a rant, it was not intended as such!

 

the ati cards also reduce the need or use of the 9400. reduceing its work load reduces stutter

 

Awesome post. Not a rant at all.

I used a Powermizer registry tool that I got from this forum to test the Powermizer on/off toggle and it seemed to help a lot - but did not solve the problem completely.

What I don't understand is how pressing the eject button causes a spike/stutter if the problem is video card related. Likewise, why does typing into a text box cause a spike?

I think most of the enthusiasm for the 'cure' revolves around people getting what they think is a better video card (a 10% improvement in return for hours on the phone and a tech coming to your house for who knows how long) or the new revision (also not likely).

 

do you get stutters when you press eject?

 

Nice post, Alexrose.

Moo,
Wait. How does it reduce the -need- for 9400m? Reducing work load reduces stutter?

And I second Lawdvig's query about the eject button. Do Radeon users get stutter when ejecting the tray?

 

the 4870 is a better card. Your in denile sorry to be the one to tell you ....

Lets count overclocked and stock ... overclocked it scores 14k 3thousand more .... even stock it does


ITS A CURE BECAUSE IT CUASES THE SYSTEM TO RELY LESS ON THE 9400M BECAUSE OF THE BIOS MODIFICATIONS

In order to get the 4870's o work on the chipset they had to do some drastic modifications to the bios regarding the chipset. I am betting that when you use 4870's it communicates with the chipset differently to not have it crap out issues due to nvidias and ATI's love for each other.

all IMO but it seems to prove its self

 

i'm confirming that i have zero issues with my nvidia too even 9400 enabled. i really like hybrid system and it's very useful for me. that's why i don't need R2 for that moment

agreed i don't need this 10% and they haven't any noticable gain for me

 

"TS A CURE BECAUSE IT CUASES THE SYSTEM TO RELY LESS ON THE 9400M BECAUSE OF THE BIOS MODIFICATIONS" - The_MooTM.

Yeah, I still don't understand how a BIOS modification removes the purpose of a low-end passive 9400.

 

As I discovered myself, the powermizer switch programme is really only suitable for single card usage, multiple users with SLI setups, including myself, found that only ONE card was disabled, one of the cards still clocked down.

Its only recently with a combination of powermizer registry tweaks and the recently included Power Management option in the NVCP that I've been able to really restrict powermizer on my machine, which has massively helped my issues (note not fixed).

Regards the disc tray, I believe someone commented that with the Nvidia cards they got DPC spikes of 16k ejecting the disc, whilst with ATI cards they're down to 4k. Now this may just be the way the system handles drive interrupts, which the Nvidia setup reacted worse to.
Hell, I remember years ago when ejecting a disc would freeze the majority of systems up completely for a few systems, due to the way the interrupt was handled and the IDE bus was used.

 

I do not know what they did honestly but they had to modify the bios to let the 9400m use the 4870's.

What ever they did help and the fact ATI's driver are not massive turds makes it better

If you cannot see the ATI cards are better then you need to keep reading ....

 

Btw, the chipset is the MCP79 SLI...aka the 730i. 9400 chipset gives the wrong impression

Now I could be wrong here but do you have to restart to switch between the 9300/280s? If not, its quite possible that whilst the Nvidia cards are present, the chipset and GPU are both enabled, just the integrated GPU is suspended. Whereas with the ATI cards, whilst they're active the 9300GPU is completely disabled at the BIOS level, which is why I restart is required to switch between the IGP and dedicated cards (even though Win7 can accept both sets of drivers at once etc).
This would explain the slightly lower load/temperatures, as you have the contrast between suspended and completely disabled.

Also regards the BIOS, this is likely true. AW/Dell would likely have had to have the BIOS optimised/tweaked to allow ATI crossfire on the board, as generally Nvidia chipsets do not allow crossfire implementation.

 

When I pressed the eject key with 280s got 16k spikes (that is audible and visible) while with ATIs I get below 4k! It's not considered a stutter, as it will not make your games lag/freeze and will not be audible during audio/video playback.
This is a total fix IMHO.
The only thing I'm afraid of is that the entire mobo/chipset line will be declared defective soon.... I need that battery life for work and if it's taken I'll have to look for a different rig, probably not even Dell....

 

I'm not in denial. And I won't be drawn into a fanboy polemic either. I don't care about synthetic benchmarks, looking at my FRAPS recordings in games I have yet to see any 4850xf scores that top mine (my rig is not OCed or modded in anyway).

http://notebookcritic.com/2009/12/16/review-alienware-m17x-all-powerful/

batman - start of the game, after cinematic first fight after Joker gets away

warhead - start of game, after bridge as you come up to the shore (kill all three patrols, jump in APV and drive to radio tower, kill guard detachment)

fc2 - firefight after jackal visit, run to river and jump in

dragon age - lothering highwaymen

l4d2 - hotel, top floor

re5 - first fight after execution until ally blows up gate

I don't buy the BIOS explanation. Maybe there is less load because of the IGP being turned off.

 

they won't take it ... they will give you a option but they cannot just take it

 

I am pretty sure that SLI is more dependent on the chipset as the data that goes between the cards gets processed by the cpu in between them where with crossfire the 2 GPUs can directly communicate cutting out the middle man of the CPU

 

Well that's a bit more information for use. I'd like to know more about this, which is why I was asking. Perhaps you were seeing false intentions.

. . .good for you?

I can read just fine, and recent posts and threads have only proven this point over and over again to the stage that it is becoming a joke. People here are not not children looking to play brand-loyalty games. What is good, is good. So when someone asks a question, perhaps they are genuinely curious about a certain feature. Perhaps you should read the words more than reading into the words; avoid interpreting peoples intentions, because it is no ones strong suit, and certainly not yours.

 

what's your stake in this? why are you getting hot? this has nothing to do with AMD vs NV. It's 100% about fixing the stutter problem.

I am glad you are happy with your AMD cards.

 

i dont have ATI cards

I had nvidia single 260. I NEVER had a stutter .. Its only SLI that has that issue.

I have tested my machine as well ...

I am not mad, just really confused why some people dont think this is a fix just puzzles me.

 

Interesting. You might be one to something here. But were there single card NV users that had problems?

 

Aye, there is typically an nvidia chip on the board (I forget the name, and it's not ACTUALLY technically necessary as has been proven before), however it would be incorrect to say the ATI cards completely ignore the CPU, or you wouldnt get a tidy GPU power upgrade in crossfire with a faster CPU (and you do).

Also, I believe there have been single card users with issues, especially in regards to powermizer. Both on this system and others.

 

i was just answering your question ?

 

Xfire benefits well from a strong CPU, even moreso than Nvidia cards, although its a mild difference. This is a GOOD thing for ATi, in my opinion.

 

Of course they do not ignore the CPU, but they rely on it less to perform crossfire is all I was implying

Of course the CPU still feeds the data to the GPUs so that they may process it and render the 3d images

 

No, even if that was true ... why would they declare the Chipset/mobo defective ?

They would declare the CPU defective .... its the 9400m and how it handles it and the drivers that never help

 

my point was the data has to go through the northbridge/chipset on its way to and from the CPU.....where with the crossfire it goes from PCI-e to PCI-e bus

I know to enable crossfire on an nvidia chipset you have to enable a special BIOS level ability for the PCIe busses to talk to eachother, let me try to find the article where I found that

Read paragraph 1 of this page:
http://www.anandtech.com/systems/showdoc.aspx?i=3094&p=3

 

Right, so it isn't the video cards (although that seems to clear the problem up for some people).

You know, none of this is official so its all just speculation and BS until John updates his post.

As far as this thread and 4870 v 280m, we have about a 10% or higher performance boost in benchmarks in favor of the 4870s and no stutter.

 

Awesome find.

 

it clears up the problem because the way the GPU's use the Chipset changes with the bios modifications they did.

............

 

Ya I found it interesting the differences between the 2 multi gpu technologies. This article is a good one to keep handy for these discussions lol

 

nVidia still doesn't have the license to make chipsets for the i5/i7 architecture, right?

 

Nice find m8! Explains our wild guessing.

 

I would assume with all the heat they've had with intel, probably not.