580M SLI throttling Discussion Thread

The throttling at stock clocks is just a side effect of aggressive BIOS triggers. So Dell will fix this, its not something unknown to them or something that's so bizzare. They released the A02 BIOS after all. Case of fine tuning.

So you all should relax, they heard us on the OSD issue and fixed it. Dell is doing the best here. Cant ask for more. I am more than certain this throttling at stock clocks will be fixed very soon.

Now for overclocking, that's not Dell's or any manufacturer's obligation to any of us. The cards have standard clocks and its their job to ensure those standard clocks will work just fine. And they will. They are listening to us. Just need to be patient.

Dell has already made official statements on power draw related protective mechanisms. Those are Nvidia design recommendations and no manufacturer will go against that. Clevo are not throttling at stock clocks, so its working as it should be. But I very much doubt Clevos are able to overclock any higher than what most see on their M18Xs.

Dell doesn't sell overclock guarantees with their GPUs. Pretty straight forward.

 

The thing is, mharidas, BIOS has nothing to do with the throttle. It's been already tested. The bios has already been pretty much dissected. The throttle is therefore prolly caused/controlled by either of these:
1) intentional VBIOS programming, or
2) faulty/incompatible VBIOS programming that glitches with Nvidia drivers, or
3) faulty hardware implementation of the Nvidia chip (i doubt that, btw), or
4) glitchy Nvidia drivers that don't work properly with Dell's design, but do work okay with e.g. Clevo.

edit: Clevo isn't throttling at OCed clocks either, BTW.

 

I agree very much with you bud... however showing someone cares is the first step to resolve something that we have all been without answers for three months. If this gets resolved, Alienware/Dell will have my business (line it has been for the past 7 years) in the future.

 

How was this "tested"? Don't take it the wrong way but people who are not with Nvidia or in the business of making the boards and assembly of such parts have no background or know-how to "test" this. Using disassemblers to unlock hidden menus and explore options is hardly testing or analysis.

Maybe I didn't make it clear, BIOS here is in reference to both BIOS. System and VBIOS.

Either BIOS can play an important role since Nividia's protective mechanisms are not completely on-chip yet. This is a design matter VLSI engineers can easily understand how Nvidia went about and what they are doing.

Hardware Reviewers have been able to fool the protective mechanism by renaming and trying other stress utilities that are not popular that can infact beat the OCP system. This alone shows its driver triggered, Just what and how much of a throttle should be done, can be upto the BIOS or be completely on chip via VRM controller.Its already known the latter is not the case so it via BIOS, They can do it by VBIOS or system BIOS or by both. There are no restrictions. Programming is programming, its all about how its chosen to be done.

There are no special dell only nvidia chips. Dell can have at the max their own custom PCB/board design, but the wafers are the same line as that of any other customer.

Clevos not hindered by OCP on overclocks? How high did they really go? So they have a PSU stronger than 330 watts? Either that or they are suffering ocp just as well, a few runs enough to get a high bench is hardly scientific. It just doesn't make sense for any vendor to go against Nvidia's recommendations on design to cater to the overclocking minority. If am Nividia I wont be pleased to get many cards back wasted because some ODM indulged in their customer's fantasies. Its a business, I would be only obliged to make the card work at stock clocks nothing more.

Throttling is happening even at stock clocks on the dell, that alone will hamper any clocks higher than standard even if overclocked by a little. Once the tuning is fixed you will find the roof of the dell cards and clevo cards in the same range. Neither of the brands will ever really get past OCP levels unless they volt mod the VRM controller, then they can yes. Simply because its a Nvidia design recommendation that every ODM will follow.

 

Wow. I guess a lot happened since yesterday. I left the thread open in my browser, so I didn't get any email notifications of the dramatic posts that evidently happened since my last post and round of PM's. Since I did not see them, I can't respond to specific concerns. So a blanket statement:

I work for Dell. I have been here 8 1/2 years. I started in Latitude tech support and have held positions in consumer tech supporting printers, Inspiron laptops, and Dimension desktops (old school). I also worked the Dell Forums for a couple of years as a liaison. From there I moved on to corp comm. I also launched the first comprehensive social media listening program at Dell which has transitioned to a new team. I now work in Dell Product Group as a social media strategist.
There are too many reasons to list for capturing systems. When we need captures, the traditional way is for engineering to tell the L3's what they need, the L3's to tell the L2's and the L2's to tell the phone techs. Sometimes things get lost in translation through those layers. When I am requesting a capture of any customer, I am working directly with the advanced quality engineer, who leads any quality engineering excursion or issue investigation. So if you call tech support asking who's this Bill guy and is his capture request legit, they never heard of me, and they probably never heard of the capture request.
If you have any trepidation at all about a capture or official communications with Dell via PM, please feel free to delete the PM and ignore. I always provide my Dell email address however, in case you want to vet me out any further than Charles and Batboy have already done in the past.

Blanket statement over. On to the business at hand. I received 2 responses to capture requests, so I think we are good for now. I will let you know if we need more.

 

hey bill, have a partial read here...this should bring you up to speed as to what we are up against. this of course is the desktop version of this problem with partial work around.

Disable GeForce GTX 580 Power Throttling using GPU-Z | techPowerUp

also interested in what these four chips on the left side of the pcb are for?
on the link above...they have 3 chips with 8 legs (circled in picture), while the bottom has 4 chips with 6 legs. dont worry about the other stuff circled. that was another experiment and it proved not to make any difference...from what we found out.



anything you can find out would be helpful...thanks


side note:
so when the techs start testing...have them also test the watts in play....

 

Do you really think it's the same root cause? Mind you, my background is tech support and social media and definitely not engineering. So forgive me if I sometimes ask a stupid question.

I read this quote from ty_ger on the TechPowerUp thread:
"It appears that NVIDIA has been pretty thorough in adding "artificial load" programs to the list of programs which trigger the OCP."

That makes me think the throttling you are seeing in games is not the same root cause. Surely Nvidia/Dell didn't put any games on "the list." Amirite?

 

we don't know for sure, but my guess would be no games...but if the watts are the trigger...and the game is capable of pushing past this...it would still technically cause the same "cause and effect"
example:
if battlefield 3 pushes the card to 110 watts per card and the ovp is set at 109 watts...then it wouldn't matter that it's not furmark running..since all it sees now are max watts. this is only a speculation guess with no real way of proving it...short of the spec sheet for this pcb.

 

Ah ok. Gotcha. So I have a question. How fast can you or anyone replicate the issue in a game? Remember, I can't really talk about synthetic benchmark applications. It has to be a real world use case that I take back to them to replicate.

 

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ask this guy. since he seems to be able to do it on call

 

But he feels he owes me nothing. Why would I want to trouble him?

What I needed was a description of what caused throttling in a game. I guess all I really need to know is "100% load and it happens in a minute or two." Any specific games I should suggest the engineers try? Just the ones mentioned previously on this thread I guess. I will let them know.

 

Just to update and clarify, the advanced quality engineer and mobility video teams are now looking into this just in case the capture requests did not give that indication sufficiently.

 

yeah, i know.. just use his words to describe the problem to the engineers.

 

haha. Actually, I did just that. I also used the quote right after that, which had exactly what I was looking for. Had I re-read that page first, I wouldn't have needed to bother asking for replication steps.

 

no worries bill. all is good.

hopefully you'll get some positive answers for us.

 

This one's gonna be fixed real fast. Turns out we don't even need the captures. More coming soon.

 

Nice....Way to start the morning!! Crysis 2 with DX11 pack causes throttling during heavy prolonged fire fights. For me this was the fastest way to replicate. In Crysis 2 you can go in guns blazing, or take the stealth route. Taking the stealth route with a lot of NPC's caused this. To accelerate the throttle, download the Crysis 2 high resolution texture pack.

 

okay....so now you have struck my curiosity.....why all of a sudden now...it's going to be fixed real fast?

care to drop a hint or two?

 

damn

now i really wanna go to the green side

 

In this particular instance, not a chance. "People are people..."

 

What's strange is that I just checked my order status to see if my M18x replacement has shipped(still in production ) and it says that the 580M's are on the Fedex truck for delivery. I guess I'll upgrade my current M18x so that some lucky outlet sniper can get a sweet rig after the motherboard has been replaced.

Maybe I'll take pics of the 580M's and their heatsinks.

 

Please do. Look for anything that may indicate a rev #. Maybe A0something-or-other.

 

fair enough.

question...are these new batch of 580m's the same as the first batch?
do we have any news on this?

 

That's what I'm wondering to the first question and no info available on the second.

 

i guess i'll find out when mine ships.

 

Hey Bill the dell guy, for better understanding what's up with the 'throttle', see the pictures below. This is an old study from August.

Spoiler

Note that I made the STALKER Clear Sky screen shots (first 5) back then running a bit higher clocks. The EVGA OC Scanner screen shot presents the throttle kicking on on the stock clocks.



red = the throttle kicks in
blue = the game is started

GPU1 = the primary card, GPU0 secondary (nvidia inspector messes that up since 285.* driver series)






The point is though, that on stock clocks STALKER Clear Sly behaves exactly the same, especially if there is a lot of volumetric light in the game's morning, so the charts produced on stock clocks would look exactly the same. I run a bit higher clocks so I didn't have to wait for too long for the throttle to kick in. When overclocked, the game causes the throttle almost instantly after it's started.

The easiest way to replicate the throttle is to use any application/game that causes 100% GPU load.

You can try EVGA OS Scanner. There are two modes in which the application runs. The regular one does not cause the throttle and both cards run at 93-96% load on avg. The 'heavy load' mode, though, engages the throttle almost instantly, since it puts the cards at 100% load. If the application was blacklisted, we wouldn't be able to run a test at all, so it can illustrate the problem quite neatly.


This is the screen shot of "heavy mode" causing the throttle. Note 100% load and oscillating voltage and clocks.

 

Thanks for the screenie. Link sent.

 

Sorry, was still editing. See the thing now.

 

http://i.imgur.com/6iN9Cl.jpg


http://i.imgur.com/F8viYl.jpg

The SLI kit showed up. The cards are still on the truck. I will post pics as soon as they arrive. Fedex has until 4:30P Alaska time to deliver the cards.

 

Same HS's as for the 69x0's

 

I was sent 2 pairs of heat sinks twice. The first pair of heatsinks NV1000W that I received and tried did not rest completely flush with the 6990M the way their original heatsinks do. After further inspection of each heatsinks I found that the 6990M's heatsink has a few differences compared to the 580M heatsinks.

 

Yes, they are different heat sinks. Before my system exchange, my Level 3 Alienware rep arranged to have a 580M SLi kit sent to me because I was having unresolved issues with the AMD components. I had already received heat sinks for AMD video cards, but the GPUs were not available. It turned out the 580M GPUs were also on backorder, so I was not able to receive them. The NVIDIA heat sinks and GPU link cable are different part numbers than the AMD parts and will not interchange according to Level 3 Tech Support.

 

(Didn't want to quote the long post in order to save space, but that's the one I'm replying to.)

You're using a modded desktop driver, correct?

This is overcurrent throttling. 55C+99% load on an OC'd card will cause the system BIOS to engage overcurrent protection and throttle, as these cards are more than capable of hitting the total system max power threshold under these conditions.

As far as EVGA OS Scanner goes, the Regular Mode is significant as it demonstrates real world gaming. The Heavy Load Mode is a stress application which will cause the issue we are seeing.

Just to be very clear: This system is not designed for GPU overclocking.

If you find anything I just said to be inaccurate or outright incorrect, please try not to get excited and just point out the error. I will get clarification. I am just trying to translate the reply I got back on this one.

 

So is it the system bios that is causing the throttle at stock clocks? If you can't answer just ignore my post . I'll live.

 

Are you talking the first 5 screen shots? They come from STALKER Clear Sky which causes the throttle also on stock clocks. I used the screen shots to illustrate how the throttle looks like, since I had them taken beforehand. I could make another test showing exactly same behavior if only 100% GPU is detected running stock clocks and the very same game.

EVGA OC Scanner was run on stock clocks.

I was on the drivers provided by Dell, AFAIR. 269.03. On every drivers since then, 285.27 beta and 285.38 beta, talking the newer series supporting Dell cards, the system/the cards behave in the very same way, in any application that causes 100% GPU load for too long. "too long" would be here longer than 60-90 seconds. Note that avg operation temperature of 580m SLI under load on stock thermal compound is some 70-73*C, so if any throttle-threshold is set to be engaged if the 55*C + 100% load condition is met, then this surely is too low.

100% load + 90*C would sound more like something applicable and useful -> this would be both efficient and safe

Can you make this happen, I mean, such a fix to be released?

I understand the difference. The problem is that apparently there are games that make the cards run at 100% load for at least some periods of time as well. This is where the throttle is engaged. This is why people report experiencing 'throttling' in games sometimes running stock clocks, e.g. if there are more intense effects during particular parts of gameplay.

I was told by an Nvidia rep that extensive 100% load would engage a protection mechanism to protect the cards from overheating, yet he was also surprised that the cards 'throttle' as soon as they hit 100% load for such a short period of time like 60 seconds. I was also told that the 'throttle' in such a case is enforced through forcing P8 performance level clocks (so 620Mhz -> 74Mhz) and is, in most cases, temperature related.

If only, as you said, the "throttle equation" is 55*C + 100% load, then these limits need to be reconsidered.

 

An important question for Bill. For the sake of clarity. What is the power limit you are talking about? How many Watts would that be? It kinda seems to me we should be rather given more powerful PSUs than restrictive GPU throttling.

 

No, but it will under some conditions.

 

I will try for a response to your 2 posts above and get back to you.

 

Thanks. This needs to be checked. As of now: if only 100% GPU load for too long = throttle. Always. Stock clocks. Temperatures below 75*C. Something's wrong for sure.

 

No worries. I may be coming to you in PM for a different request shortly. (not a system capture)

 

BTW, hey skygunner, can you measure the thickness of the thermal pads there? 1mm right?

 

told you temp was part of the problem and that you needed to below a certain point. if it was below this (crazy 55C (smh!)) then the throttle would take far longer to show (if at all). sort of like the turbo timer and the extreme cpus...once you drop below it's trip point...the timer is really no longer in play to throttle the cpu. almost what is going on here.

 

I'm not nearly as savvy as you guys, but one thing about the 55C thing. What I understood from the email I was reading, the temp is not the operative variable. It's OC+100% load that initiates the throttling. Am I just totally off base here?

 

I tried to measure the time limit for the throttle to kick in. With a stopwatch. Always longer than 60 seconds. This means an application/game starts, the cards hit 100% load, 60+ seconds passes, the throttle is engaged.

I had no idea who came with the idea of 55*C there. Especially as the fans are engaged when the temp exceeded 63*C. It just doesn't make any sense.

But this also means that as soon as the limit is risen the problem will be fixed... unless a new problem will be too weak PSUs.

 

That would be much more like: stock+100% load = throttle.

The ideal solution would be to get rid of the limit whatsoever.

the equation "100% load = throttle" on a gaming laptop like m18x is just... completely pointless. It is ultra counter productive. A temperature limit like 90-95*C would be safe and efficient.

 

PM coming. Please reply at your earliest convenience.

 

Thank you all for your efforts to resolve the issue that concerns us all!

I hope you will continue to keep us informed of your work

If it were possible I would have given 10 points of rep!

 

They do look to be 1mm. The pads thickness is exactly the same as the 6990M/6970M's heatsinks pads.

 

power supply seems to be on track. we have really tried to push that and so far...best we could come up with is 330 running occt (scook9) i couldn't even break 310 over clocking the whole laptop and running benches or in game benchmarks (1 to 5 minutes of actual game play). so im going to go out on a limb and say..this 330 watt psu...which can probably hit closer to 400 watts before shutting off...is fine for the most part.

although..what we dont know...i was the actual power draw of the card is...i dont remember any of you guys running test with a kilowatt meter...if you did...ill look for the links. but im still think same had the highest watt usage out of the whole m18x community.(but he had dual 6990m's rated at 100W each)

 

No problem. They pay me to do this, but I enjoy it anyway. I've been making my living in social media since early 2007, and this is by far my all time favorite community.