Radeon 5970 Overclocking: The VRM Temperature Bottleneck

[mikeyakame]

@mikeyakame

You like good cars

But your wording about HD5970 and failing under OC is harsh. It's not failing only throttling

Cheers lad so do yoU Classic jap cars are awesome!

I think I took it a little far with that, but what I meant to say was this.

i use this example with OE tuning in mind, no modifications etc. All as per the factory.

Lets say you take 2JZ-GTE and throw it on an Engine Dyno, now different manufacturers make different engine dynos, and of course the programming to cycle the engine through different loads and conditions will be different for each and of course also reprogrammable. Now if I used two different engine dynos, one represents a typical stress load and the other represents a unique stress load, the typical one doesn't turn up any issues but the unique one causes my engine's ECU to fuel cut from running lean.

I would be going straight to Toyota and voicing my problem to them, because an Engine Dyno is an Engine Dyno and how it loads an engine should be irrelevant, therefore the fuel mapping should be adequate that under any standard conditions it doesn't risk blowing my engine up. Sure they might be some unique ones but selling me a performance vehicle with a performance engine I expect this kind of testing to be done.

It would be inexcusable for Toyota to say well the Engine Dyno you used is an Engine Power Virus because we figured that 99% of our customers won't ever test the engine under those conditions. Engine Dyno is an Engine Dyno how I stress the engine should be irrelevant as long as I'm doing it in a responsible manner. No mechanic or engine builder would accept this as an excuse and Toyota would be called out for it and made to fix it or else lose respect of industries, yet doing the same thing with a Video card is considered alright because you aren't paying $100K+ for it new.

Engineering should consider all scenarios whether they are unique or not, and especially more so if you sell a product as one you can thrash or tune.

My thoughts have nothing to do with any manufacturer, rather the nature of an industry. I'd have the same grief if Nvidia did this, if Intel did this, even if Toyota, Honda or Nissan did this. I think this is what a lot of the guys not defending the matter here are trying to say, it's not about fanboyism it's about delivering a product engineered to do what it is designed to do under typical and atypical scenarios, not doing that is inexcusable.

[Manicdan]

i dont think engines make a good comparison for this,

try it with tires. factory rated tires can handle the 3200Lbs the car weights up to 90MPH where the car electronically limits your speed (this is a real scenario for some cars btw)

if you remove that limiter and push your car to 120MPH, your tires then rip apart and you die in a giant fireball. your wife then calls them up and says its the car companies fault. is she right? or is it the owners fault for letting his car go to new speeds and not getting Z rated tires. (this is with the 4800 cards)

the 5800 example would be the car automatically shuts off the fuel as it feels unbalancing in the tires, and knows they are about to fail. sure the car can still go faster, but not fast enough to cause anyone to die in a giant fireball

hope you like my example

[mikeyakame]

i dont think engines make a good comparison for this,

try it with tires. factory rated tires can handle the 3200Lbs the car weights up to 90MPH where the car electronically limits your speed (this is a real scenario for some cars btw)

if you remove that limiter and push your car to 120MPH, your tires then rip apart and you die in a giant fireball. your wife then calls them up and says its the car companies fault. is she right? or is it the owners fault for letting his car go to new speeds and not getting Z rated tires. (this is with the 4800 cards)

the 5800 example would be the car automatically shuts off the fuel as it feels unbalancing in the tires, and knows they are about to fail. sure the car can still go faster, but not fast enough to cause anyone to die in a giant fireball

hope you like my example

But this is happening under reference conditions. If it happens overclocking it's a whole different ballgame. That's what I'm trying to get at.

If that same fuel cut kicks in at 70mph when your car is designed to do up to 90mph on those tires it is delivered with, would you consider this acceptable or would you be inclined to say its a manufacturer fault for not correctly implementing the fuel cut.

[LordEC911]

ill agree with you on some things here...the 295 was the fastest card up until this past month..however it wasnt the best overall card out..the 5870 was..and competed VERY well against a 295..

fermi is a lot of speculation and until it comes out..you can only hope itll live up to the expectations.

also another thing ati has done is become VERY powerful while keeping power consumption down a lot...i mean the 5970 quite a bit faster but uses around the same power...thats AMAZING in it self..also they brought eyefinity which is great as well..

another thing..13 months??? where did u get that from..if im not mistaken..didnt it launch in Jan 09??? and was best up until Nov 09...
im not a wiz with numbers but that doesnt look like a year in between those two dates..
and before that for 5 months ati had the crown..
so it was like 5 months ati 9-10 months nvidia back to ati until atleast march if fermi reveals its goodies by then ...which will probably be a single gpu..might not beat the 5970 even then..

as far as sticking with their arch..i guess u cant be MAD at them for it..but i wouldnt care to buy a product that is rebranded..that would upset me vs buying something new...but i guess u can praise them for that idk?

Don't try to refute his claims with facts... That is just unfair.

[Manicdan]

But this is happening under reference conditions. If it happens overclocking it's a whole different ballgame. That's what I'm trying to get at.

i think the biggest argument (or debate) is if furmark and occt are allowed to be used for testing or comparisons. dnet is looking to be a great example for failure under stock conditions since its a real program. did ATI ever give feedback as to their opinion for that program throttling and what they intend to do about it?

[mikeyakame]

i think the biggest argument (or debate) is if furmark and occt are allowed to be used for testing or comparisons. dnet is looking to be a great example for failure under stock conditions since its a real program. did ATI ever give feedback as to their opinion for that program throttling and what they intend to do about it?

I haven't heard or read anything about it. They haven't spoken a word of it, other than those 2 programs first mentioned are dangerous for testing. If dnet delivers the same problem and it is software for general usage, 24/7 or not then I would be inclined to say they need to address the problem.

[LordEC911]

I think we have to take things like furmark into consideration because we never know how stressful future programs and games will be. Also if it is not stable in something brief like OCCT or furmark, who is to say we are not causing longterm damage to the card over a longer period of time.

People saying we will never see loads like furmark simply want progress to be held back for the sake of their argument. Something like this program can easily be fixed if another phase of power was added to the card. When the cards are 600 a piece and such a add on would be insignificant to the overall cost, they should do it for the benefit of the customer, especially if these cards are being marketed to overclock.

Honestly, I think most people would love if a game was programed so well, that it was using 100% of the videocards power. This might be a tad unrealistic because its difficult to programs in such a manner when there are so many hardware configurations there are, unlike consoles.

However, I can see a program(not game), especially if open Cl does take off, which could be as stressful as furmark. If these cards ever want to be taken seriously in the supercomputer world, they better be beefed up because there will be programs that use 100%.

It was not until the 4870 series, which had crashing with furmark, that we started calling it a power virus on AMD accord.

However what is the point of a stress program if it does not show long term stability. A stress program has to prove stable in the short term at above average loads to show a piece of hardware can stay stable in the long term at lower loads. If elevator manufacturers only tested to the maximum rating, most of us would feel alot less safe in a crowded elevators. Or if tires were only tested up there speed rating, we would see alot more tires burning out.

Programs like Prime95 or Intelburn in, test beyond realistic loads, but do we call them power viruses? No we don't.

How do you know the VRMs are underbuilt?
I had some very nice posts in the OCCT thread that went unnoticed and were buried by the common sheeple, so I feel your pain.

The thing is, other than a few synthetic programs, no real app has come close to stessing the cards this much, not even the GPGPU programs that run quite well on AMD/ATi cards. This dnet thing is new to me so I will have to look into that, haven't heard about it before.

Power virus was brought up due to the fact that it very accurately described these aforementioned programs. They seemed to have been specifically coded to exploit ATi's shader architecture, especially in the case of OCCT since it was a Nvidia sponsered app.

No longterm study has been done on the longevity of these cards being stressed. Any comment made on that subject is pure baseless speculation with no evidence or facts.

the 295 has no thermal problems though, in the dual or single PCB version.

Mine idles at 45C and never goes above 75C even after 4hrs of Crysis.

Here you are, crying about other crying about a bias synthetic test, and then say your GTX295 doesn't have any thermal problems because you don't have high temps in Crysis. What a freaking coincidence...

If you look at anhilators posts, they are generally quite neutral. I don't thinks its a matter of putting faith particularly in nvidia. Its that the general attitude is Nvidia is bad, ATI/AMD is good.

E.g underbuilt VRM, terrible quanities, AMD officially raising prices(on the 5870, 5970 and 5950) and using slag marketing tactics and on this forum, they are praised for it or don't under go criticism.

And if a criticism is valid, it somehow gets turned into not their fault or the blame is put on someone else. E.g it works fine under stock settings or games are that stressful in real situations(even those it is marketed to be an overclocked card or future games may prove otherwise), Its TSMC fault(they should have stockpiled longer), the markets calls for price increases(although it has never ever happened before), using slag marketing tactics(this one NV gets called y and some sarcastic comment about whoop ass; when AMD does it cool and everyone laughs at NV).

NV on the other hand does anything and they are reamed out to kingdom come. Some people who have generally a pretty good attitude in general towards both companys like SKYMTL, are criticized for being anti AMD, when its pretty obvious from their comments of past and their reviews that they are not.

This board has turned so pro-AMD, Anti NV that the news section has turned into flame-opulous. Some people even wished death(or joked about it), on Jen Huang, which is Ridiculously over the top or NV to go out of business. It was never this bad before and it started really when AMD started making a comeback. If you look at when the zero tolerance for flaming came into effect it was when AMD videocards became good again.

You keep talking about underbuilt VRMs... how do you know this is the problem? Quantities are purely thanks to TSMC promising a certain capacity and not delivering, not sure how you blame that on AMD/ATi. What are these "slag marketing tactics?" Quotes/links would be nice.

The funny part about you talking about AMD blaming others, that is where the blame actually lies. Nvidia on the other hand is purely blaming everyone but themselves for their problems, even though other manufacturers are NOT having the same problems. Ex- "We cannot release GF100 because 40nm is leaky and has too low of yields" but yet there are a handful of other GPUs on 40nm that might have had some problems but were/are being produced. "Bumpgate wasn't our fault, it was TSMC, AIBs, OEMs, etc." Yet they are the only ones with the problem...

Bringing up SKYMTL is sorta weird. Look at his post history, over the last few months he has almost nothing good to say about anything AMD/ATi related. His website is honestly quite neutral but for whatever reason that isn't being shown in his posts here.

Dude, I am pretty sure you were here more than a year ago when anything AMD/ATi was pretty much crapped on by the same dozen people flaming and trolling every single thread. Things were MUCH worse than the anti-Nvidia bias going on now on the forum. These things are constantly flip-flopping around but IMO it is not even close to as bad as the R600/RV670 time period.

Edit- Also AMD/ATi did take the blame on the Furmark/OCCT VRM problem on the 4 series. They said they didn't see the need to increase the cost and the price for the cards to statisfy an outlier of a situation and that has no effect on the actual operation/performance of the card.

[safan80]

It runs over 80c at stock, thats thermal problems as far as i'm concerned.



Considering nobody here is mentioning how ambient temperatures or fan speed play a role in stability (other than the article) this thread is pretty much moot..

both my factory OCed 285s get over 80C, and that's with the cooler at 100%. I bet if people were to run the 5970 cooler at 90% it would manage to stay below 80C without problems at stock. The thing is most don't want to do take advantage of the ATI cooling system by raising the fan speed. the heat and VRM issue is not just an ATI problem it's an Nvidia one also.

It took AMD a full 13 months (even w/ GDDR5/DX10.1/DX11) after the 295 was launched to catch up with it. Let me go ahead and avoid the fanboy comment by noting the 295 was the fastest video card in the world up until the 5970, me owning one has nothing to do with this fact.

So it appears to me NVIDIA was sticking with a very solid design and using it as a placeholder until the GT300 is ready for the masses, seems like a very good idea to me instead of rushing a product simply for profits in a non-competitive market (DX11). If any ATI fan boys would read about the GT300/Fermi they would see it is a completely new approach to GPUs and is by far more innovative than anything ATI has done in the last few years.

For my resolution the GTX 295 is a joke the stock cooler can barely keep it cool.
the GTX295 is gtx 260 based, if Nvidia's cooler had enough power if would been 285 based like the Mars. The Mars are overheating in a lot of cases. The only reason the 5970 does not come standard at the 5870 level is because they were trying to keep the video below 300 watts.

With all that said Nvidia between now and the 295 should been finished with Fermi. They went to the point of showing completely fake Fermi cards, and yes they've changed the focus of Fermi to avoid competing with ATI because Nvidia's Fermi can not compete.

Marketing BS is the only Nvidia has going for themselves at the moment, and for those that believe it I hope you get what you pay for.

[JoeBar]

^^

Well said...

[dan7777]

im currently running the 295 and am wondering which card to get 5870-or 5850 and later i plan to get anthor 1 down the line.

[Manicdan]

both my factory OCed 285s get over 80C, and that's with the cooler at 100%. I bet if people were to run the 5970 cooler at 90% it would manage to stay below 80C without problems at stock. The thing is most don't want to do take advantage of the ATI cooling system by raising the fan speed. the heat and VRM issue is not just an ATI problem it's an Nvidia one also.

in the anand article they said keeping the 5970 fan at higher speeds, like 70-80% was enough to keep dnet running fine. ill re-read it for exacts

[Manicdan]

From Anand's article

In the case of FurMark this was very quick and even at 100% fan speed the cooler could still not keep the VRMs cool enough to allow full-time 850MHz operation. The Dnet client on the other hand was much slower to ramp up, and we ultimately found that 70% fan speed was enough to keep our hottest bank of VRMs below the threshold, stabilizing at 116C.

If the fan speed were to ramp up in the case of near-critical temperatures in the VRMs, then the Dnet client likely would have ran without an issue the first time, as this would have pushed the fan to 70%.

the Dnet client is not a power virus. It’s a real application, one that AMD even used in their 5800 presentation back in September. Thus as far as we’re concerned, our 5970 is only good for 775MHz, the lowest clock speed where the VRMs stayed under 120C

so the answer is even Dnet allows the 5870 be OCed up to 850MHz as long as the fan is at 70%, if you can deal with 100% who knows what you can do. the only other cases with proven throttling are both moot points as far as i think anyone should be concerned.

[BababooeyHTJ]

if Furmark is using 100% of the card, then how does OCCT produce temperatures 5-10C hotter than Furmark?

when overclocked the card was throttling down in two cases

I don't see any evidence that this happens at stock speeds. Therefore I don't see a problem. Honestly, who buys a dual gpu card expecting high clocks on the stock cooler?

Which vrms are this card using the insanely hot volterras that are on my 280?

Edited for space

How could anyone ignore your posts in that OCCT thread? It was every other post. You felt that OCCT is a "power virus" did you expect a response every time that you mentioned that. I think Shinti bought up a good point in a surprisingly neutral way about how do we know that this doesn't effect any other apps? I mean why do we need to rename Furmark exe to get it to run properly on the 4xxx series? I'm sure that both companies are guilty of that one.

Once again what does Nvidia have to do with this thread? Do you have any evidence other than the fact that the dude who wrote OCCT was given a Nvidia card that it is a Nvidia sponsored app? I think it more goes to show how hard it is to work with ATI as mentioned by quite a few people with experience doing so in the Batman thread.

What because SkyMTL doesn't b**w ATI as enthusiastically as some people here he is some Nvidia fanboy now?

[LordEC911]

How could anyone ignore your posts in that OCCT thread? It was every other post. You felt that OCCT is a "power virus" did you expect a response every time that you mentioned that. I think Shinti bought up a good point in a surprisingly neutral way about how do we know that this doesn't effect any other apps? I mean why do we need to rename Furmark exe to get it to run properly on the 4xxx series? I'm sure that both companies are guilty of that one.

Once again what does Nvidia have to do with this thread? Do you have any evidence other than the fact that the dude who wrote OCCT was given a Nvidia card that it is a Nvidia sponsored app? I think it more goes to show how hard it is to work with ATI as mentioned by quite a few people with experience doing so in the Batman thread.

What because SkyMTL doesn't b**w ATI as enthusiastically as some people here he is some Nvidia fanboy now?

No one wanted to discuss my posts, my last few, they just buried it or trolled/flamed.

Bringing up Shintai and neutral/non-bias in the same sentence is something I thought would never happen. We have evidence of the fact that other than a few very specifically coded products, it doesn't affect other apps, especially GPGPU. You have to rename furmark because AMD/ATi decided to save cards and throttle the performance through software, which is now worked into the hardware with the 5series.

Nvidia was brought up because others were making the comparison, obviously in a negative light.

Does there need to be more evidence? Why do the AMD/ATi cards outpace the Nvidia cards so much? The code isn't fully utilizing Nvidia's architecture and he even admitted so and would look into. Some months later, this problem still hasn't been fixed.

Amazing at how many devs, that are NOT under Nvidia's thumb, have the complete opposite to say about AMD/ATi. Hell, even Crytek, who likes to go both ways, doesn't seem to have a problem getting help from either one.

As for SkyMTL, I have already said I don't know what the deal is. He has abosutely nothing positive to say about anything AMD/ATi related the last few months. It has nothing to do with being enthusiastic about AMD/ATi... and I also said nothing about him being an Nvidia fanboy. Why are you putting words in my mouth? Why are you resorting to these underhanded tactics? Why can't we have a civil discussion about the topic?

[STEvil]

but your concern and whats ok for the card is two different things..
the 4870x2 reaches upwards of 90c..which is probably to much for ur concern..however the card performs problem free at those temps...im sure amd knows just a little bit more about this then us? just an assumption

I have a 4870X2 and have modified the fan profiles to keep the GPU's and VRM's under 75c with 20c ambients. When ambient temperatures rise in the summer I use a different profile... its a little louder but keeps the card stable (I have had crashing issues at 85c+).

[BababooeyHTJ]

No one wanted to discuss my posts, my last few, they just buried it or trolled/flamed.

Bringing up Shintai and neutral/non-bias in the same sentence is something I thought would never happen. We have evidence of the fact that other than a few very specifically coded products, it doesn't affect other apps, especially GPGPU. You have to rename furmark because AMD/ATi decided to save cards and throttle the performance through software, which is now worked into the hardware with the 5series.

Nvidia was brought up because others were making the comparison, obviously in a negative light.

Does there need to be more evidence? Why do the AMD/ATi cards outpace the Nvidia cards so much? The code isn't fully utilizing Nvidia's architecture and he even admitted so and would look into. Some months later, this problem still hasn't been fixed.

Amazing at how many devs, that are NOT under Nvidia's thumb, have the complete opposite to say about AMD/ATi. Hell, even Crytek, who likes to go both ways, doesn't seem to have a problem getting help from either one.

As for SkyMTL, I have already said I don't know what the deal is. He has abosutely nothing positive to say about anything AMD/ATi related the last few months. It has nothing to do with being enthusiastic about AMD/ATi... and I also said nothing about him being an Nvidia fanboy. Why are you putting words in my mouth? Why are you resorting to these underhanded tactics? Why can't we have a civil discussion about the topic?

Your last posts around the time that the thread started to die down? No one trolled or flamed your last few posts there from what I remember. I do agree that you bought up some good points one of which is the fact that Furmark also stresses the 4xxx series cards harder than GT200. Look at the framerates in each bench. Different architectures and drivers are going to perform differently in different apps, you know that. You also know that Furmark which OCCT pretty much is sees much higher framerates on 4xxx series hardware. Also we know that my 280 and a 4870 draw a very similar amount of current under load. My 280 has what 4 or 5 of the same vrms dedicated for the core while the 4870 has 3? I'm asking because I could be wrong. We don't even know what that 82A in GPU-Z/Rivatuner even means. That was never answered in the thread.

Which devs are under Nvidia's thumb? Did my 9800gtx outperform my 4870 in FSX because Microsoft is under Nvidia's thumb? I just don't get you point there unless you are trying to imply that the dude who wrote OCCT is "under Nvidia's thumb" which is a ridiculous statement.

I also think that Shinti's point in that thread about not being able to be sure what type of driver tweeks are being done in games much like in Furmark from both companies. He actually bought up evidence of Nvidia doing this in the past.

What do you want SkyMTL to say? I'm not going to argue with you on this "point" since I really don't want to seem like I'm bashing ATI.

Also I don't see how any of this is "on topic". I said my piece about this not being a problem imo since it apparently doesn't happen at stock clocks. It is a dual gpu card after all. I think that I'm being civil here but some of the statements that you are making contridict what you said in that OCCT thread.

[LedHed]

I don't see any evidence that this happens at stock speeds. Therefore I don't see a problem. Honestly, who buys a dual gpu card expecting high clocks on the stock cooler?

I did and do, 295 delivers great clock speeds with just modifying the stock fan speed profile and software voltage tuner; the 295 was pretty much made to OC. (see sig for clocks) I've rarely seen the GPU go over 75C in gaming at the below clocks.

[BababooeyHTJ]

I did and do, 295 delivers great clock speeds with just modifying the stock fan speed profile and software voltage tuner; the 295 was pretty much made to OC. (see sig for clocks) I've rarely seen the GPU go over 75C in gaming at the below clocks.

and how dose that card perform against a 5970? How long after GT200 launch did that card hit the market? It's not even on the same die size as the original GT200. Also how do you even know your vrm temps?

So, you have seen the GPU over 75c? At what fan speed, even though there is no way for you to know you vrm temps which this thread is about? I know for a fact that those GPU temps are limiting your OC.

I know that the vrms on my GTX280 can get hot as he**.

[LedHed]

. Also how do you even know your vrm temps?

RivaTuner shows all VRM temps and my fan speed is based off them.

*Note no where in this thread did I say the 295 was faster than the 5970, the point of this thread is a thermal discussion. I don't know why people keep trying to turn it into ATI vs NVIDIA.

[clonez]

why is everybody posting car allegories

sure a car with normal tires can go faster than 90 mph (heck, i drove over 120 mph in DRIVING SCHOOL )
but this is about a graphics card not being able to oc properly under normal circumstances, you just just get what you pay for.. wait a month or two and youll get dual 8 pin + WC.. thats the real monster (nonetheless achievable with current cards)^^

[safan80]

why is everybody posting car allegories

sure a car with normal tires can go faster than 90 mph (heck, i drove over 120 mph in DRIVING SCHOOL )
but this is about a graphics card not being able to oc properly under normal circumstances, you just just get what you pay for.. wait a month or two and youll get dual 8 pin + WC.. thats the real monster (nonetheless achievable with current cards)^^

Well I'm going to do most of my gaming during the holiday, so the dual 8 pin will be out of the question.

[BababooeyHTJ]

RivaTuner shows all VRM temps and my fan speed is based off them.

*Note no where in this thread did I say the 295 was faster than the 5970, the point of this thread is a thermal discussion. I don't know why people keep trying to turn it into ATI vs NVIDIA.

Didn't know that any 55nm GT200 card had monitors for the vrm temps.

My point was that your card runs hotter than a single card therefore doesn't oc as well as a single card. Also was based off of a much more mature architecture with a much closer competitor (4870x2) at the time.

Also if you can afford a 5970, you can afford better cooling. Do we blame Intel because we can't get max clocks on the stock cooler? Did I blame Asus when the crappy NB cooling held back my clocks?

The card runs fine under these extreme tests at stock clocks which are still impressive therefor I don't see the problem.

[LedHed]

most video card owners use stock cooling, it's a luxury most can't afford. I saw my 8800 GTS 512 reach speeds so high with a volt-mod that I won a G92 OCing competition and won my TPower i45 on stock cooling at first. I also OC my 295 and the temps are very reasonable with custom fan profiles in RivaTuner.

[LordEC911]

Also I don't see how any of this is "on topic". I said my piece about this not being a problem imo since it apparently doesn't happen at stock clocks.

Agreed.

Would have been nice to have a heatpipe extend out to the second set of VRMs so it would have more direct contact with the vapor chamber, rather than just using the cooling bracket.

[BababooeyHTJ]

most video card owners use stock cooling, it's a luxury most can't afford. I saw my 8800 GTS 512 reach speeds so high with a volt-mod that I won a G92 OCing competition and won my TPower i45 on stock cooling at first. I also OC my 295 and the temps are very reasonable with custom fan profiles in RivaTuner.

What exactly is your point? What do you think long periods of furmark would have done to my card on the stock cooler? Again, if you can afford a $600+ card you can afford better cooling. So this card has a temp bottleneck? So does yours. It's max clocks (hell lets say stock clocks) are still a better bang for your buck than a GTX295 with lets say your max clocks. Anyone who has seen my post history knows that I'm no ATI fanboy, not by a long shot, but stop comparing your 295 to a 5970. Just stop.