R9 290X Variance

[3DVagabond]

So that makes four websites that found differences between press and retail cards? So if you don't use the AC 24/7 and have high ambient temps, you're basically screwed if you own a 290 or 290X.



I'm frankly astonished that this kind of crap got past AMD's QA. What a botched launch, I can't believe they didn't do extensive testing of their product in a variety of ambient conditions. 50C is probably the norm for a good portion of the hot and humid southeastern US, unless they run the AC 24/7.

Have you ever lived in the southern states? I lived in E Tx. and in the summer the AC was on 24/7. When it's 100F+ at midnight do you think people turn off the AC? lol

As far as being "basically screwed" just flip the switch to uber if the card is throttling to maintain temps. Or, you can get really really exotic, I mean multiple computer science degrees exotic, and manually adjust fan speeds.

Pumping heat into the case to raise the ambient temperature and setting the cards to the reduced fan setting is not going to get you good results though. Did you really need a review site to point that out for you?

Think about some of the ways we've seen these sites test these cards. Pumping heat into the case to raise ambient temps, reducing the fan settings to adjust for same sound levels, putting there hand over the blower intake, have you ever seen this type of contortions to make a card fail? Ever? Did you see any sites buying retail 680's to see if they boosted to the same levels as the review samples when they first came out and were boosting way above spec like this?

[H]OCP

(Kyle saw a GTX 680 sample card reach over 1300MHz running live demos but it could not sustain this clock.)

Wake up and smell the coffee, man!

 

[blastingcap]

Have you ever lived in the southern states? I lived in E Tx. and in the summer the AC was on 24/7. When it's 100F+ at midnight do you think people turn off the AC? lol

As far as being "basically screwed" just flip the switch to uber if the card is throttling to maintain temps. Or, you can get really really exotic, I mean multiple computer science degrees exotic, and manually adjust fan speeds.

Pumping heat into the case to raise the ambient temperature and setting the cards to the reduced fan setting is not going to get you good results though. Did you really need a review site to point that out for you?

Think about some of the ways we've seen these sites test these cards. Pumping heat into the case to raise ambient temps, reducing the fan settings to adjust for same sound levels, putting there hand over the blower intake, have you ever seen this type of contortions to make a card fail? Ever? Did you see any sites buying retail 680's to see if they boosted to the same levels as the review samples when they first came out and were boosting way above spec like this?

[H]OCP
Wake up and smell the coffee, man!

After having to clean three PCs in a row this past week, I know that dust gets all over the place after a while and if a card is throttling right out of the gate at 40% (even after just a few test runs let alone an hour or more of gaming), that is a bad sign for future usage in less-ideal conditions like when the case filters are cloggy and the video card fans are dusty. I don't know about anyone else, but I'd like to have a decent margin for error without being forced to crank up the fan speed and thus raise noise and lower life expectancy of the fans. Which is why I STILL haven't bought the R9 290 which would be the natural upgrade path for me. I'm still waiting for AIB coolers.

 

[3DVagabond]

After having to clean three PCs in a row this past week, I know that dust gets all over the place after a while and if a card is throttling right out of the gate at 40% (even after just a few test runs let alone an hour or more of gaming), that is a bad sign for future usage in less-ideal conditions like when the case filters are cloggy and the video card fans are dusty. I don't know about anyone else, but I'd like to have a decent margin for error without being forced to crank up the fan speed and thus raise noise and lower life expectancy of the fans. Which is why I STILL haven't bought the R9 290 which would be the natural upgrade path for me. I'm still waiting for AIB coolers.

There's nothing you are saying there that I'll disagree with. I would recommend a case with better filters and you up your maintenance schedule, though

Did you read my post though?

 

[blackened23]

Think about some of the ways we've seen these sites test these cards. Pumping heat into the case to raise ambient temps, reducing the fan settings to adjust for same sound levels, putting there hand over the blower intake, have you ever seen this type of contortions to make a card fail? Ever? Did you see any sites buying retail 680's to see if they boosted to the same levels as the review samples when they first came out and were boosting way above spec like this?

Maybe you need to read the LR article again because that isn't what happened. You still keep telling us that it is apparently justified that people are buying cards based on reviews they see - if someone buys a card and sees the "quiet" benchmarks on the web, shouldn't they expect to get that performance when the purchase a card at retail? You're sitting here telling me it's justified for cards to have 15-20% performance variances as compared to what web reviews state. I actually find that hilarious. Please tell me how, as a consumer, that is a justified situation - the only ones who win in that situation (with AMD delivering a deceptive/defective product) are marketers, shareholders, and AMD employees. The consumer doesn't win in this situation - You think it's completely okay for someone as a consumer to receive a product that isn't performing as it should in quiet mode?

If someone is a consumer please tell me how this is remotely justified as a consumer purchase. Please. Why should 15%-20% performance variances between quiet benchmarks in 290X reviews, and getting a 15-20% slower card at retail? Not everyone wants to run uber fan speeds to correct the problem, a large contingent of users will simply run defaults and/or find the quiet mode benchmarks on the web to be perfectly fine. Little do they know that their retail card doesn't perform anywhere near that - NOT EVERYONE will even KNOW how to correct the problem, not everyone reads news on the web or browses forums like this on a daily basis - fact of the matter is, there are people who bought 290X cards who received cards that perform below the level they should. I don't know if it was intentional on AMD's part, but regardless AMD did a disservice to those customers.

 

[Silverforce11]

TBH, have any reviewer tested reference Kepler cards inside a case with 45C ambients and compare its performance to that of an open bench or well ventilated 21C ambient setup??

I would imagine the results would look completely different.

Warning issued for thread crapping.
-- stahlhart

 

[blackened23]

My personal experience with Kepler is that it hardly ever throttles, and when it does, it is so miniscule that it doesn't affect performance. I've only had cards that throttle 2 bins at maximum, which is 26mhz - and that was in super demanding games with vsync turned off.

That's a far different situation than the 290 throttling down to 727 to 650MHz. I mean, you have a GTX 670 - isn't your experience more or less similar? I've never talked to anyone with a Kepler GPU where throttling was a substantial problem. It can happen to reference cards, but the extent of the throttling (as mentioned) is so small that it doesn't affect performance. Certainly, it isn't causing 15% or more performance losses.

 

[3DVagabond]

My personal experience with Kepler is that it hardly ever throttles, and when it does, it is so miniscule that it doesn't affect performance. I've only had cards that throttle 2 bins at maximum, which is 26mhz - and that was in super demanding games with vsync turned off.

That's a far different situation than the 290 throttling down to 727 to 650MHz.

You always say that and I posted results from Hardware.fr that clearly showed otherwise. They did just the opposite with their GK110 cards though as we are seeing done with Hawaii. They added extra cooling to stop the throttling. To their benefit they showed both results. Out of box and what performance would be like by either adding an extra fan blowing on the card or upping the fan speed on the cooler. They didn't slow the fan down, pump extra heat into the case, etc... to create an unnatural condition to bring on exaggerated throttling. I'm sorry. I've always considered you pretty objective, but if you can't admit there's something going on here we've never seen sites do before, you need to show this same type of treatment to nVidia cards.

Infraction issued for thread crapping.
-- stahlhart

 

[blackened23]

You sidestepped the question of how this is a justificable situation with AMD delivering a defective (or at least, underperforming compared to where it should be) product at retail. And no, I stand by the fact that Kepler doesn't throttle to an extent that causes dramatic performance differences. I have seen this with my own two eyes using the Kepler in harsh ambient conditions - plenty of websites have analyzed performance over time with the kepler, and what you're stating just doesn't happen.

Do you see any differences there? A 1 bin throttle versus nearly 250mhz. No wonder AMD didn't publish a base clock for the 290X.

 

[Silverforce11]

You sidestepped the question of how this is a justificable situation with AMD delivering a defective (or at least, underperforming compared to where it should be) product at retail. And no, I stand by the fact that Kepler doesn't throttle to an extent that causes dramatic performance differences. I have seen this with my own two eyes using the Kepler in harsh ambient conditions - plenty of websites have analyzed performance over time with the kepler, and what you're stating just doesn't happen.

Do you see any differences there? A 1 bin throttle versus nearly 250mhz. No wonder AMD didn't publish a base clock for the 290X.

Thats what Quiet Mode is for, why are you harping on about it? If you dont want increased throttling, you use Uber Mode.

The difference between AMD and NV is AMD caps the fan speed and clocks will scale based on temps. NV caps clocks with a certain boost speed, and increases fan speed to achieve it if temps are high. Put a 780 into a high ambient situation, its fans are going to get very loud.

 

[blackened23]

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_780_Ti/32.html

780ti review, reference, AUTO FAN:

NVIDIA's GeForce GTX 780 Ti successfully captures the single GPU performance crown back from AMD. The card delivers impressive performance thanks to the GK110 GPU which has all its 2,880 shaders enabled. Averaged over all our benchmarks, performance is almost 10% higher than the GTX Titan and 8% higher than the R9 290X in its noisy "Uber" mode. The performance gap is 13% versus the R9 290X in quiet mode, or the R9 290, and the card even manages these performance levels with minimal throttling. Some investigation yielded an average clock rate of 993 MHz, which means that, thanks to NVIDIA GPU Boost 2.0, the card always stays well above NVIDIA's guaranteed 876 MHz the card will never go below. On AMD's R9 290 Series, we've seen major clock drops from time to time, depending on the heat situation, dropping the card below 700 MHz in some cases, though it is officially specified to run at "up to 1000 MHz."

Like I said. Kepler doesn't have the performance variances that the 290 has. @Silverforce: because the quiet mode benchmarks on the web don't represent quiet mode performance for retail purchased products. SKYMTL at hardwarecanucks, as mentioned earlier, has 4 retail cards and NONE of them perform similarly, there is a 15% performance delta.

That's the issue. The quiet mode benchmarks appearing on the web are essentially a farce, because retail boards aren't performing at that level. If you look at various 290X reviews, you would think that the performance loss for "quiet mode" would be fairly minimal compared to uber mode - but that really isn't the case. Most of these cards purchased at retail are performing well below what all of the published reviews indicate with wide performance variances. That's the problem: all of the existing 290X quiet mode benchmarks on the web are deceptive due to the fact that retail bought products perform up to 15%-20% slower in quiet mode.

 

[Silverforce11]

Has anyone tried to force NV reference 780 to stay at 40% max fanspeed and throw it into a high temperature ambient and see what it throttles to??

The throttling you are claiming is overblown, its extremely easy to go into Overdrive, set a higher fan speed cap and it won't throttle in games, if you throw it into a case with high ambient, it will just crank up the fanspeed speed higher, sucking out all the hot air and dumping it out.

 

[ShintaiDK]

You sidestepped the question of how this is a justificable situation with AMD delivering a defective (or at least, underperforming compared to where it should be) product at retail. And no, I stand by the fact that Kepler doesn't throttle to an extent that causes dramatic performance differences. I have seen this with my own two eyes using the Kepler in harsh ambient conditions - plenty of websites have analyzed performance over time with the kepler, and what you're stating just doesn't happen.

Do you see any differences there? A 1 bin throttle versus nearly 250mhz. No wonder AMD didn't publish a base clock for the 290X.

Thats a good example of the issue in question. The Boost vs Baseclock delta. In AMDs case, there is no official baseclock tho. This is also why I get worried when I see 280X reference cards. The delta for Tahiti chips got expanded from 50Mhz to 150Mhz. On Hawaii its just a plain disaster that AMD dont want to add a number to. It resembles of the FX series with missing specs.

The higher the delta, the less confidence the maker got in its product. And it shows by the variance of cards. The delta can also be abused, as we have seen, to inflate review site scores with the press cards.

In your example the GTX loses its boost, thats bad in itself. But the variance is rather low with something like 40-45Mhz and the clock stays stable. The 290X on the other hand loses 150-200Mhz and starts a stuttering behaviour due to the constant clock variance that powertune desperately tries to raise.

 

[blackened23]

Has anyone tried to force NV reference 780 to stay at 40% max fanspeed and throw it into a high temperature ambient and see what it throttles to??

The throttling you are claiming is overblown, its extremely easy to go into Overdrive, set a higher fan speed cap and it won't throttle in games, if you throw it into a case with high ambient, it will just crank up the fanspeed speed higher, sucking out all the hot air and dumping it out.

Did your 670 ever throttle by 300mhz? I'd be willing to bet your experience with the 670 matches my kepler experience. Rarely throttles if ever, and when it does, it is only by 1-2 bins. Right? I can almost guarantee that because I used the 680 - 780 fairly extensively, and the temps where I live get fairly harsh. Of all of my friends using Kepler GPUs, performance variance isn't an issue, even with auto fan. Honest question. Did your kepler ever throttle by 200-300mhz? I'm 99.9999999% sure that answer is no - i'm pretty certain your kepler experience has been similar to my own. If it wasn't, i'll stand corrected.

Also, you don't compare fan percentages. You compare similar sound levels, that's the entire purpose of "quiet" mode. You can leave the 780 at auto fan and it will never approach the sound levels of the uber mode 290X, and it won't demonstrate the same level of performance variance of a quiet mode 290X.

 

[Silverforce11]

That's the issue. The quiet mode benchmarks appearing on the web are essentially a farce, because retail boards aren't performing at that level. If you look at various 290X reviews, you would think that the performance loss for "quiet mode" would be fairly minimal compared to uber mode - but that really isn't the case. Most of these cards purchased at retail are performing well below what all of the published reviews indicate with wide performance variances. That's the problem: all of the existing 290X quiet mode benchmarks on the web are deceptive due to the fact that retail bought products perform up to 15%-20% slower in quiet mode.

No, its overblown. Quiet Mode specifically mentions a lower fan speed cap AND all reviews explain the thermal throttling on R290/X. Reviewers CLEARLY point out their testing conditions, ie. 21C ambient, open bench or well ventilated cases.

If there's higher ambients for the end user, they have the simple option of increasing fan speed or setting it to Uber Mode.

 

[Silverforce11]

No my 670 never got below its rated boost speed, but it got FREAKEN LOUD when temps were high. Way louder than review sites with their optimal testing conditions. I would be interested in seeing a reputable review cap 780ti fanspeed to 40% and see how it does in a hot case.. any takers on the outcome??

 

[blackened23]

No my 670 never got below its rated boost speed, but it got FREAKEN LOUD when temps were high. Way louder than review sites with their optimal testing conditions.

Well, the 670 reference definitely had a cheap cooler on it, it's a far cry from the Titan shroud. I still don't think Kepler "variance" is an issue. I have never seen any type of appreciable variance and I do live in the southeastern portion of the US - temps here get crazy during the summer. I run AC nearly all the time and the computer room in my home still gets extremely warm during the summer. Despite this I haven't noted any significant throttling. Throttling a little? Sure. Not 300mhz, though. Not anything causing a 15% performance loss.

Anyway, I just want AMD to fix this issue. In all likelihood, i'm going to get a 290 for litecoin mining in rig #2 in my computer room, and I don't want to deal with throttling / variance nonsense. AMD just needs to do the right thing and ensure that these performance variances, even in quiet mode, don't happen. Know what i'm saying? It's a valid complaint - people should not buy a product at retail that underperforms what press web reviews would indicate.

 

[Silverforce11]

See, you don't get the difference.

Kepler variance is noise. Because the fan will spin as high as it needs to maintain a certain defined boost clock. As far as I know, nobody has reviewed them with a low fanspeed cap to see what happens to the boost blocks.

R290/X variance is performance, due to a capped fanspeed. This variance can also be changed to that of noise, very simple in Overdrive menu, its just a max fanspeed slider, pull it to the right and thats it, no performance variance.

Edit: Out of the box, Kepler also have a small performance variance along with the noise. Since if ambient is low, it will boost higher than its rated clocks, such as in optimal review conditions. Though the variance is small, 2-5% tops. For the record, my 670 got to 80% fanspeed playing Final Fantasy 14 due to high summer ambients, it was louder than my R290 with 47% fan-cap (which doesn't throttle in games, yet).

 

[3DVagabond]

See, you don't get the difference.

snip*As far as I know, nobody has reviewed them with a low fanspeed cap to see what happens to the boost blocks.

This is true!

Blackened, sorry but I'm through with you. You carry on with your agenda and continue to ignore or deny anything that is presented to you that doesn't agree with your position. I'm not going to simply have you repeat yourself like it's any more credible because you said it 100 times.

 

[ICDP]

Has anyone tried to force NV reference 780 to stay at 40% max fanspeed and throw it into a high temperature ambient and see what it throttles to??

The throttling you are claiming is overblown, its extremely easy to go into Overdrive, set a higher fan speed cap and it won't throttle in games, if you throw it into a case with high ambient, it will just crank up the fanspeed speed higher, sucking out all the hot air and dumping it out.

Yes, me.

http://forums.anandtech.com/showpost.php?p=35615542&postcount=108

GTX780 MSI Gaming editions hit 95c in Furmark after a few minutes and throttled to 914 MHz. I stopped it before my card caught fire

Essentially many people are claiming Hawaii is broken because they are comparing how Nvidia boost works and assuming AMDs should be the same. They completely refuse to acknowledge that this is how Hawaii was designed. It is why it has a quiet and an uber mode.

Nvidia boost works ever so slightly differently than AMD boost and people need to grasp this basic fact before declaring Hawaii "broken".

Nvidia:

• Temperature target is set at 80c default
• The fan will keep ramping up to max (80% default) required to keep that temperature target.
• Fan noise is not considered a priority.
• When/If the fan reaches max allowed (80% default on my GTX780) and the temperature is still rising the core clock begins dropping in 13MHz increments in an effort to stabilise at the temperature target.
• The GPU has a minimum core clock.

AMDD:

• Temperature target is set at 95c default
• The fan will usually only ramp up to the maximum determined by the quiet (40%) or uber (55%) mode setting.
• If the minimum core clocks are reached (290 = 662 and 290X = 727) then the fan will start ramping up beyond the the preset mode.
• Noise is considered a priority in quiet mode, if you use this mode expect throttling by a large margin in some cases.
• Noise is not considered a priority in Uber mode and throttling is not an issue.
• The user has a choice that can be met with the flick of a single switch.
• If/When the temperature reaches 95c and the fan is also at the highest for the chosen mode, then the core clock begins dropping to maintain the chosen temperature.
• The R9 290 has a minimum core clock of 662 MHz, the R9 290X has a minimum core clock of 727.
  

With both types of GPU it is possible to override these setting using 3rd party software (or with AMD the built in OC utility) to adjust the max temperature and fan profiles to suit your needs.

They are almost but not quite the same, AMD simply has simply set a lower default max fan speed in quiet mode to keep noise in check and has not set a minimum clock. Obviously the Nvidia ref GTX780 cooler and the after-market AIB coolers are far more efficient than the one on the R9 290/X reference cards so throttling and noise is less of an issue. Having said that I have used enough reference Nvidia GPUs to know that 80% fan speed can easily be reached in prolonged gaming and that they are loud.

It has already been show many times that uber mode on the R9 290X eliminates or reduces the throttling to almost nothing. Admittedly at a cost of excessive noise.

I have been helping a friend flash his R9 290 to an R9 290X and can say without doubt that Uber mode does not throttle by any significant measure in any of the prolonged testing we have done.

 

[Silverforce11]

Great post ICDP, thanks for the input.

AMD also seem to have a special setting, if it detects high ambient temps, it will override the Quiet/Uber mode and allow the fan to ramp up higher if its needed to keep temps <=95C. Thats what happened with my R290s when I close my case, leaving the settings on auto, it reached 75% fan speed mining LTC. If i leave the case open, it stays at 47% fan maxspeed but throttles all the way down to 662mhz.

Such a design is actually very helpful, higher ambient cases = crap airflow. But higher fanspeed on blowers = sucking hot air inside the case and dumping it out. It's a beneficial design which actually shows a great deal of care for crapping OEM systems.

Put a 780 into a crap case and it will try to ramp its fan to the max maintaining the high boost clocks.. what happens once fans are maxed and temps keep going up? Will the core down clock to keep it safe or will it burn??

 

[ICDP]

Great post ICDP, thanks for the input.

AMD also seem to have a special setting, if it detects high ambient temps, it will override the Quiet/Uber mode and allow the fan to ramp up higher if its needed to keep temps <=95C. Thats what happened with my R290s when I close my case, leaving the settings on auto, it reached 75% fan speed mining LTC. If i leave the case open, it stays at 47% fan maxspeed but throttles all the way down to 662mhz.

Such a design is actually very helpful, higher ambient cases = crap airflow. But higher fanspeed on blowers = sucking hot air inside the case and dumping it out. It's a beneficial design which actually shows a great deal of care for crapping OEM systems.

I didn't know this, thanks for the info. It seems AMD have adopted the Nvidia approach to reduce RMAs. I have no problem with technology being incorporated to keep temperatures in check. Though I do feel a drop from 947 to 662 is extreme.

Put a 780 into a crap case and it will try to ramp its fan to the max maintaining the high boost clocks.. what happens once fans are maxed and temps keep going up? Will the core down clock to keep it safe or will it burn??

I assume it will keep dropping until the base clock is reached and after that the temperature keeps going up. Eventually the GPU would just lock up. If this is how it works then I can see how the AMD keep dropping core clock method would be better for people who leave their PC running. Such as when LTC mining.

 

[blackened23]

This is true!

Blackened, sorry but I'm through with you. You carry on with your agenda and continue to ignore or deny anything that is presented to you that doesn't agree with your position. I'm not going to simply have you repeat yourself like it's any more credible because you said it 100 times.

I don't have an agenda other than wanting a better product, as far as agenda goes I can say something along those lines based on what I've seen you post here but I generally don't do that. It is rude. Right now, you're being rude. Did you accuse me of "agendas" when I talked about the 7970? FYI, I really liked that product, and the 7970 did not have performance variances - it was a great value (post release), had great performance, overclocked like crazy, and had consistent performance without necessitating super high fan speeds. I really liked the 5870-7970 cards. Yet, suddenly I have an agenda when i'm pointing out a flaw in the 290X - flaws that did not exist with any prior AMD card. Suddenly, i'm the bad guy. Look, it's pretty simple: the 290X has variance issues that never existed before, and websites are just now catching on to this issue - these are problems that need to be made known to AMD. When LR told AMD of these issues they denied a problem existed. I considered the 5870, 6970 and 7970 to be great cards. I can't say the same for the 290X. Not yet anyway, not until these issues are fixed. And it's doubtful they will be fixed since AMD is completely denying the problem. Suddenly, apparently, pointing something like that out makes me the bad guy.

Instead of being rude, just agree that these are valid concerns that AMD needs to fix, period. I want to buy one of these cards for LTC mining, and I don't want to play the retail lottery in doing so; AMD needs to fix this problem.

I mean I could sit here and ask rhetorical questions all day about agendas and bias. But I don't do that. I dare you to tell me that these aren't valid concerns about performance variances in retail cards - please tell me why this situation should exist? Why? I've already shown that the situation with Kepler is far different - and you have users here who use AMD cards stating the same. That Kepler cards are meeting the advertised boost clocks with minimal throttling, and the TPU 780ti review shows the same.

 

[parvadomus]

@blackend23 u funny guy. You still dont understand how amd boost work, and thats why u seem to have an agenda. Anyways this topic is over at least for me. boost works as intended but the reference cooler dont let it shine, quite the opposite indeed. But I wouldnt care less, why you do? You end up being boring.
PS: I dont remember you making 500 posts about the crappy 670 cooler.

Infraction issed for personal attack and thread crapping.
-- stahlhart

 

[Erenhardt]

I think you two (ICDP and blackened23) should make a bet. Will gtx780 die in stress test with capped fan at 40%, or will it show significant performance variance?
What we know is Hawaii will not let melt itself, will cap fan/downclock - whichever is priority.
blackened23, are you saying that a card which can make suicide run by default is a good design and not faulty product, but one that will prevent damage is bad?
Who will pay for the damage in case of failure?

Infraction issued for callout and thread crapping.
-- stahlhart

 

[stahlhart]

Thread reopened. Review the content of the OP and stay on topic if you're going to post in this thread, and stop trying to derail it.
-- stahlhart