Thanks. From that article it looks like a third party cooler is needed to keep GK110 under the 80 degree limit at max clocks. So there is room for chilly nVidia cards to report misleading performance numbers as well.
AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread
Thanks. From that article it looks like a third party cooler is needed to keep GK110 under the 80 degree limit at max clocks. So there is room for chilly nVidia cards to report misleading performance numbers as well.
NV's temp target != AMD's temp target. For AMD it is simply the setpoint for the fan control. But with nV cards it shifts the temperature, at which the clock speed gets reduced.
But that is simply wrong for stock settings and plenty of tests prove that. You need a more efficient cooling for that to be true (and/or a significantly raised temperature "target").
Are you saying that 290x won't downclock after hitting 95c at 40%/55% fan speed? How is that different to nvidia's downclocking at 80c?
See prior post.
Also, while they are similar in theory the reality seems to be a bit different.
Why is the fan so loud? www.youtube.com/embed/VY5fU9mcX4w
it sucks.....a lot of air into tight spaces that creates sound that the ear dont like.
its cheap, does it job unless you go outside specs like need more cooling for OC or the uber mode. its not silent, which is a huge deal to many.
AMD's solution downclocks the GPU only, if it hits 95°C and there is no headroom left for speeding up the fan (determined by the maximum fan speed limit set in the control panel). But it is just (not pure) concidence, that the temperature target for the fan control and the (not changeable) threshold for downclocking are the same.
Setting a lower temperature target usually decreases the performance of an nV GPU (as it downclocks already at lower temperatures) while it increases the performance of AMD GPUs (as the temperature is kept lower or at least the fan profile gets more aggressive, speeding up earlier) because the temperature may not reach the downclocking threshold (which is basically fixed at 95°C). That are two fundamentally different things. As I said, nVidia's temp target shifts the temperature thresholds at which the downclock happens (which is why you want to increase it for longer boosts also at higher temperatures), while in AMD's case it is really just the temperature target for the fan control as the name implies and has no influence on the temperature threshold for downclocking. So setting the temp target on a 290X to let's say 85°C will just influence the fan control (it tries to keep the GPU at 85°C instead of 95°C), but the downclocking still happens only at 95°C (if it reaches them with the maximal allowed fan speed).
Btw., what that über mode switch is doing, is simply setting the fan speed limit to 55% (instead of the default 40%), nothing else. It doesn't change clock speeds, voltages or power targets, nothing. That's why it effectively changes nothing at all, if you flip the switch after you have changed some of the limits manually in the CCC (as the BIOS values don't overwrite them, it's just a different default for the fan speed limit if you left the CCC untouched).
??? Setting a lower temperature target on 290x will absolutely lower clocks and performance. For the same performance and power/heat output the 290x's fan will simply spin faster and louder to maintain a lower temperature threshold. Sorry, this isn't magic - you can't get lower temperatures and higher performance without changing the cooler or living with ear bleeding fan speeds.
nVidia cards down clock when they hit the temperature threshold. AMD cards down clock when they hit the temperature threshold. What is this difference you're seeing?
Are you assuming that you can't raise fan speeds on nVidia cards as well? You've really lost me...
No, it won't. Read again what I wrote. You simply missed the fact that the temperature target setting for the 290X is basically just for the fan control and doesn't have an influence on the temperature threshold for lowering the clocks as it does on nV GPUs. That's the difference. You simply have to keep the 290X below 95°C and it won't throttle (as long as the also existing power limit is not reached, which is routinely only possible in Furmark or equivalents), irrespective of the target temperature for the fan control.
AMD's powertune uses a more orthogonal set of parameters accessible to the user as nV does.One can define a maximum allowable noise threshold (fan speed limit), a desired working temperature for the fan control (but it doesn't throttle to keep to that limit), and a maximum instantaneous power consumption (power limit). The threshold for downclocking is fixed at 95°C. It downclocks if it reaches that temp with maxed out fan (at the set fan speed limit), not the set temperature target.
NV on the other hand allows you to set a power limit (averaged over 100ms or so, iirc) and to shift the temperature threshold for downclocking (that' the slightly misleadingly named temperature target). And of course you can change the fan profile with external tools, if you want. It downclocks, if you reach the temperature target (it's a little more complicated as it progressively reduces the accessible boost clock when approching the temperature target, iirc).
Last edited by a moderator:
I get that you're saying the fan threshold and downclock threshold are decoupled. Do you have a source or any evidence for that by the way?
Fortunately, someone has done the experiment on our behalf and lowered the temperature target. As expected noise goes up and performance is lower.
Let's assume what you claim was true. It would mean that lowering the temp target to 70°C would drive fan speeds higher but not affect clock speeds until the card hit 95°C. However, fan speeds are capped. There is absolutely no way to limit the card to 70°C and stay under max rpm without also lowering clocks/heat output. You have to first find a better cooler as evidenced by Guru3D's numbers.
This is the same behavior as nVidia cards.
It was described that way in some reviews, for instance THG's.
But...
I have to admit, that this test speaks against it. There are a few points which tell me, it was not a very realistic test (they must have had an especially cool running card or it was pretty cold in their testing environment for not seeing any marked difference between quiet and uber mode), but anyway.
It may be a moot point now, but while for a specific fixed setting the behaviour of the cards may be similar, but the function of the temp target is different. It is at least coupled to the fan control for AMD GPUs (I thought it does this exclusively), while it is not in nV's case.
gamervivek
Regular
Always attribute to malice what can be attributed to stupidity, for in either case they don't deserve their positions and they should be punished and punished well.
Dave Baumann, couple simple questions to resolve a discussion on another forum:
- 208W is said to be the max consumption of the Hawaii GPU with PowerTune at 0%, right? It's the chip only, not the whole card?
- Adjusting volts via 3rd party software can't make the GPU (or card if the first question is cards consumption, not GPUs) go past the 312W max consumption of the Hawaii GPU (or the card?) with PowerTune set to 50%, right?
- If the first questions answer is 'it's for the gpu only', how much, roughly, does rest of the board (memories etc) consume at stock clocks and volts?
- 208W is said to be the max consumption of the Hawaii GPU with PowerTune at 0%, right? It's the chip only, not the whole card?
- Adjusting volts via 3rd party software can't make the GPU (or card if the first question is cards consumption, not GPUs) go past the 312W max consumption of the Hawaii GPU (or the card?) with PowerTune set to 50%, right?
- If the first questions answer is 'it's for the gpu only', how much, roughly, does rest of the board (memories etc) consume at stock clocks and volts?
OpenGL guy
Veteran
290X has 1:8 DP.
Since you're here now, could you answer those couple questions I addressed to Dave?
Tech Report also thinks it's 1/4 rate DP (ctrl-F "double"), so bad assumption or PR miscommunication?
Or the Radeon version is detuned?
Or the Radeon version is detuned?
After thinking a bit about the priority thing with Powertune, it would actually be nice to offer a setting in the control center for what takes precedence over what.
Let's assume the power limit has always the highest priority, 727 MHz is some kind of base clock as mentiond in some reviews, and the 95°C are the max ceiling for a save operation. Then there are six possible permutations of the three parameters (clock speed, temperature and fan speed) on the priority list. Just lets get through all of them.
(i)
fan speed
temperature
clock
keeps fan speed always below limit, sacrifices clock for temperature
lowers clock when reaching temp target (or power limit),
if not enough (reaches 727 MHz), violate temp target,
speed up fan only when at 727MHz and 95°C
(ii)
fan speed
clock
temperature
keeps fan speed always below limit, sacrifice temperature for clock speed
temperature target basically just there for fan control, if temp target is violated no downclock
reduce clock only, if 95°C (or power limit) are hit,
speed up fan only, if 95°C cannot be held at 727 MHz
(iii)
clock
fan speed
temperature
always keeps clock as high as possible, fan speed more important than temperature
temperature target basically just there for fan control, if temp target is violated no downclock
increase fan above set limit, if absolute limit (95°C) is reached,
only reduce clock when power limit is reached or 95°C cannot be kept at 100% fan speed (which is quite unlikely)
(iv)
clock
temperature
fan speed
always keeps clock as high as possible, temperature more important than fan speed
basically disables the fan speed limit
temp target is for fan control only, no downclock if violated
only reduce clock when power limit is reached (or 95°C cannot be kept at 100% fan speed, which is very unlikely)
(v)
temperature
clock
fan speed
try to keep set temperature at all costs, clock more important than fan speed
basically disables the fan speed limit
if fan@100% is not enough to keep the temp limit at full clocks, reduce clock
only violate temperature limit if it cannot be held at 100% fan and 727MHz (unlikey if not set very low)
(vi)
temperature
fan speed
clock
try to keep set temperature at all costs, fan speed more important than clock
first reduce clock to keep temperature with the max allowed fan speed,
if 727 MHz are not low enough, increase fan speed to stay at the temperature target
only violate temperature limit if it cannot be held at 100% fan and 727MHz (unlikey if not set extremely low)
Let's assume the power limit has always the highest priority, 727 MHz is some kind of base clock as mentiond in some reviews, and the 95°C are the max ceiling for a save operation. Then there are six possible permutations of the three parameters (clock speed, temperature and fan speed) on the priority list. Just lets get through all of them.
(i)
fan speed
temperature
clock
keeps fan speed always below limit, sacrifices clock for temperature
lowers clock when reaching temp target (or power limit),
if not enough (reaches 727 MHz), violate temp target,
speed up fan only when at 727MHz and 95°C
(ii)
fan speed
clock
temperature
keeps fan speed always below limit, sacrifice temperature for clock speed
temperature target basically just there for fan control, if temp target is violated no downclock
reduce clock only, if 95°C (or power limit) are hit,
speed up fan only, if 95°C cannot be held at 727 MHz
(iii)
clock
fan speed
temperature
always keeps clock as high as possible, fan speed more important than temperature
temperature target basically just there for fan control, if temp target is violated no downclock
increase fan above set limit, if absolute limit (95°C) is reached,
only reduce clock when power limit is reached or 95°C cannot be kept at 100% fan speed (which is quite unlikely)
(iv)
clock
temperature
fan speed
always keeps clock as high as possible, temperature more important than fan speed
basically disables the fan speed limit
temp target is for fan control only, no downclock if violated
only reduce clock when power limit is reached (or 95°C cannot be kept at 100% fan speed, which is very unlikely)
(v)
temperature
clock
fan speed
try to keep set temperature at all costs, clock more important than fan speed
basically disables the fan speed limit
if fan@100% is not enough to keep the temp limit at full clocks, reduce clock
only violate temperature limit if it cannot be held at 100% fan and 727MHz (unlikey if not set very low)
(vi)
temperature
fan speed
clock
try to keep set temperature at all costs, fan speed more important than clock
first reduce clock to keep temperature with the max allowed fan speed,
if 727 MHz are not low enough, increase fan speed to stay at the temperature target
only violate temperature limit if it cannot be held at 100% fan and 727MHz (unlikey if not set extremely low)
Last edited by a moderator: