TomsHardware and Techreport have been holding AMD's feet to the fire recently over perceived performance differences between stock retail cards and AMD press samples sent out to news sites. Following reports of lower than anticipated performance in retail R9 290X both sites put non-press cards through their paces and found unanticipated differences between the clock speeds of retail and press cards, running on the order of 6% in some places. Although to some end-users that would seem to be a superficial difference, at the very high end consumers have a right to expect a certain performance level, especially when the difference between competing cards may also be on the order of single-digit values.
The R9 290X is already known to be more of a slave to the so-called 'Silicon Lottery' than its competitors due to operating at variable rather than fixed frequencies (hence the 'up to 1GHz' statements in reference materials). GPU frequencies realised are based chiefly on temperatures and fan profiles, which is the province of AMD's PowerTune utility; if a GPU can run at higher frequencies for given fan speeds and temp. maximums, it will do so. Equally it will throttle frequencies down if already at the target temp. and maximum fan speed.
NVIDIA cards also have some inherent performance variation due to GPU BOOST, but will always ramp above a fixed baseline frequency stated in reference material. In this respect therefore they have a certain level of guaranteed performance which an end-user can expect (although review sites have also criticised NVIDIA for not including a 'disable GPU BOOST mode' to rigorously assess this level).
The R9 290X potentially being aggressively thermally limited isn't helped by the loud reference cooler, dual-mode BIOS and lack of partner cooler designs. Note though that it is performance which is crucial, not frequencies; it doesn't matter if frequencies frequently change so long as framerate expectations are met.
In response to criticism AMD released the following statement on Friday:
Based on feedback from the enthusiast community, we’ve implemented an all new PowerTune mechanism in the AMD Radeon R9 290 series that exploits the full capability of the individual GPUs rather than clamping performance to a least-common-denominator type of capability level. This has the advantage of improving overall performance but does result in some degree of performance variability. These changes will also result in some degree of run-to-run test variability based on environmental and operational conditions in un-controlled test environments.
The range of performance differential is not expected to meaningfully change the user experience but we’ve taken note of recent reports that the degree of variability is higher than expected. Reasonably we would expect the variability to occur both above and below the performance of the press samples, however it appears that most reported performances are biased towards the low side. We are actively investigating these reports and we will update when we have completed our investigation.
It is important to note that is it to be expected that the more thermally limited the setting the more variation can naturally occur. AMD Radeon R9 290X, by default, ships with two default BIOS settings for different acoustic levels and the "Uber" setting was designed to limit the level of thermal throttling and comparisons in this test mode are more consistent. Furthermore, AMD have implemented comprehensive PowerTune controls within the OverDrive panel of Catalyst Control Center and users can tweak to their own desired balance between acoustics and performance.
The range of performance differential is not expected to meaningfully change the user experience but we’ve taken note of recent reports that the degree of variability is higher than expected. Reasonably we would expect the variability to occur both above and below the performance of the press samples, however it appears that most reported performances are biased towards the low side. We are actively investigating these reports and we will update when we have completed our investigation.
It is important to note that is it to be expected that the more thermally limited the setting the more variation can naturally occur. AMD Radeon R9 290X, by default, ships with two default BIOS settings for different acoustic levels and the "Uber" setting was designed to limit the level of thermal throttling and comparisons in this test mode are more consistent. Furthermore, AMD have implemented comprehensive PowerTune controls within the OverDrive panel of Catalyst Control Center and users can tweak to their own desired balance between acoustics and performance.
Not an admittance of fault, but an acknowledgement that investigation is warranted.
Our review of the R9 290X is available here.
Sources: TechReport, TomsHardware, PCPER
