Brent_Justice
Moderator
- Joined
- Apr 17, 2000
- Messages
- 17,755
Intro
This topic is a pet peeve of mine, but is also relevant in the soon to come AMD Nano reviews. It also applies to all NVIDIA and AMD video card testing with today's latest power management and GPU Boost features. I am talking about, warming up your video card prior to testing performance. Let's talk about that.
How GPU Clock Speed Varies
In today's modern GPUs the clock speed is in essence a variable. The clock speed in AMD and NVIDIA GPUs is dynamic and able to change based on many factors including thermals, power and even down to small differences in binning. We aren't going to get specific into how that works today, today I want to talk to you about testing video cards, and the right way to do it knowing how GPU clock speed works in modern GPUs.
For NVIDIA, this is called GPU Boost. You start with a base and boost clock, and the GPU dynamically changes the clock speed upwards as needed. However, it is also capable of going the opposite direction, and lowering the clock speed if needed. Typically, in gaming, you will find the real-world in-game frequency higher than the boost clock. For AMD, it works differently.
AMD provides an "Up To" clock speed rating. This is what the GPU is capable of reaching, but it may often not. Instead, the clock speed is below the rated clock speed, and has the ability to reach up to it, if possible. These are two very different ways of looking at clock speed, but just note that in both methods the clock speed can change over time while gaming.
Clock Speed over Time
This is the heart of the matter. The Clock Speed can change over time while gaming. For NVIDIA, this means a lower boost clock speed. For AMD, this can mean the clock speed is under the rated "up to" clock speed. This time is also variable, it could be after 5 minutes of gaming, or after 15 or 30 minutes of gaming. The reason why it changes is due to the thermal properties, mainly. As you play games over time, the video card heats up, the components in your system heats up, the ambient air temperature heats up. This heating up, can cause your clock speed, to go down over time.
For example, your video card may start out at 950MHz for the first 5 minutes of playing a game. However, after 15 minutes it may drop to 900MHz. After 30 minutes it may drop to a much lower 800MHz. This means you are losing performance over time, but it also means that if you tested for performance at the 5 minute mark, versus the 30 minute mark, your results would be very different.
In modern GPUs, it is IMPERATIVE to test performance after the video card as "warmed up" after say 15-30 minutes of playing a game. Else, you are only reporting the higher performance, and ignoring the real-world performance after playing games for a while.
How to Test
This is how GPU testing should be done:
A.) The very first thing should be to test and find out what the real-world in-game clock frequency is while gaming over time, not the base clock, not the boost clock. This requires usage of a third party utility like Afterburner. In Afterburner, you can enable an OSD on your screen that shows the actual GPU clock frequency. The first thing you need to do is play a game for 30 minutes and monitor this GPU frequency. You need to see what it starts off as, and how it changes over time, and the final clock speed it settles to consistently. It is the consistent clock speed you are looking for. This then needs to be reported in the review. This is the clock speed you need to test at.
B.) Then, when you do your testing, benchmarks, gaming, whatever, do it after this time period of playing games that have warmed up your GPU to its consistent real-world clock speed. This then results in your performance indicating the real-world performance regular gamers who install this video card, will get while gaming.
IF you do a short benchmark, say 5 minutes, this will show an inflated result. 5 minutes will not be long enough to warm up the card to its real-world frequency. 5 minutes will show you the high end 950MHz performance, but will not show you the performance when after gaming for 15-30 minutes it drops to 800MHz. The benchmark isn't long enough. Therefore, you must also have a test that lasts long enough to show the real-world performance. Testing from a "cold start" will always show inflated results.
Summary
Therefore, A.) the reviewer must find out what the dynamic clock speeds are and report the actual clock speed after a long period of gaming, B.) test video cards after this warm up period, C.) use tests that are long enough to realize these real-world clock frequencies and never do a cold start test.
This is how "FAIR" review sites, test cards. This is how we, at HardOCP, as a "FAIR" website, test video cards. This is how we would have tested Nano.
As you look forward to reviews this week, keep this topic in mind.
This topic is a pet peeve of mine, but is also relevant in the soon to come AMD Nano reviews. It also applies to all NVIDIA and AMD video card testing with today's latest power management and GPU Boost features. I am talking about, warming up your video card prior to testing performance. Let's talk about that.
How GPU Clock Speed Varies
In today's modern GPUs the clock speed is in essence a variable. The clock speed in AMD and NVIDIA GPUs is dynamic and able to change based on many factors including thermals, power and even down to small differences in binning. We aren't going to get specific into how that works today, today I want to talk to you about testing video cards, and the right way to do it knowing how GPU clock speed works in modern GPUs.
For NVIDIA, this is called GPU Boost. You start with a base and boost clock, and the GPU dynamically changes the clock speed upwards as needed. However, it is also capable of going the opposite direction, and lowering the clock speed if needed. Typically, in gaming, you will find the real-world in-game frequency higher than the boost clock. For AMD, it works differently.
AMD provides an "Up To" clock speed rating. This is what the GPU is capable of reaching, but it may often not. Instead, the clock speed is below the rated clock speed, and has the ability to reach up to it, if possible. These are two very different ways of looking at clock speed, but just note that in both methods the clock speed can change over time while gaming.
Clock Speed over Time
This is the heart of the matter. The Clock Speed can change over time while gaming. For NVIDIA, this means a lower boost clock speed. For AMD, this can mean the clock speed is under the rated "up to" clock speed. This time is also variable, it could be after 5 minutes of gaming, or after 15 or 30 minutes of gaming. The reason why it changes is due to the thermal properties, mainly. As you play games over time, the video card heats up, the components in your system heats up, the ambient air temperature heats up. This heating up, can cause your clock speed, to go down over time.
For example, your video card may start out at 950MHz for the first 5 minutes of playing a game. However, after 15 minutes it may drop to 900MHz. After 30 minutes it may drop to a much lower 800MHz. This means you are losing performance over time, but it also means that if you tested for performance at the 5 minute mark, versus the 30 minute mark, your results would be very different.
In modern GPUs, it is IMPERATIVE to test performance after the video card as "warmed up" after say 15-30 minutes of playing a game. Else, you are only reporting the higher performance, and ignoring the real-world performance after playing games for a while.
How to Test
This is how GPU testing should be done:
A.) The very first thing should be to test and find out what the real-world in-game clock frequency is while gaming over time, not the base clock, not the boost clock. This requires usage of a third party utility like Afterburner. In Afterburner, you can enable an OSD on your screen that shows the actual GPU clock frequency. The first thing you need to do is play a game for 30 minutes and monitor this GPU frequency. You need to see what it starts off as, and how it changes over time, and the final clock speed it settles to consistently. It is the consistent clock speed you are looking for. This then needs to be reported in the review. This is the clock speed you need to test at.
B.) Then, when you do your testing, benchmarks, gaming, whatever, do it after this time period of playing games that have warmed up your GPU to its consistent real-world clock speed. This then results in your performance indicating the real-world performance regular gamers who install this video card, will get while gaming.
IF you do a short benchmark, say 5 minutes, this will show an inflated result. 5 minutes will not be long enough to warm up the card to its real-world frequency. 5 minutes will show you the high end 950MHz performance, but will not show you the performance when after gaming for 15-30 minutes it drops to 800MHz. The benchmark isn't long enough. Therefore, you must also have a test that lasts long enough to show the real-world performance. Testing from a "cold start" will always show inflated results.
Summary
Therefore, A.) the reviewer must find out what the dynamic clock speeds are and report the actual clock speed after a long period of gaming, B.) test video cards after this warm up period, C.) use tests that are long enough to realize these real-world clock frequencies and never do a cold start test.
This is how "FAIR" review sites, test cards. This is how we, at HardOCP, as a "FAIR" website, test video cards. This is how we would have tested Nano.
As you look forward to reviews this week, keep this topic in mind.