The GK110 vs Hawaii Bench-off thread has been lacking results that are run by the same person and in the same setup as well as actual game benchmarks in general. Since I ended up owning both an unlockable R9 290 and a reference GTX 780 more or less at the same time, I decided to make something out of it. I tried to get hold of as many even remotely relevant benchmarks that I could. The limiting factors were these: I didn't want to purchase anything new, I didn't have an optical drive in my test rig, I was only interested in ready-to-run benchmarks and I tried to avoid known CPU limited benchmarks. Excluded were Arkham series Batmans (have these on DVD), Crysis (the benchmark tool didn't work right), Crysis 2 (on DVD), GTA IV + Episodes (known CPU bottleneck and known to be bad for AMD), ARMA II (tried it, CPU limited to the extent of equal results with each GPU), Valley (thoroughly tested elsewhere and favours Nvidia) and 3DMark11 (tested elsewhere, somewhat CPU bound). What I ended up with are these:
Benchmarks:
Heaven Benchmark (standalone benchmark)
Sleeping Dogs (built-in benchmark)
Sniper Elite V2 (standalone)
Metro 2033 (built-in)
3DMark Fire Strike (standalone)
Hitman Absolution (built-in)
Alien vs Predator (standalone)
STALKER: Call of Pripyat (standalone)
Final Fantasy XIV: A Realm Reborn (standalone)
Mafia 2 (built-in)
Just Cause 2 (built-in)
Resident Evil 6 (standalone)
Passion Leads Army a.k.a PLA (standalone)
Test setup:
Club3D R9 290 (stock BIOS and Asus 290X BIOS with 2816 shaders reported by GPU-Z)
MSI Geforce GTX 780 (reference model)
Asus P9X79
i7-4820k @4.6GHz
16GB DDR3 2133MHz
Rosewill Armor Evo (3x140mm & 1x230mm fans)
Ambient temp ~19C
1920x1080 resolution
Win 8.1 Pro Preview (for a clean benchmarking installation)
Catalyst 13.12 WHQL
GeForce 331.82 WHQL
Afterburner 3.0.0 beta 17
GPU clocks
290&290X:
Stock voltage
Core clock: 1090MHz
Memory clock: 1300MHz
GTX 780:
Voltage +13mV (1.175V actual)
Core clock: 1176MHz (actual)
Memory clock: 1725MHz
Notes:
These are the confirmed stable 24/7 core clocks my cards would run. The preliminary stability testing was done by playing 1-2 hour sessions of Far Cry 3 and Sleeping Dogs.
For 290/290X Sleeping Dogs turned out to be the ultimate stability test: any clocks that could handle it would handle any other benchmark. 1100MHz was stable as in never crashed, but it was temp sensitive: going above 70 degrees I started to get random artifacts and I wasn't willing to ramp the fan up far enough to counter that. The same is true with both BIOSes, the stock 290 BIOS just takes a tiny bit longer to get to those temps. For the memory clocks, I didn't try anything beyond 1300MHz. My personal limiting factor when overclocking the reference Hawaii card was noise: the first thing I did was finding a fan setting that I could tolerate. That turned out ot be exactly 55% (same as 290X uber setting) which for my card meant 2950-2990 RPM. Not pleasant since the fan has a whining sound signature, but borderline tolerable even on silent parts of gaming with headphones on. The max temperatures can be seen in the monitoring data in the screen captures (mouseover at the highest temp).
For GTX 780 the preliminary testing turned out insufficient. 1189MHz core was very stable in both Sleeping Dogs and FC3 but once it was time to bench Final Fantasy XIV, I got a driver crash combined with fatal DX error and an unresponsive system. Twice in a row. Within two minutes. I dropped the core to 1176MHz and it looped without any further drama for half an hour. If someone wants to give their Kepler OC a good whooping, FFXIV seems like a nice no-cost test. For memory, I aimed for a 15% overclock that would take the memory bandwidth to the same level as I had with the 290/X. As it turns out, my 780 made it with no problems on either stability or performance. Again, I left it there and didn't venture any further. The main OC limiting factor with the 780 was the max power limit of 106%. I could have run most of the benchmarks with more voltage (1.2V) and thus higher stable clocks, but they would've only been stable in the sense of not crashing. Especially longer Sleeping Dogs sessions would throttle the clocks for hitting the power limit. I did get power limit hits even with 1176MHz@1.175V but they were of the "soft" kind that didn't throttle my clocks (can be seen in screen captures, I left the mouse cursor over the highest power limit reading). I used a fixed fan speed of 70% which equals to about the same RPM as with the Radeon (2930-3000 RPM). Special note: this is not equal in noise! The 780 fan is much, much quieter at 3000RPM. There was just absolutely no need for higher fan speeds as the temps remained below 70 degrees through all the tests (and gaming for that matter).
All cards were tested on fresh driver install after a complete driver removal with DDU 9.9.
All available graphics settings were maxed out except for Sniper Elite V2 which was run at stock settings. The driver settings were all in their stock settings.
There are a few benchmark results in the screen captures that are not included in the comparison. 3DMark was run for all the tests but I consider the Ice Storm and Cloud Gate to be irrelevant. The only STALKER: Call of Pripyat benchmark that I included is the sun shafts, which is the most demanding one. It was also the only one that I could get to run full screen from the very beginning, the benchmark tool is buggy in that way.
Finally, I'm not composing the result in charts for a reason. With the modern GPUs and their power balancing tech, it is -in my opinion- crucial to see the actual runtime clocks to get the whole big picture of the performance. This is why I've decided to show you each result with the monitoring data attached to it.
Enough notes yet? Let's get to the point then.
Results:
Heaven Benchmark
R9 290: 53.6 fps Warning: Spoiler! (Click to show)
R9 290X: 55.0 fps Warning: Spoiler! (Click to show)
GTX 780: 61.1 fps Warning: Spoiler! (Click to show)
Sleeping Dogs
R9 290: 70.1 fps Warning: Spoiler! (Click to show)
R9 290X: 72.6 fps Warning: Spoiler! (Click to show)
GTX 780: 72.8 fps Warning: Spoiler! (Click to show)
Sniper Elite V2 benchmark
R9 290: 68.2 fps Warning: Spoiler! (Click to show)
R9 290X: 71.3 fps Warning: Spoiler! (Click to show)
GTX 780: 85.3 fps Warning: Spoiler! (Click to show)
Metro 2033
R9 290: 53.98 fps Warning: Spoiler! (Click to show)
R9 290X: 54.66 fps Warning: Spoiler! (Click to show)
GTX 780: 53.24 fps Warning: Spoiler! (Click to show)
3DMark Fire Strike Graphics Score
R9 290: 11338 Warning: Spoiler! (Click to show)
R9 290X: 11645 Warning: Spoiler! (Click to show)
GTX 780: 11275 Warning: Spoiler! (Click to show)
Hitman Absolution
R9 290: 56.62 fps Warning: Spoiler! (Click to show)
R9 290X: 57.55 fps Warning: Spoiler! (Click to show)
GTX 780: 53.95 fps Warning: Spoiler! (Click to show)
Alien vs Predator
R9 290: 102.1 fps Warning: Spoiler! (Click to show)
R9 290X: 106.1 fps Warning: Spoiler! (Click to show)
GTX 780: 101.0 fps Warning: Spoiler! (Click to show)
STALKER: Call of Pripyat Sun Shafts benchmark
R9 290: 76 fps Warning: Spoiler! (Click to show)
R9 290X: 80 fps Warning: Spoiler! (Click to show)
GTX 780: 95.0 fps Warning: Spoiler! (Click to show)
Final Fantasy XIV: A Realm Reborn Benchmark score
R9 290: 14518 Warning: Spoiler! (Click to show)
R9 290X: 14752 Warning: Spoiler! (Click to show)
GTX 780: 15976 Warning: Spoiler! (Click to show)
Mafia 2 Benchmark (sorry about the notepad result presentation, alt-tabbing didn't work here so couldn't get the actual result screen in there)
R9 290: 135.5 fps Warning: Spoiler! (Click to show)
R9 290X: 139.3 fps Warning: Spoiler! (Click to show)
GTX 780: 138.4 fps Warning: Spoiler! (Click to show)
Just Cause 2 Concrete Jungle benchmark
R9 290: 107.49 fps Warning: Spoiler! (Click to show)
R9 290X: 108.87 fps Warning: Spoiler! (Click to show)
GTX 780: 87.88 fps Warning: Spoiler! (Click to show)
Resident Evil 6 benchmark score
290: 14160 Warning: Spoiler! (Click to show)
R9 290X: 14673 Warning: Spoiler! (Click to show)
GTX 780: 15848 Warning: Spoiler! (Click to show)
Passion Leads Army benchmark
R9 290: 105.8 fps Warning: Spoiler! (Click to show)
R9 290X: 105.6 fps Warning: Spoiler! (Click to show)
GTX 780: 143.0 fps Warning: Spoiler! (Click to show)
Conclusion:
I'm actually not going to make the final conclusion just yet. I haven't gone through the results as closely as I would have liked to, the second round of testing with the 780 took its toll on my available time. I'm just going to say that based on the numbers I posted, the average performance results go like this: 290 as a base line (100%), the 290X scores +2.7% and the GTX 780 +7.8%.
If someone actually has the time to look at the results and spots errors or irregularities, please let me know. I would like to have the data as solid as possible. What I can say is that Passion Leads Army has a rather big impact on the overall score. AMD drivers clearly aren't optimized for it, since it's the only benchmark where there's no difference between 290 and 290X. If I exclude PLA, the average performance difference from 290 is +2.9% for the 290X and +5.5% for the 780.
I will say this: the 290 is no longer mine. Not because of the performance as much as because of the noise. Also, I watched through all the benchmarks in effort to see the smoothness factor in action. There was absolutely no difference there for the naked (but rather sensitive) eye. Where there was a difference, however, was with the preliminary testing with Far Cry 3. I have come to understand that FC3 has been suffering from serious stuttering with various AMD setups all along, which was why I was surprised to see [H]ardOCP saying that they've had a better gaming experience with 290X CF than with Titan SLI. Now I know that the testing scenarios are very different (single card vs CF/SLI, single display vs multi monitor etc.) but I must say that to me, the difference was very obvious but in the different direction. My 290(X) had a very distinctive pause-like stutter every couple of seconds and general lag in some areas, whereas my 780 is practically free of stuttering. This was the one outlier on smoothness that I came across.
Edited by specopsFI - 12/21/13 at 4:11am
Benchmarks:
Heaven Benchmark (standalone benchmark)
Sleeping Dogs (built-in benchmark)
Sniper Elite V2 (standalone)
Metro 2033 (built-in)
3DMark Fire Strike (standalone)
Hitman Absolution (built-in)
Alien vs Predator (standalone)
STALKER: Call of Pripyat (standalone)
Final Fantasy XIV: A Realm Reborn (standalone)
Mafia 2 (built-in)
Just Cause 2 (built-in)
Resident Evil 6 (standalone)
Passion Leads Army a.k.a PLA (standalone)
Test setup:
Club3D R9 290 (stock BIOS and Asus 290X BIOS with 2816 shaders reported by GPU-Z)
MSI Geforce GTX 780 (reference model)
Asus P9X79
i7-4820k @4.6GHz
16GB DDR3 2133MHz
Rosewill Armor Evo (3x140mm & 1x230mm fans)
Ambient temp ~19C
1920x1080 resolution
Win 8.1 Pro Preview (for a clean benchmarking installation)
Catalyst 13.12 WHQL
GeForce 331.82 WHQL
Afterburner 3.0.0 beta 17
GPU clocks
290&290X:
Stock voltage
Core clock: 1090MHz
Memory clock: 1300MHz
GTX 780:
Voltage +13mV (1.175V actual)
Core clock: 1176MHz (actual)
Memory clock: 1725MHz
Notes:
These are the confirmed stable 24/7 core clocks my cards would run. The preliminary stability testing was done by playing 1-2 hour sessions of Far Cry 3 and Sleeping Dogs.
For 290/290X Sleeping Dogs turned out to be the ultimate stability test: any clocks that could handle it would handle any other benchmark. 1100MHz was stable as in never crashed, but it was temp sensitive: going above 70 degrees I started to get random artifacts and I wasn't willing to ramp the fan up far enough to counter that. The same is true with both BIOSes, the stock 290 BIOS just takes a tiny bit longer to get to those temps. For the memory clocks, I didn't try anything beyond 1300MHz. My personal limiting factor when overclocking the reference Hawaii card was noise: the first thing I did was finding a fan setting that I could tolerate. That turned out ot be exactly 55% (same as 290X uber setting) which for my card meant 2950-2990 RPM. Not pleasant since the fan has a whining sound signature, but borderline tolerable even on silent parts of gaming with headphones on. The max temperatures can be seen in the monitoring data in the screen captures (mouseover at the highest temp).
For GTX 780 the preliminary testing turned out insufficient. 1189MHz core was very stable in both Sleeping Dogs and FC3 but once it was time to bench Final Fantasy XIV, I got a driver crash combined with fatal DX error and an unresponsive system. Twice in a row. Within two minutes. I dropped the core to 1176MHz and it looped without any further drama for half an hour. If someone wants to give their Kepler OC a good whooping, FFXIV seems like a nice no-cost test. For memory, I aimed for a 15% overclock that would take the memory bandwidth to the same level as I had with the 290/X. As it turns out, my 780 made it with no problems on either stability or performance. Again, I left it there and didn't venture any further. The main OC limiting factor with the 780 was the max power limit of 106%. I could have run most of the benchmarks with more voltage (1.2V) and thus higher stable clocks, but they would've only been stable in the sense of not crashing. Especially longer Sleeping Dogs sessions would throttle the clocks for hitting the power limit. I did get power limit hits even with 1176MHz@1.175V but they were of the "soft" kind that didn't throttle my clocks (can be seen in screen captures, I left the mouse cursor over the highest power limit reading). I used a fixed fan speed of 70% which equals to about the same RPM as with the Radeon (2930-3000 RPM). Special note: this is not equal in noise! The 780 fan is much, much quieter at 3000RPM. There was just absolutely no need for higher fan speeds as the temps remained below 70 degrees through all the tests (and gaming for that matter).
All cards were tested on fresh driver install after a complete driver removal with DDU 9.9.
All available graphics settings were maxed out except for Sniper Elite V2 which was run at stock settings. The driver settings were all in their stock settings.
There are a few benchmark results in the screen captures that are not included in the comparison. 3DMark was run for all the tests but I consider the Ice Storm and Cloud Gate to be irrelevant. The only STALKER: Call of Pripyat benchmark that I included is the sun shafts, which is the most demanding one. It was also the only one that I could get to run full screen from the very beginning, the benchmark tool is buggy in that way.
Finally, I'm not composing the result in charts for a reason. With the modern GPUs and their power balancing tech, it is -in my opinion- crucial to see the actual runtime clocks to get the whole big picture of the performance. This is why I've decided to show you each result with the monitoring data attached to it.
Enough notes yet? Let's get to the point then.
Results:
Heaven Benchmark
R9 290: 53.6 fps Warning: Spoiler! (Click to show)
R9 290X: 55.0 fps Warning: Spoiler! (Click to show)
GTX 780: 61.1 fps Warning: Spoiler! (Click to show)
Sleeping Dogs
R9 290: 70.1 fps Warning: Spoiler! (Click to show)
R9 290X: 72.6 fps Warning: Spoiler! (Click to show)
GTX 780: 72.8 fps Warning: Spoiler! (Click to show)
Sniper Elite V2 benchmark
R9 290: 68.2 fps Warning: Spoiler! (Click to show)
R9 290X: 71.3 fps Warning: Spoiler! (Click to show)
GTX 780: 85.3 fps Warning: Spoiler! (Click to show)
Metro 2033
R9 290: 53.98 fps Warning: Spoiler! (Click to show)
R9 290X: 54.66 fps Warning: Spoiler! (Click to show)
GTX 780: 53.24 fps Warning: Spoiler! (Click to show)
3DMark Fire Strike Graphics Score
R9 290: 11338 Warning: Spoiler! (Click to show)
R9 290X: 11645 Warning: Spoiler! (Click to show)
GTX 780: 11275 Warning: Spoiler! (Click to show)
Hitman Absolution
R9 290: 56.62 fps Warning: Spoiler! (Click to show)
R9 290X: 57.55 fps Warning: Spoiler! (Click to show)
GTX 780: 53.95 fps Warning: Spoiler! (Click to show)
Alien vs Predator
R9 290: 102.1 fps Warning: Spoiler! (Click to show)
R9 290X: 106.1 fps Warning: Spoiler! (Click to show)
GTX 780: 101.0 fps Warning: Spoiler! (Click to show)
STALKER: Call of Pripyat Sun Shafts benchmark
R9 290: 76 fps Warning: Spoiler! (Click to show)
R9 290X: 80 fps Warning: Spoiler! (Click to show)
GTX 780: 95.0 fps Warning: Spoiler! (Click to show)
Final Fantasy XIV: A Realm Reborn Benchmark score
R9 290: 14518 Warning: Spoiler! (Click to show)
R9 290X: 14752 Warning: Spoiler! (Click to show)
GTX 780: 15976 Warning: Spoiler! (Click to show)
Mafia 2 Benchmark (sorry about the notepad result presentation, alt-tabbing didn't work here so couldn't get the actual result screen in there)
R9 290: 135.5 fps Warning: Spoiler! (Click to show)
R9 290X: 139.3 fps Warning: Spoiler! (Click to show)
GTX 780: 138.4 fps Warning: Spoiler! (Click to show)
Just Cause 2 Concrete Jungle benchmark
R9 290: 107.49 fps Warning: Spoiler! (Click to show)
R9 290X: 108.87 fps Warning: Spoiler! (Click to show)
GTX 780: 87.88 fps Warning: Spoiler! (Click to show)
Resident Evil 6 benchmark score
290: 14160 Warning: Spoiler! (Click to show)
R9 290X: 14673 Warning: Spoiler! (Click to show)
GTX 780: 15848 Warning: Spoiler! (Click to show)
Passion Leads Army benchmark
R9 290: 105.8 fps Warning: Spoiler! (Click to show)
R9 290X: 105.6 fps Warning: Spoiler! (Click to show)
GTX 780: 143.0 fps Warning: Spoiler! (Click to show)
Conclusion:
I'm actually not going to make the final conclusion just yet. I haven't gone through the results as closely as I would have liked to, the second round of testing with the 780 took its toll on my available time. I'm just going to say that based on the numbers I posted, the average performance results go like this: 290 as a base line (100%), the 290X scores +2.7% and the GTX 780 +7.8%.
If someone actually has the time to look at the results and spots errors or irregularities, please let me know. I would like to have the data as solid as possible. What I can say is that Passion Leads Army has a rather big impact on the overall score. AMD drivers clearly aren't optimized for it, since it's the only benchmark where there's no difference between 290 and 290X. If I exclude PLA, the average performance difference from 290 is +2.9% for the 290X and +5.5% for the 780.
I will say this: the 290 is no longer mine. Not because of the performance as much as because of the noise. Also, I watched through all the benchmarks in effort to see the smoothness factor in action. There was absolutely no difference there for the naked (but rather sensitive) eye. Where there was a difference, however, was with the preliminary testing with Far Cry 3. I have come to understand that FC3 has been suffering from serious stuttering with various AMD setups all along, which was why I was surprised to see [H]ardOCP saying that they've had a better gaming experience with 290X CF than with Titan SLI. Now I know that the testing scenarios are very different (single card vs CF/SLI, single display vs multi monitor etc.) but I must say that to me, the difference was very obvious but in the different direction. My 290(X) had a very distinctive pause-like stutter every couple of seconds and general lag in some areas, whereas my 780 is practically free of stuttering. This was the one outlier on smoothness that I came across.
Edited by specopsFI - 12/21/13 at 4:11am










