Yes, you've mentioned a few that would benefit. You haven't mentioned any new ones though which is your claim. Crysis 2 and Metro 2033 aren't new games. BF3 MP would benefit more from a CPU, SP more from a GPU but the point is moot since he isnt playing BF3. So far the games the OP is interested in would benefit more from a CPU. MP3 is pure guess work on your part as to what it will benefit from more so I'm going to take a guess too and say it'll benefit more from a CPU. If he said he was looking for the best possible performance from Crysis 2 and Metro 2033 I'd be in full agreement with you.
Yes, you've mentioned a few that would benefit. You haven't mentioned any new ones though which is your claim. Crysis 2 and Metro 2033 aren't new games. BF3 MP would benefit more from a CPU, SP more from a GPU but the point is moot since he isnt playing BF3. So far the games the OP is interested in would benefit more from a CPU. MP3 is pure guess work on your part as to what it will benefit from more so I'm going to take a guess too and say it'll benefit more from a CPU. If he said he was looking for the best possible performance from Crysis 2 and Metro 2033 I'd be in full agreement with you.
cmdrdredd
Lifer
- Dec 12, 2001
- 27,052
- 357
- 126
cmdrdredd
Lifer
- Dec 12, 2001
- 27,052
- 357
- 126
OP: I want to apologize for this taking a while. I was doing some overclock testing with the GTX 670 and I found that when I had EVGA PrecisionX or MSI Afterburner on, some games and benchmarks would not run. I'm not the only one, EVGA forums pulled up similar things. So I don't know how I can overclock this GPU without using those programs and without causing a crash. So all of my results are done with stock GTX 670 clock speeds. I do have the Superclocked version which bumps the core and memory clocks up a bit at stock though.
All tests are at 1920x1200 4xAA/16xAF with maximum graphical detail settings on my system in my sig. The exception is Crysis 2 where I turned AA off completely and Far Cry 2 where I enabled 8xAA. Crysis 2 did run using DX11 and the high resolution textures. Also 3dMark where I cannot change the presets because you have to buy the pro version. All games used the built in benchmark programs except for Crysis 2 for which I used the Adrenaline Benchmark Application. Batman: Arkham City was run without Physx on in order to give a true baseline. I could not get the new drivers for the GTX 670 to install when I had the 9600GT card installed for physx. I did not test Battlefield 3 because there is no really good way to create a benchmark that is repeatable. Performance varies online and I don't find playing through a section in FRAPS presents the best guage of the game's performance. I will say that while the FPS didn't jump significantly, I could turn on MSAA without issue when I had the GTX 670 installed.
3DMark 11
6950 = 5246
GTX670 = 7627
3DMark Vantage
6950 = 19178
GTX 670 = 30714
Far Cry 2
6950 = 69.83fps Min/87.24fps Avg/123.19fps Max
GTX 670 = 74.89fps Min/106.76fps Avg/167.30fps Max
Crysis
6950 = CPU Bench 36.045fps GPU Bench 33.778fps
GTX 670 = CPU Bench 48.827fps GPU Bench 46.867fps
Crysis 2
6950 = 16.2fps Min/32.6fps Avg
GTX 670 = 25.2fps Min/58.7fps Avg
Dirt 3
6950 = 45.17fps Min/54.88fps Avg
GTX 670 = 55.75fps Min/75.56fps Avg
Batman: AC
6950 = 4fps Min/59fps Avg/81fps Max
GTX 670 = 19fps Min/83fps Avg/119fps Max
What we can gather from these results is the following. The minimum FPS jumps pretty significantly. It makes most games more playable and where some games were unplayable on a HD6950, they are playable with the GTX670. You can see from Crysis 2 for example that the high levels of tessellation cause a severe strain on the 6950. It should be noted that the Catalyst Control Panel has a tessellation slider and I did not use it. I left it on "AMD Optimized". In Batman Arkham City there is a point where both cards recorded very low FPS. This is due to the benchmark loading up and during game play neither card would drop to the minimum fps shown above. While I don't have Metro 2033 I can only assume you'd see the same type of increase you see with Crysis 2. Unplayable becomes playable etc. So while the averages didn't go up too drastically, they went up enough to be noticeable during game play.
What do we learn about the q9550 overclocked paired with a GTX 670? Lets look at Anandtech's own benchmarks of the GTX 670 using a 4.3Ghz i7-3960x compared to a couple of my results.
My dirt 3 score of 75.56fps is very much below their score of 123.2fps. There is still plenty of room to push the frames with a stronger CPU. It should be noted that if you run a 60Hz monitor, you would likely have vsync on and you won't see such high framerates. The minimums are even more telling, 55.75fps vs 102.8fps. So again we have a good boost from the CPU. If you take my overclocked 6950 results (unlocked shaders) and look at Dirt 3, I scored a 54.88avg and 45.17min. Anandtech's score with the i7 CPU was 72.2fps avg and 57.5fps min with an HD6970(a bit faster than my 6950 but you get a decent idea). So you can see that a CPU really does help, more than a GPU in this game.
Lets look at Batman. I have to assume their score in the charts are averages. With a GTX 670 on a 4.3Ghz i7 they are getting 94fps while I score 83fps. There is some room here for improvement with a stronger CPU. Maybe not as much as some people would have you believe but it's there. Their 6970 scored 60fps and my 6950 scored 59fps average.
Guru3D tests a couple other games and benchmarks we can use to compare to my scores. They use an i7 at 3.7Ghz. In FarCry 2 they scored 129fps average and I scored 107fps. So again we see there is some room for improvement. Their 6970 scored 83fps while my 6950 scored 87fps. Using this we can see that the biggest improvement comes from GPU, but the CPU would cause it to be a more drastic improvement when you change the GPU out as well. This game is older, but it seems as if the engine used scaled well with GPU power.
3DMark is a synthetic test and isn't a good judge of true gaming performance but it does tell you the potential of a GPU when paired with a specific CPU. Guru3D scored 28865 with GTX 670 in 3DMark Vantage and my score was 30714. Did 100Mhz give that much variation? Was there something else going on? This is why I don't fully trust this benchmark but I included it for comparison anyway. With a 6970 they scored 21419 and I scored 19178 which seems to be in line with what I expected. In 3DMark 11 their GTX 670 score of 8691 bests my 7627 score pretty handily. Here we see where a CPU, even clocked slightly lower can significantly affect performance. Their 6970 scored 5449 while my score of 5246 with a 6950 is showing less of a difference.
Guru3D also tested Crysis 2 which will be the final compare I give. They scored 61fps with the GTX 670 while I scored 58.7fps. Their 6970 scores 36fps while I get 32.6fps. So for this game you can see the GPU is the bottleneck for sure. The CPU does impact performance slightly but not significantly.
Overall I'd say that judging the results I obtained the GPU helps slightly more than a CPU would. That said, the HD6950 would show a good increase in fps when paired with a stronger CPU. The GTX670 was a good upgrade in my opinion because the settings that become playable increases. I can now turn on AA and still have good frame rates at my resolution. You can see in some games like Batman that the GPU is basically maxed out with work and you have to change that to get a significant jump. Just changing the CPU doesn't do as much. So a Q9550 @ 3.8Ghz does bottleneck the GTX670 but for some games the GPU changes the experience more than the CPU does. It should be noted that along with a new CPU you get higher memory bandwidth on the Motherboard because of the switch to DDR3 as well as the benefit of SATA 6Gbps and newer PCIe revisions that offer a little more bandwidth. PCIe 1.1 on the P35 chipset loses roughly 5% performance when compared to a motherboard running PCIe 3.0 and a GTX680 (similar enough to 670 for comparison). I hope this information proves useful and I am sorry for the long post. Feel free to post questions if you have them and I'll try to help.
All tests are at 1920x1200 4xAA/16xAF with maximum graphical detail settings on my system in my sig. The exception is Crysis 2 where I turned AA off completely and Far Cry 2 where I enabled 8xAA. Crysis 2 did run using DX11 and the high resolution textures. Also 3dMark where I cannot change the presets because you have to buy the pro version. All games used the built in benchmark programs except for Crysis 2 for which I used the Adrenaline Benchmark Application. Batman: Arkham City was run without Physx on in order to give a true baseline. I could not get the new drivers for the GTX 670 to install when I had the 9600GT card installed for physx. I did not test Battlefield 3 because there is no really good way to create a benchmark that is repeatable. Performance varies online and I don't find playing through a section in FRAPS presents the best guage of the game's performance. I will say that while the FPS didn't jump significantly, I could turn on MSAA without issue when I had the GTX 670 installed.
3DMark 11
6950 = 5246
GTX670 = 7627
3DMark Vantage
6950 = 19178
GTX 670 = 30714
Far Cry 2
6950 = 69.83fps Min/87.24fps Avg/123.19fps Max
GTX 670 = 74.89fps Min/106.76fps Avg/167.30fps Max
Crysis
6950 = CPU Bench 36.045fps GPU Bench 33.778fps
GTX 670 = CPU Bench 48.827fps GPU Bench 46.867fps
Crysis 2
6950 = 16.2fps Min/32.6fps Avg
GTX 670 = 25.2fps Min/58.7fps Avg
Dirt 3
6950 = 45.17fps Min/54.88fps Avg
GTX 670 = 55.75fps Min/75.56fps Avg
Batman: AC
6950 = 4fps Min/59fps Avg/81fps Max
GTX 670 = 19fps Min/83fps Avg/119fps Max
What we can gather from these results is the following. The minimum FPS jumps pretty significantly. It makes most games more playable and where some games were unplayable on a HD6950, they are playable with the GTX670. You can see from Crysis 2 for example that the high levels of tessellation cause a severe strain on the 6950. It should be noted that the Catalyst Control Panel has a tessellation slider and I did not use it. I left it on "AMD Optimized". In Batman Arkham City there is a point where both cards recorded very low FPS. This is due to the benchmark loading up and during game play neither card would drop to the minimum fps shown above. While I don't have Metro 2033 I can only assume you'd see the same type of increase you see with Crysis 2. Unplayable becomes playable etc. So while the averages didn't go up too drastically, they went up enough to be noticeable during game play.
What do we learn about the q9550 overclocked paired with a GTX 670? Lets look at Anandtech's own benchmarks of the GTX 670 using a 4.3Ghz i7-3960x compared to a couple of my results.
My dirt 3 score of 75.56fps is very much below their score of 123.2fps. There is still plenty of room to push the frames with a stronger CPU. It should be noted that if you run a 60Hz monitor, you would likely have vsync on and you won't see such high framerates. The minimums are even more telling, 55.75fps vs 102.8fps. So again we have a good boost from the CPU. If you take my overclocked 6950 results (unlocked shaders) and look at Dirt 3, I scored a 54.88avg and 45.17min. Anandtech's score with the i7 CPU was 72.2fps avg and 57.5fps min with an HD6970(a bit faster than my 6950 but you get a decent idea). So you can see that a CPU really does help, more than a GPU in this game.
Lets look at Batman. I have to assume their score in the charts are averages. With a GTX 670 on a 4.3Ghz i7 they are getting 94fps while I score 83fps. There is some room here for improvement with a stronger CPU. Maybe not as much as some people would have you believe but it's there. Their 6970 scored 60fps and my 6950 scored 59fps average.
Guru3D tests a couple other games and benchmarks we can use to compare to my scores. They use an i7 at 3.7Ghz. In FarCry 2 they scored 129fps average and I scored 107fps. So again we see there is some room for improvement. Their 6970 scored 83fps while my 6950 scored 87fps. Using this we can see that the biggest improvement comes from GPU, but the CPU would cause it to be a more drastic improvement when you change the GPU out as well. This game is older, but it seems as if the engine used scaled well with GPU power.
3DMark is a synthetic test and isn't a good judge of true gaming performance but it does tell you the potential of a GPU when paired with a specific CPU. Guru3D scored 28865 with GTX 670 in 3DMark Vantage and my score was 30714. Did 100Mhz give that much variation? Was there something else going on? This is why I don't fully trust this benchmark but I included it for comparison anyway. With a 6970 they scored 21419 and I scored 19178 which seems to be in line with what I expected. In 3DMark 11 their GTX 670 score of 8691 bests my 7627 score pretty handily. Here we see where a CPU, even clocked slightly lower can significantly affect performance. Their 6970 scored 5449 while my score of 5246 with a 6950 is showing less of a difference.
Guru3D also tested Crysis 2 which will be the final compare I give. They scored 61fps with the GTX 670 while I scored 58.7fps. Their 6970 scores 36fps while I get 32.6fps. So for this game you can see the GPU is the bottleneck for sure. The CPU does impact performance slightly but not significantly.
Overall I'd say that judging the results I obtained the GPU helps slightly more than a CPU would. That said, the HD6950 would show a good increase in fps when paired with a stronger CPU. The GTX670 was a good upgrade in my opinion because the settings that become playable increases. I can now turn on AA and still have good frame rates at my resolution. You can see in some games like Batman that the GPU is basically maxed out with work and you have to change that to get a significant jump. Just changing the CPU doesn't do as much. So a Q9550 @ 3.8Ghz does bottleneck the GTX670 but for some games the GPU changes the experience more than the CPU does. It should be noted that along with a new CPU you get higher memory bandwidth on the Motherboard because of the switch to DDR3 as well as the benefit of SATA 6Gbps and newer PCIe revisions that offer a little more bandwidth. PCIe 1.1 on the P35 chipset loses roughly 5% performance when compared to a motherboard running PCIe 3.0 and a GTX680 (similar enough to 670 for comparison). I hope this information proves useful and I am sorry for the long post. Feel free to post questions if you have them and I'll try to help.
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
Sweet benches by cmdrdredd.
Q9550 @ 3.8ghz + GTX670 > SB/IVB + HD6950 OC in GPU demanding games (which is most of the reason people buy a new GPU in the first place).
The exceptions are MMOs, strategy games and games such as Resident Evil 5, GTAIV, Dirt 3, etc. that benefit from a quad-core CPU (but Q9550 is a quad so this isn't as big of an issue like we are dealing with an E8400 here). But a lot of the times one can enable higher levels of AA or enable super-sampling shifting the bottleneck to the GPU. Q9550 @ 3.5ghz won't extract maximum value from a GTX670 but in games such as Batman AC, Metro 2033, Crysis 1/Warhead, Crysis 2, BF3 with 4xMSAA, HD6950 is choking.
Also, while Q9550 @ 3.5ghz might not provide the same minimums in BF3 multiplayer, HD6950 is lucky to get 30 fps avg in BF3:
The answer is: it depends on the game. Q9550 @ 3.5ghz is roughly equivalent to Core i7 930 / 860 @ 3.0ghz. It's not that bad. Nehalem had about a 15-17.5% increase in IPC over 45nm C2Q.
I say get the GPU upgrade and get a full platform upgrade with Haswell. If the GPU upgrade solves your problems, you just saved yourself from getting IVB/SB.
Another quick way to test if the Q9550 @ 3.5ghz is a bottleneck in specific games is to lower its speed to 2.8ghz. Then perform the same test with Q9550 @ 3.5ghz and HD6950 at 0AA or 8AA. If the FPS drop with a CPU @ 2.8ghz is more severe, the game needs a much faster CPU. If HD6950 is choking with AA, then a GPU upgrade would be more beneficial.
Overall, with high level of tessellation and deferred MSAA in modern games, there is almost no chance that SB / IVB + 6950 will beat Q9550 @ 3.5ghz + GTX670 in FPS, racing, action adventure and RPG games @ 1920x1080.
Q9550 @ 3.8ghz + GTX670 > SB/IVB + HD6950 OC in GPU demanding games (which is most of the reason people buy a new GPU in the first place).
The exceptions are MMOs, strategy games and games such as Resident Evil 5, GTAIV, Dirt 3, etc. that benefit from a quad-core CPU (but Q9550 is a quad so this isn't as big of an issue like we are dealing with an E8400 here). But a lot of the times one can enable higher levels of AA or enable super-sampling shifting the bottleneck to the GPU. Q9550 @ 3.5ghz won't extract maximum value from a GTX670 but in games such as Batman AC, Metro 2033, Crysis 1/Warhead, Crysis 2, BF3 with 4xMSAA, HD6950 is choking.
Also, while Q9550 @ 3.5ghz might not provide the same minimums in BF3 multiplayer, HD6950 is lucky to get 30 fps avg in BF3:
The answer is: it depends on the game. Q9550 @ 3.5ghz is roughly equivalent to Core i7 930 / 860 @ 3.0ghz. It's not that bad. Nehalem had about a 15-17.5% increase in IPC over 45nm C2Q.
I say get the GPU upgrade and get a full platform upgrade with Haswell. If the GPU upgrade solves your problems, you just saved yourself from getting IVB/SB.
Another quick way to test if the Q9550 @ 3.5ghz is a bottleneck in specific games is to lower its speed to 2.8ghz. Then perform the same test with Q9550 @ 3.5ghz and HD6950 at 0AA or 8AA. If the FPS drop with a CPU @ 2.8ghz is more severe, the game needs a much faster CPU. If HD6950 is choking with AA, then a GPU upgrade would be more beneficial.
Overall, with high level of tessellation and deferred MSAA in modern games, there is almost no chance that SB / IVB + 6950 will beat Q9550 @ 3.5ghz + GTX670 in FPS, racing, action adventure and RPG games @ 1920x1080.
Last edited:
When a 6950 has trouble breaking 30 fps, you disable HBAO/SSAO or whatever they call it and enjoy the game at 50fps+. When a Q9550 has trouble breaking 30 fps, your Q9550 has trouble breaking 30 fps. 
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
When a 6950 has trouble breaking 30 fps, you disable HBAO/SSAO or whatever they call it and enjoy the game at 50fps+. When a Q9550 has trouble breaking 30 fps, your Q9550 has trouble breaking 30 fps.
But how many modern games are there outside of Shogun 2 where Q9550 @ 3.5ghz has trouble breaking 30 fps?
I'll start with a stock Q9550:
69 fps in Deus Ex Human Revolution
101 fps in Hard Reset
46 fps F1 2011
56 fps in BF3
127 fps in Anno 2070
60 fps in Sonic Generations
82 fps in COD MW3
40 fps in SKYRIM (before the CPU patch that fixed this)
.... I could keep going but it would be pointless to add Dragon Age 2, Bullet Storm, Witcher 2, Crysis 2, Batman AC, Dirt 3 since we know they are more GPU limited.
Also, the examples I linked use Q9550 @ 2.8ghz. OP's Q9550 is at 3.5ghz.
A GPU upgrade is far better for the OP than a CPU upgrade because HD6950 is the bigger bottleneck in the majority of games. If we turn off some graphical features, then the OP might as well ride it for another year with this system since in that case HD6950 OCed is good enough
With a Q9550 @ 3.5ghz ~ Core i7 870 2.93ghz. That CPU provides 85% of the performance of $1000 3960X with a GTX680 at 1080P.
The funny part is Q9550 @ 3.5ghz is still better for games than any stock clocked AMD CPU out for sale today.
Last edited:
Your 56 fps BF3 is for SP not MP
My q6600 @ 3.4 just bearly managed a 30 fps. Min fps would obviously dip below that, and did so quote often, especially on 64 player maps. A 9550 would be marginally better but still have low minimums.
How many games can a 6950 not break 30 fps on? Zero, because those settings can be manipulated.
Bottom line is if he wants more eye candy get a GPU if you want smoother gameplay get a CPU. Like I said there are exactly 0 games that a 6950 cannot provide perfectly smooth game play in) the same cannot be said for the CPU. If the OP is trying to turn up the eye candy and is having trouble doing so with adequate performance, GPU is what he should get. If the OP is looking for more performance regardless of his settings, meaning even at lower settings the performance isnt where he wants it, CPU will do that.
My q6600 @ 3.4 just bearly managed a 30 fps. Min fps would obviously dip below that, and did so quote often, especially on 64 player maps. A 9550 would be marginally better but still have low minimums.
How many games can a 6950 not break 30 fps on? Zero, because those settings can be manipulated.
Bottom line is if he wants more eye candy get a GPU if you want smoother gameplay get a CPU. Like I said there are exactly 0 games that a 6950 cannot provide perfectly smooth game play in) the same cannot be said for the CPU. If the OP is trying to turn up the eye candy and is having trouble doing so with adequate performance, GPU is what he should get. If the OP is looking for more performance regardless of his settings, meaning even at lower settings the performance isnt where he wants it, CPU will do that.
Last edited:
Wow thanks for the benchmarks cmdrdredd!
It's true that recently I've only been playing online MMO's and that it's hard to tell what future games will be dependent on. I wanted to wait because really I have low expectations for AAA single player games, and I don't expect my system to have trouble with single player at least until the next generation of consoles come out and developers develop accordingly. Therefore it seems that yes for the games I'm playing at the moment a new CPU will provide the biggest boost to minimum fps when online.
However if a GTX 670 can provide me with a consistent 5-10 minimum fps boost, I think that should be enough to tide me over until Haswell. From the benchmark cmdredd has provided it seems that it can.
I live in Mauritius (a tiny island in the ocean), so my access to a second hand market is limited. But I did find someone locally willing to take my 6950 for USD 200 so I'm leaning towards that route. It's much harder for me to resell my CPU.
I think I'm going to get the GTX 670 and ask a family member who is visiting bring it over for me, thus saving a shitload on shipping, and then import the CPU and mobo if I find the GPU just isn't enough.
Again, thanks for the benchmarks cmdrdredd!
It's true that recently I've only been playing online MMO's and that it's hard to tell what future games will be dependent on. I wanted to wait because really I have low expectations for AAA single player games, and I don't expect my system to have trouble with single player at least until the next generation of consoles come out and developers develop accordingly. Therefore it seems that yes for the games I'm playing at the moment a new CPU will provide the biggest boost to minimum fps when online.
However if a GTX 670 can provide me with a consistent 5-10 minimum fps boost, I think that should be enough to tide me over until Haswell. From the benchmark cmdredd has provided it seems that it can.
I live in Mauritius (a tiny island in the ocean), so my access to a second hand market is limited. But I did find someone locally willing to take my 6950 for USD 200 so I'm leaning towards that route. It's much harder for me to resell my CPU.
I think I'm going to get the GTX 670 and ask a family member who is visiting bring it over for me, thus saving a shitload on shipping, and then import the CPU and mobo if I find the GPU just isn't enough.
Again, thanks for the benchmarks cmdrdredd!
AtenRa
Lifer
- Feb 2, 2009
- 14,001
- 3,357
- 136
Most games need a higher performance GPU as they are GPU limeted. With the GTX670 you'll be able to raise the IQ settings in games and still have higher performance than with the HD6950.
BF3 MP, especially in open 64 player maps (Caspian Border for example) it needs 6-8 threaded CPUs + high performance GPUs. If you play BF3 MP more than other games buy an 8-core AMD FX or Intel 8 threaded CPU.
BF3 MP, especially in open 64 player maps (Caspian Border for example) it needs 6-8 threaded CPUs + high performance GPUs. If you play BF3 MP more than other games buy an 8-core AMD FX or Intel 8 threaded CPU.
Yes get the GTX 670 right now do MB later.
I have a I7920/3.5Ghz and son had a dual core socket 775.
I upgraded his to a GTX 570 and a used Q 9450 (stock no OC)
and it beats my I7 in most games.
I have a 5870 in my I7. No doubt the GTX 570 would do a little bit better in my I7 box but not a ton better.
So a slower cpu with a faster vpu will beat a faster cpu with slower vpu in most games. He does not have any issues with any games. Plays Lost planet 2 at 60fps while my I7/5870 can only muster 22fps at the same settings.
Go for the video card first for sure.
I have a I7920/3.5Ghz and son had a dual core socket 775.
I upgraded his to a GTX 570 and a used Q 9450 (stock no OC)
and it beats my I7 in most games.
I have a 5870 in my I7. No doubt the GTX 570 would do a little bit better in my I7 box but not a ton better.
So a slower cpu with a faster vpu will beat a faster cpu with slower vpu in most games. He does not have any issues with any games. Plays Lost planet 2 at 60fps while my I7/5870 can only muster 22fps at the same settings.
Go for the video card first for sure.
TRENDING THREADS
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 10K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 7K
-
-
Facebook
Twitter