Not getting the full potential out of my 7870?

[Red Hawk]

I just installed a Radeon HD 7870 in my PC, and while it has certainly improved over my old 5770 (particularly in Batman: Arkham City) it doesn't feel like the improvement is as great as it should have been. I kept reading in reviews that the 7870 and 7850 essentially had double the performance of the 5700 series, but it doesn't feel like that. Several games still feel sluggish or jerky; for example, Crysis stays around just 25 FPS when I'm just standing around at 2x MSAA and all other settings max, a marginal improvement over 18-20 FPS with my 5770. Any ideas of what could be causing this, or if there's something else bottlenecking my PC? I did a clean install of the 7870, uninstalling all AMD software and uninstalling the graphics card and driver from Windows' Device Manager before removing it and installing the 7870. I'm running Catalyst 12.8.

 

[balane]

I have the exact same video card. I run it at 1220/1450 on stock voltage and it helped a lot. Also, get the 12.9 drivers and 12.9 CAP, they are improved. Very happy with mine at 1920x1200 (3570, 8GB)

 

[Termie]

First thing to do would be run 3dMark11 to make sure you're in the right ballpark for the graphics and overall score. The overall score should be around 6500.

Did you use the Batman benchmark on your old card by any chance? I think the 7870 should be about 3x faster with high tessellation selected in that game.

 

[KingFatty]

Definitely get specific benchmark numbers.

If your card is in agreement with all the benchmarks, then your card is fine and nothing wrong with the hardware. If so, then the issue is your subjective impression of the increase, and that can be addressed by talking to a professional counselor to help you through coming to grips with GPUs that just don't measure up.

only kidding! I'm sure there is some issue you can fix with the GPU to get your frames where they are supposed to be.

 

[VulgarDisplay]

I just installed a Radeon HD 7870 in my PC, and while it has certainly improved over my old 5770 (particularly in Batman: Arkham City) it doesn't feel like the improvement is as great as it should have been. I kept reading in reviews that the 7870 and 7850 essentially had double the performance of the 5700 series, but it doesn't feel like that. Several games still feel sluggish or jerky; for example, Crysis stays around just 25 FPS when I'm just standing around at 2x MSAA and all other settings max, a marginal improvement over 18-20 FPS with my 5770. Any ideas of what could be causing this, or if there's something else bottlenecking my PC? I did a clean install of the 7870, uninstalling all AMD software and uninstalling the graphics card and driver from Windows' Device Manager before removing it and installing the 7870. I'm running Catalyst 12.8.

You sure you were running the same exact in game settings on the 5770? Because with 2xAA in Crysis my 7970 doesn't seem to really keep my frames acceptable at all times. The game is still just that demanding.

 

[Red Hawk]

First thing to do would be run 3dMark11 to make sure you're in the right ballpark for the graphics and overall score. The overall score should be around 6500.

Did you use the Batman benchmark on your old card by any chance? I think the 7870 should be about 3x faster with high tessellation selected in that game.

I ran 3DMark 11, and got a score of 4304...oh dear...

All scores are down. The CPU Physics score is actually down more, at 3433 when 3DMark tells me it should be 8900. What in the world could be causing this? Could something be causing my PC to throttle itself?

 

[Termie]

I ran 3DMark 11, and got a score of 4304...oh dear...

All scores are down. The CPU Physics score is actually down more, at 3433 when 3DMark tells me it should be 8900. What in the world could be causing this? Could something be causing my PC to throttle itself?

Your CPU physics score should be much higher - I'm not sure about 8,900, but at least twice as high as it is. Clearly, something is wrong with the CPU. What was the graphics score? If it's around 6,000, you know the GPU is working fine.

Did you change anything else when you replaced the video card? Any changes in the BIOS? Power supply? The first thing I'd do is set the BIOS to all default setting - clear the CMOS if that's the easiest way. Can you confirm the speed the CPU is operating at? Run 3dMark11 in a window and have CPU-z open to watch the frequency.

Another possibility is failing memory. Maybe try to run 3dMark11 with one stick and then with the other.

 

[Red Hawk]

Ok, I disabled a few startup services and more importantly I reset all the settings in my motherboard's BIOS (except for hard drive mode -- default is IDE and the OS was installed in AHCI, so it needs to be AHCI in order to boot). That means stock CPU clocks and Turbo Boost enabled. Definite improvement, as the physics score is now 6552 and 3DMark is saying the target should be 5400 (I guess 3DMark tried guessing the boost from overclocking the first time around). Graphics score seems to have improved, but it's still below expected. Total score is 5089 and 3D mark reports that the expected total should be 5600.

Dunno what was wrong about my overclocking settings that made performance drag rather than go up. Next I'll get rid of the card's factory 10 MHz overclock on both core clocks and memory clocks to see if that changes anything. If I'm still falling short, maybe I'll install the 12.9 beta drivers.

EDIT: Huh. After removing the factory overclock (and disabling Gigabyte's Easytune utility which I noticed had not been disabled on startup) the score is now 6371. Really weird. How could overclocking have been causing performance to drop? I thought that that problems overclocking could cause are graphical glitches and crashes, not lower performance.

 

[Termie]

Ok, I disabled a few startup services and more importantly I reset all the settings in my motherboard's BIOS (except for hard drive mode -- default is IDE and the OS was installed in AHCI, so it needs to be AHCI in order to boot). That means stock CPU clocks and Turbo Boost enabled. Definite improvement, as the physics score is now 6552 and 3DMark is saying the target should be 5400 (I guess 3DMark tried guessing the boost from overclocking the first time around). Graphics score seems to have improved, but it's still below expected. Total score is 5089 and 3D mark reports that the expected total should be 5600.

Dunno what was wrong about my overclocking settings that made performance drag rather than go up. Next I'll get rid of the card's factory 10 MHz overclock on both core clocks and memory clocks to see if that changes anything. If I'm still falling short, maybe I'll install the 12.9 beta drivers.

EDIT: Huh. After removing the factory overclock (and disabling Gigabyte's Easytune utility which I noticed had not been disabled on startup) the score is now 6371. Really weird. How could overclocking have been causing performance to drop? I thought that that problems overclocking could cause are graphical glitches and crashes, not lower performance.

Boom, you're done. Nothing wrong with the GPU at all. That CPU overclock was really messed up, and I'm glad you caught the Easytune software - I should have mentioned that you should uninstall any motherboard utitlies but you hadn't mentioned using them.

At this point, you can return your GPU to its factory settings (10MHz OC), and start over on that CPU overclock, using the BIOS only.

 

[Red Hawk]

I had been doing overclocking through the BIOS, Easytune was just a utility that I figured was harmless to let sit in my startup list. Perhaps not. I reenabled the CPU OC to 4 GHz (only by changing the maximum multiplier to 40) and disabled Turbo Boost. Reran 3DMark11, and scores are still normal. I'm going to rerun the test one more time with the 10 MHz GPU overclocks to make sure those are fine. Thanks for recommending that I use 3DMark11 for testing!

 

[gpse]

why disable turbo boost?

 

[Termie]

I had been doing overclocking through the BIOS, Easytune was just a utility that I figured was harmless to let sit in my startup list. Perhaps not. I reenabled the CPU OC to 4 GHz (only by changing the maximum multiplier to 40) and disabled Turbo Boost. Reran 3DMark11, and scores are still normal. I'm going to rerun the test one more time with the 10 MHz GPU overclocks to make sure those are fine. Thanks for recommending that I use 3DMark11 for testing!

You're welcome! It's funny - 3dMark11 gets a bad rap for being overused, but it's the best benchmark I know of to test whether CPUs and GPUs are operating in their normal range.

Now that this is all figured out, let us know how Crysis runs. I bet you'll see that the 7870 is a big improvement over the 5770! Actually, it's possible you weren't getting everything out of that 5770 either, but that's water under the bridge...

 

[Red Hawk]

Odd. At normal 7870 stock clocks the card runs as expected now. Bump core and memory clocks up 10 MHz to the default factory overclock, and 3DMark11 drops below 5000. Oh well, I'm happy as is and the 10 MHz boost probably shouldn't give much of an improvement anyways. Maybe I'll revisit overclocking when newer drivers come out.

As for Crysis, it's running much better now. Frame rates are at a steady 30 FPS when walking around at the beginning, even if I bump it up to 4x or 8x MSAA.

Edit: And just to illustrate how much of a difference the problem before was making, I benchmarked the beginning scene of Deus Ex: Human Revolution before and after with Fraps. Before, the min/max/average was 44, 86, and 61.08. Now it's 72, 131, and 101.55. Oy vey.

 

[Termie]

Odd. At normal 7870 stock clocks the card runs as expected now. Bump core and memory clocks up 10 MHz to the default factory overclock, and 3DMark11 drops below 5000. Oh well, I'm happy as is and the 10 MHz boost probably shouldn't give much of an improvement anyways. Maybe I'll revisit overclocking when newer drivers come out.

As for Crysis, it's running much better now. Frame rates are at a steady 30 FPS when walking around at the beginning, even if I bump it up to 4x or 8x MSAA.

That is really, really strange. When you have the time, try a few more runs of 3dMark11 at different speeds. If it's consistently must slower at 1010MHz than at 1000MHz, I'd definitely try something higher to see if the card can hold an overclock. Should be able to do 1100MHz easily. If it doesn't run correctly at 1010 (the factory settings), I'd consider RMA'ing it, but it will be a pain explaining the situation, since it's obviously not crashing.

 

[Red Hawk]

Ok, I did 3DMark11 at each factory overclock individually with the other at stock, benchmarked at 1100 MHz core clock with memory at stock, and lastly did 1100 MHz core clock with the factory setting for memory. Results: Neither core nor memory dragged down the score when overclocked individually. It ran fine at 1100 MHz individually, and the memory overclock didn't bring down the score when run alongside the 1100 MHz core overclock. The sharp drop in scores and frame rate only, and consistently, occurs when both core and memory clocks are set to the factory overclock of 1010 MHz and 1210 MHz respectively. This is just weird.

Edit:

why disable turbo boost?

Because the clock I'm setting the CPU to (4 GHz) is higher than any default Turbo Boost setting, so there's no need for it.

 

[Termie]

Ok, I did 3DMark11 at each factory overclock individually with the other at stock, benchmarked at 1100 MHz core clock with memory at stock, and lastly did 1100 MHz core clock with the factory setting for memory. Results: Neither core nor memory dragged down the score when overclocked individually. It ran fine at 1100 MHz individually, and the memory overclock didn't bring down the score when run alongside the 1100 MHz core overclock. The sharp drop in scores and frame rate only, and consistently, occurs when both core and memory clocks are set to the factory overclock of 1010 MHz and 1210 MHz respectively. This is just weird.
...

All I can say is that you've discovered a portal to another universe, where 10 + 10 = -100

Anyway, while you could probably RMA to Asus and note that the card bends space and time, you might as well be happy with a very solid card and a nice and easy 1100Mhz manual overclock.

BTW, that's not the first time an Asus card bends time and space, LOL:

I knew the Asus GTX 660 TOP was quiet from using it and from the readings at idle, but its noise levels under load are something else entirely. Somehow, the card bends time and space, shaving 0.8 decibels off of its noise level at idle—and lowering the noise floor of the entire test system and environment at the same time. The really freaky thing is that this strange result pretty much matches my subjective impressions. I need to check the label on that new allergy medicine.

http://techreport.com/review/23527/review-nvidia-geforce-gtx-660-graphics-card/10

 

[tweakboy]

I have the exact same video card. I run it at 1220/1450 on stock voltage and it helped a lot. Also, get the 12.9 drivers and 12.9 CAP, they are improved. Very happy with mine at 1920x1200 (3570, 8GB)

Your a very happy 3570 user.. God bless you!

 

[Red Hawk]

All I can say is that you've discovered a portal to another universe, where 10 + 10 = -100

Anyway, while you could probably RMA to Asus and note that the card bends space and time, you might as well be happy with a very solid card and a nice and easy 1100Mhz manual overclock.

BTW, that's not the first time an Asus card bends time and space, LOL:

Cool! Maybe I can make a TARDIS out of it. 😛 But yeah, not worth an RMA. Even if the memory overclock didn't work at all, I wouldn't go to the trouble just because I didn't get the 10 MHz of memory speed. I bought the card for the deal it was on, not the factory overclock, anyways.

 

[Termie]

Cool! Maybe I can make a TARDIS out of it. 😛 But yeah, not worth an RMA. Even if the memory overclock didn't work at all, I wouldn't go to the trouble just because I didn't get the 10 MHz of memory speed. I bought the card for the deal it was on, not the factory overclock, anyways.

Hold on a second...

Ok, I did 3DMark11 at each factory overclock individually with the other at stock, benchmarked at 1100 MHz core clock with memory at stock, and lastly did 1100 MHz core clock with the factory setting for memory. Results: Neither core nor memory dragged down the score when overclocked individually. It ran fine at 1100 MHz individually, and the memory overclock didn't bring down the score when run alongside the 1100 MHz core overclock. The sharp drop in scores and frame rate only, and consistently, occurs when both core and memory clocks are set to the factory overclock of 1010 MHz and 1210 MHz respectively. This is just weird.

...

You apparently found that the memory overclock worked fine, as long as you used anything other than the factory core clock.

That's what's so incredibly weird about this situation. While a memory overclock certainly can cause error correction to kick in, reducing your fps, it probably wouldn't do so selectively, only with one core clock but not a higher core clock. Anyway, point is that you should go ahead and test that memory overclock again, maybe even dialing in something like a 50MHz to see if fps drops or if it continues to rise.

 

[Silverforce11]

If its giving you lower scores, run at 1.2ghz vram. Sometimes 10mhz is the difference with ECC on GDDR5.

Most of the gains are had from core OC on the 78xx series.

 

[zaydq]

Sometimes memory and core clocks don't work well with each other, just fiddle with em (preferably higher 😛) and find a stable match. Weird that factory OC is wonky but sometimes it happens, even manufacturers can goof an OC.