TR Forums

Hi Everyone.  Long time reader ( since around 2006 ) but very infrequent poster.  I have a question for everyone regarding the bandwidth throughput of a GTX 1070 and how it relates to performance degradation.  I recently ( yesterday ) upgraded from a GTX 660 Ti to a GTX 1070.  I'm currently running a Core i7 950 ( Bloomfield ) with 12GB of DDR3 2100 on a Gigabyte GA-X58A-UD3R.  This motherboard only supports a maximum of PCIE 2.0 which only has data rates of 500MB/s, Total bandwidth rates of 16GB/s, and data transfer rates of 5GT/s, according to the PCIE2.0 specifications.  Also, according to the motherboard specification page it specifically says if you are running a single card that they highly suggest you put it in the first PCIE 16x slot.  Unfortunately, the first slot on my motherboard does not function ( it's a long story ) so I have it in the second slot.  I believe the reason behind wanting it in the first slot is closer proximity to the CPU but I'm not sure what this translates to in real world performance if it runs in the second slot.

Does anyone have any comments or suggestions on whether or not I am being artificially throttled by an old motherboard or perhaps some free benchmark tools to test this with along with some scores of what I should be getting with this card?

Edit:  I forgot to note I am running a Dell U2711 at 2560x1440.

Thanks in advance.

Every generation of motherboards and graphics cards, multiple websites test of whether PCIe 3.0 offers any meaningful benefits.

The answer is no, time and time again.
The answer is yes, in certain specific instances that are getting rarer and less likely to occur each generation.

If you are not running SLI in SFR (split-frame rendering mode) you don't need slot bandwidth
If you are not using CUDA to number crunch large datasets, you don't need slot bandwidth
If you are not gaming at resolutions so large you run out of local GPU video memory, then you don't need slot bandwidth.

At present, with a single GPU, A PCIe 2.0 x16 slot provides 99% of the speed of a PCIe 3.0 x16 slot. The differences are likely to be within the margin of measurement error.
Even if you are running in a PCIe 2.0 X8 slot, you're still getting 96% of the maximum performance your GPU has to offer.
In fact, you can get 85% of your GPUs performance from a PCIe 3.0 x1 (Same bandwidth as a PCe 2.0 x2 or a PCIe 1.1 x4 slot)

Thanks for the detailed response, I do appreciate it.

I am not running SLI.
I am not crunching datasets.
I game at 2560x1440 which isn't that high in my opinion compared to what most modern day games are reviewed at.

My second slot is a PCIe 2.0 x16 slot, not a x8 though as you indicated.

I just find it kinda hard to believe that going from a 660 Ti to a 1070 wasn't as big of an upgrade as I thought it was going to be.

https://www.techpowerup.com/reviews/NVI ... ng/21.html
PCIe 2.0 x16 is 99% the speed of PCIe 3.0 x16 at 1440p.

The i7 950 is your bottleneck. It was a great CPU, but that was 7 years ago or something, right?
An i3 6320 would run circles around it for gaming.

If you want to check, find any review of a 1070 that uses a game benchmark you own, and run the benchmark. If your 1440p results are way off theirs, the difference is the CPU.

Thanks for the link.  Interesting to note that they only found a 1% difference at 2560x1440 between PCIe 2.0 and 3.0

Don't get me wrong.  I was able to increase the graphics settings in most of my games from either low or medium to ultra and was still able to maintain a constant FPS of around 50-60, so it was an improvement, I just thought I would have gotten a bit more out of it.  Note that at around 50-60 FPS the games play just fine, unlike previously when I tried to set them that high.

I agree with you on the i7 950.  My next CPU/motherboard/RAM upgrade is probably going to be Kaby Lake in the spring when Intel releases the desktop CPUs since I'm not in too big of a rush right now.  

Thanks again for the detailed information.  I appreciate it.

jwolberg wrote:

Don't get me wrong.  I was able to increase the graphics settings in most of my games from either low or medium to ultra and was still able to maintain a constant FPS of around 50-60, so it was an improvement, I just thought I would have gotten a bit more out of it.  

You are aware that you won't exceed 60FPS if you have vertical sync enabled right? It won't go past the 60hz refresh rate of your monitor.  

Kougar wrote:

jwolberg wrote:

Don't get me wrong.  I was able to increase the graphics settings in most of my games from either low or medium to ultra and was still able to maintain a constant FPS of around 50-60, so it was an improvement, I just thought I would have gotten a bit more out of it.  

You are aware that you won't exceed 60FPS if you have vertical sync enabled right? It won't go past the 60hz refresh rate of your monitor.  

This- also, I ran SLI 670s and then 970s with a 2500k- that's half the bandwidth per card that you are dealing with, and it worked well, well enough that the CPU was pegged and at 4.5GHz, was actually holding the cards back!

What games specifically are you running? If they are CPU intensive, a good enough GPU will perform just as well as a top of the line one with no measurable difference. If the games are DX 9, they are far more responsive to a CPU upgrade, assuming the starting video card was good enough and not underpowered.

Arclight wrote:

What games specifically are you running? If they are CPU intensive, a good enough GPU will perform just as well as a top of the line one with no measurable difference. If the games are DX 9, they are far more responsive to a CPU upgrade, assuming the starting video card was good enough and not underpowered.

I can say for my situation that the performance was the same between BF4 (relatively light, but still enough to top off my 2500k) to Witcher 3.

You upgraded to a GPU that's allowing to run games at ULTRA settings at native 1440p resolution and maximum 60Hz refresh rate of your monitor.....how is that disappointing?

I'll just echo others here, but my first thought was: what games are you playing at 1440p and ultra settings and "only" getting 50-60fps?
In TR's review of the GTX 1070 where they tested Rise of Tomb Raider and Witcher 3 at the same settings you're mentioning, they got very similar results with their i7-5960X.

Also check if Vsync is on (which would limit your fps to 60fps since that's your monitor's max refresh rate).  Might want to try out NVidia's new Fast-Sync feature instead.

As my rule of thumb, "Ultra" settings usually doesn't add a whole lot of detail beyond "High" that's noticeable to the user, but oftentimes drastically increases the workload on the GPU.    

jwolberg wrote:

Thanks for the link.  Interesting to note that they only found a 1% difference at 2560x1440 between PCIe 2.0 and 3.0

Don't get me wrong.  I was able to increase the graphics settings in most of my games from either low or medium to ultra and was still able to maintain a constant FPS of around 50-60, so it was an improvement, I just thought I would have gotten a bit more out of it.  Note that at around 50-60 FPS the games play just fine, unlike previously when I tried to set them that high.

One thing to note... If you test your old card at low settings and then test the new card at low settings you won't improve much as the horsepower is just not needed. But when you start increasing the graphics detail you will notice a bigger difference.

Kougar:  I understand vertical sync and I have it disabled in the games I play.

Arclight:  I play mostly BF4, Civ5, and most recently the BF1 Demo ( using DX12 ).  I've been waiting for The Witcher and Fallout 4 to go on sale to pick those up.

DPete27:  Thanks for the link to the 1070 review.  I read it when it came out but it was nice to take a look at it again knowing the FPS rates they got with a much more powerful CPU than I am running in my current computer.  I really think I am being held back now by my old CPU.  It is currently being maxed out at 100% when I game now in most of the games I play.  As I noted above, VSync is disabled in the games I play.  I will try enabling Fast Sync through the NVIDIA Control Panel and see if it makes any real difference.


Thanks again everyone for the input.

I would also, like Ctispy_, point tomyour CPU holding your GPU back. An i7 950 is still great for many tasks, gaming not being one. Just get an i7 6700K or something and retire that old horse.
Edit: BTW, if you're seeing GPU utilization below 99% (e.g. in GPU Z), the amount of work your GPU can get done would be limited due to your CPU not being able to issue frames or artificial limita like fps caps. If you're seeing massive/irritating stuttering and hitching and GPU utilization bouncing around all over the place, that would point to your CPU being the bottleneck. Like I said, any modern i5/i7 would get the job done, post-SB that is. PCI-E 2.0 is not a significant limiting factor, at least for gaming.

If you are comfortable overclocking and have sufficient cooling to do it you could also alleviate some of the CPU bottleneck by doing that, which would be a lot cheaper than upgrading your CPU/motherboard/memory.  i7 950 was a pretty good overclocker, usually able to achieve a 30-40% overclock/performance improvement.  Even if you got a new heatsink to overclock with that heatsink could probably be used with your next CPU/motherboard as well.

I have a decent cooler for my CPU and I'm not averse to overclocking.  My issue is that I don't really have a good fundamental grasp of the various configuration settings to tweak to get a good, stable overclock.  If you have a link to any good guides for my specific CPU or any other overclocking guides in general I would absolutely appreciate it.

As far as the CPU goes, I do intend on upgrading in the spring to Kaby Lake.

Check out the 2 links in the 2nd post here:  http://www.overclock.net/t/845674/i7-950-good-oc-guide

First one is a general "Newbie guide to overclocking".  Second one is specifically about your generation of CPUs/motherboards.  The main settings you'd be tweaking will be FSB, CPU multiplier, Vcore (CPU voltage), and RAM divider/multiplier....then testing with a stability testing program like Prime95 or IntelBurnTest/Linx and keeping an eye on temperatures with a temperature monitor program.

Your CPU is actually the last generation where overclocking was a little bit complicated unfortunately.  All the stuff from Sandy Bridge and newer has been been much simpler since you don't really need to mess with the FSB anymore.  Kaby Lake will be the same in all likely hood.

If you have any specific questions I'm sure someone will be able to help you out too.

USAFTW wrote:

I would also, like Ctispy_, point tomyour CPU holding your GPU back. An i7 950 is still great for many tasks, gaming not being one. Just get an i7 6700K or something and retire that old horse.
Edit: BTW, if you're seeing GPU utilization below 99% (e.g. in GPU Z), the amount of work your GPU can get done would be limited due to your CPU not being able to issue frames or artificial limita like fps caps. If you're seeing massive/irritating stuttering and hitching and GPU utilization bouncing around all over the place, that would point to your CPU being the bottleneck. Like I said, any modern i5/i7 would get the job done, post-SB that is. PCI-E 2.0 is not a significant limiting factor, at least for gaming.

I had GPU-Z running while gaming and logging.  The GPU was at 100% load the entire time I was in game and only dropped to 0% when I exited out.

travbrad:  Thanks for the links.  I will check them out.  If you are curious, I will post back to let you know how high I got it.

Your CPU may be the bottleneck rather than the bandwidth to the card. @Chrispy_ is on point.

Do you have before/after benchmark numbers? I recently upgraded my brothers machine with a Asus 1070 Turbo from a 780 Reference and he is getting double minimum and average FPS at 1080p in Dying Light; he has a 4770K and 16GB of DDR3, forget the speed. That's the most intensive game he owns so that's all I have to go off of.

EVERYTHING Max
GTX 0780: 25min/70max/40avg
GTX 1070: 65min/160max/90avg

TardOnPC wrote:

Your CPU may be the bottleneck rather than the bandwidth to the card. @Chrispy_ is on point.

Do you have before/after benchmark numbers? I recently upgraded my brothers machine with a Asus 1070 Turbo from a 780 Reference and he is getting double minimum and average FPS at 1080p in Dying Light; he has a 4770K and 16GB of DDR3, forget the speed. That's the most intensive game he owns so that's all I have to go off of.

EVERYTHING Max
GTX 0780: 25min/70max/40avg
GTX 1070: 65min/160max/90avg

I did not get any benchmark numbers as I wasn't entirely familiar with any programs that could provide any other than just being able to comfortably play on higher settings with smoother framerates.  I still have my old card and would be more than happy to put it back in and run some benchmarks before and after if you have any suggestions of programs to use?

Nothing fancy. FRAPS is good enough. I haven't used it's video recording but FPS logging and screenshots work well for me.

jwolberg wrote:

Kougar:  I understand vertical sync and I have it disabled in the games I play.

Disabling it in the games is one thing, but that does nothing if vertical sync is still enabled in your NVIDIA driver settings.   GPU driver settings will always override the in-game settings far as I am aware.

Games like Civ V should be getting absurd FPS rates with vertical sync off on your GPU, mine are locked at 60 despite a much slower GPU @ larger resolution.

Kougar wrote:

jwolberg wrote:

Kougar:  I understand vertical sync and I have it disabled in the games I play.

Disabling it in the games is one thing, but that does nothing if vertical sync is still enabled in your NVIDIA driver settings.   GPU driver settings will always override the in-game settings far as I am aware.

Games like Civ V should be getting absurd FPS rates with vertical sync off on your GPU, mine are locked at 60 despite a much slower GPU @ larger resolution.

I have it configured currently to "Use the 3D application setting".

Just an FYI - for some unknown reason NVIDIA never enables MSI (message signaling interrupts) for their enthusiast/consumer cards (they do for their professional quadro lines). There is a slight chance that your card is being held back by when sharing an IRQ with another device. Enabling MSI via a registry tweak can avoid this particular issue. Unfortunately, the registry tweak must be reapplied every time you install a new driver.

I don't expect you to get much performance out of it, but it could alleviate audio and video stutters and unusual pauses during gameplay.

For more information:

http://forums.guru3d.com/showthread.php?t=378044

Rectal Prolapse wrote:

Just an FYI - for some unknown reason NVIDIA never enables MSI (message signaling interrupts) for their enthusiast/consumer cards (they do for their professional quadro lines). There is a slight chance that your card is being held back by when sharing an IRQ with another device. Enabling MSI via a registry tweak can avoid this particular issue. Unfortunately, the registry tweak must be reapplied every time you install a new driver.

I don't expect you to get much performance out of it, but it could alleviate audio and video stutters and unusual pauses during gameplay.

For more information:

http://forums.guru3d.com/showthread.php?t=378044

Is it working for you? Ever tested it?

seankay wrote:

Is it working for you? Ever tested it?

Yes, works fine. Although on my Sandy Bridge machine it didn't make much difference except that it virtually eliminates audio stutter on the desktop (NVIDIA HDMI audio controller also set to MSI). It never really had a problem to begin with, but it made a huge difference on my older Core2Duo machine - almost no video stutter in games anymore, with an older NVIDIA 670 card.
I suspect it will make a bigger difference on newer motherboards that pack too many features at once (dual NICs, wifi, USB3, SATA, onboard audio, Thunderbolt, more power savings states, etc. that share too damn many IRQs at once because driver writers are too half-assed to use MSI!).

jwolberg wrote:

I did not get any benchmark numbers as I wasn't entirely familiar with any programs that could provide any other than just being able to comfortably play on higher settings with smoother framerates.  I still have my old card and would be more than happy to put it back in and run some benchmarks before and after if you have any suggestions of programs to use?

Unigine Valley is a good simple benchmark.
It's pretty old now but it's DX11 which is what practically every game runs right now, so it's still valid.
Relatively small download and quick install - just pick the Exreme HD preset and hit run, then click the benchmark button top left. It'll spit out an FPS number which you can compare against other sites (here is just one example giving the 1070 Founder's Edition 99FPS on the default Extreme HD setting)
The other popular tool is 3D Mark (Firestrike) but that's far more demanding on your GPU and less likely to show up the effect of a CPU bottleneck.

jwolberg wrote:

The GPU was at 100% load the entire time I was in game

When in game what clock is your GPU running at? What is your GPU temp?

End User wrote:

jwolberg wrote:

The GPU was at 100% load the entire time I was in game

When in game what clock is your GPU running at? What is your GPU temp?

Core clock was bouncing around 1850-1880Mhz and temp was mid to high 60Cs according to GPU-Z

Chrispy_:  Thanks for the links.  I will post back my results from Unigine Valley after I run it shortly.

Here's some benchmarks I just ran using Unigine Valley.  Note that the default resolution for "Extreme HD" is 1920x1080 which is not my monitor's native resolution, so I also ran it at 2560x1440.

Unigine Valley Benchmark 1.0

FPS:
76.2
Score:
3188
Min FPS:
18.9
Max FPS:
133.1
System

Platform:
Windows 8 (build 9200) 64bit
CPU model:
Intel(R) Core(TM) i7 CPU 950 @ 3.07GHz (3064MHz) x4
GPU model:
NVIDIA GeForce GTX 1070 21.21.13.7270 (4095MB) x1
Settings

Render:
Direct3D11
Mode:
1920x1080 8xAA fullscreen
Preset
Extreme HD



Unigine Valley Benchmark 1.0

FPS:
54.0
Score:
2259
Min FPS:
23.5
Max FPS:
109.2
System

Platform:
Windows 8 (build 9200) 64bit
CPU model:
Intel(R) Core(TM) i7 CPU 950 @ 3.07GHz (3064MHz) x4
GPU model:
NVIDIA GeForce GTX 1070 21.21.13.7270 (4095MB) x1
Settings

Render:
Direct3D11
Mode:
2560x1440 8xAA fullscreen
Preset
Custom
Quality
Ultra


For some reason it detected my system as Windows 8 when I reality I am running Windows 10.  No idea on that one.

Rectal Prolapse wrote:

seankay wrote:

Is it working for you? Ever tested it?

Yes, works fine. Although on my Sandy Bridge machine it didn't make much difference except that it virtually eliminates audio stutter on the desktop (NVIDIA HDMI audio controller also set to MSI). It never really had a problem to begin with, but it made a huge difference on my older Core2Duo machine - almost no video stutter in games anymore, with an older NVIDIA 670 card.
I suspect it will make a bigger difference on newer motherboards that pack too many features at once (dual NICs, wifi, USB3, SATA, onboard audio, Thunderbolt, more power savings states, etc. that share too damn many IRQs at once because driver writers are too half-assed to use MSI!).

If that is the case then I would love to test it out on a better board. Gonna go to my tech guy about this pretty soon