Gaming in an upgraded Mac Pro 1.1 - Is the CPU my bottleneck?

[Guigue]

Hi!

I wasn’t sure which thread to post in since this could end up being related to multiple (CPU upgrade, NVIDIA in older Mac Pros, Mavericks on older Mac Pros), so instead I decided to create a new one.

I’m in the process of upgrading my aging Mac Pro 1.1. So far, I have:

- Upgraded the firmware from 1.1 to 2.1
- Installed Mavericks 10.9.1
- Installed an EVGA GTX 660

Other than that, I have 8 GB or RAM (more on the way though) and the original 2 x 2.66 GHz Dual Core Xeon.

Everything is running smoothly, except my gaming performances. I’m mainly using Borderlands 2 and Starcraft 2: Heart of the Swarm as my guides.

When in Borderlands, I often get stutter at average settings. I tried lowering them, and while thing would get better, I could always sense that things wouldn’t go smoothly in certain area. I ended up with super low settings and the game would still slow down when moving around looking at specific places.

I opened activity monitor to see if something was wrong, and I noticed that the CPU use for Borderlands is 120%, no matter what I do.

Since I feel like I pretty much took the graphic card out of the equation by lowering the settings to the absolute minimum, I’m wondering if the CPU is the bottleneck. I am very surprised that, even though they’re old processors, Borderlands 2 wouldn’t play on Quad Core 2.66 GHz Xeons at the lowest settings…

Any idea what could be done? To either test out and isolate the problem, or fix it. I have replacement processors coming (changing the 5130 for 5355), but since they have the same clock speed I don’t expect any improvement

Thanks!

 

[BSDanalyst]

Hi!

I wasn’t sure which thread to post in since this could end up being related to multiple (CPU upgrade, NVIDIA in older Mac Pros, Mavericks on older Mac Pros), so instead I decided to create a new one.

I’m in the process of upgrading my aging Mac Pro 1.1. So far, I have:

- Upgraded the firmware from 1.1 to 2.1
- Installed Mavericks 10.9.1
- Installed an EVGA GTX 660

Other than that, I have 8 GB or RAM (more on the way though) and the original 2 x 2.66 GHz Dual Core Xeon.

Everything is running smoothly, except my gaming performances. I’m mainly using Borderlands 2 and Starcraft 2: Heart of the Swarm as my guides.

When in Borderlands, I often get stutter at average settings. I tried lowering them, and while thing would get better, I could always sense that things wouldn’t go smoothly in certain area. I ended up with super low settings and the game would still slow down when moving around looking at specific places.

I opened activity monitor to see if something was wrong, and I noticed that the CPU use for Borderlands is 120%, no matter what I do.

Since I feel like I pretty much took the graphic card out of the equation by lowering the settings to the absolute minimum, I’m wondering if the CPU is the bottleneck. I am very surprised that, even though they’re old processors, Borderlands 2 wouldn’t play on Quad Core 2.66 GHz Xeons at the lowest settings…

Any idea what could be done? To either test out and isolate the problem, or fix it. I have replacement processors coming (changing the 5130 for 5355), but since they have the same clock speed I don’t expect any improvement

Thanks!

No it is not. The hardware should not even come close to being a problem. The problem is in the OS. Run bootcamp for gaming.

 

[cryingrobot]

Our threads seemed to have crossed as I posted something very similar. I have a very similar issue. Upgraded 1,1 to 10.9 using tiamos method and installed a gtx 660 to run borderlands 2. I also have an iMac with the 775m gpu to give some comparison. My 1,1 has the 2.66 Octo with 16 gigs memory and running an ssd for the boot.

1. On my 1,1 bl2 boot amp runs quite well. Highest settings with physx on 1920 by 1080 monitor. It did crash a couple of times...maybe due to physx.

2. On my 1,1 on osx, again highest settings and looks fine but no physx. However, FPS seemed kind of choppy despite settings. This might have been an issue with the monitor.

3. On the iMac, just no comparison. Runs on highest setting at native 2440? Resolution and the smoothness of the FPS is very noticeable.

Others have mentioned the 660 should be a significant performance upgrade from the 775m but it looks like it is not as smooth and doesn't have the same fps. I've also read that the 1,1 PCI slot is such a bottleneck that a 660 is completely wasted. IMO, it's a significant upgrade from the 4870 I was using and seems to be able to run highest settings but the choppiness indicates there is some kind of bottleneck.

 

[ybz90]

Depends on the game, but there is a very good chance your CPU could bottleneck. I would suspect some Blizzard games are like this.

----------

No it is not. The hardware should not even come close to being a problem. The problem is in the OS. Run bootcamp for gaming.

This is untrue. The 2006 Xeons have pretty low single threaded performance compared to today's computers. Based on what the OP describes from a usability standpoint, it has nothing to do with the OS.

 

[fhenry]

probably true for gaming, less for other things like octane which works without any bottleneck !

 

[branana]

In my experience (4,1 to 5,1, hexacore 3.2ghz), all games are pretty CPU bottlenecked. Most of these games only use OpenGL 3 which from what I gather is terrible compared to OpenGL 4 and DirectX 11 at feeding the video card.

I have a GTX 680 and through all games I only get about 60% GPU utilization. (Xcom, Civ 5, Dota 2 etc) no matter if I play at high res high detail or low res low details (A dead give away for CPU bottleneck)

Unfortunately if you must get the most out of your hardware for gaming, DirectX is your only option until games start coming out supporting OpenGL 4.

 

[cryingrobot]

Interesting. Ive heard there is also a 5% bottleneck on the PCI. Is that 60% CPU bottleneck on top of the PCI bottleneck?

 

[leman]

In my experience (4,1 to 5,1, hexacore 3.2ghz), all games are pretty CPU bottlenecked. Most of these games only use OpenGL 3 which from what I gather is terrible compared to OpenGL 4 and DirectX 11 at feeding the video card.

Only if the programmer is terrible

I have a GTX 680 and through all games I only get about 60% GPU utilization. (Xcom, Civ 5, Dota 2 etc) no matter if I play at high res high detail or low res low details (A dead give away for CPU bottleneck)

No, its simply because all these games are a joke from the GPU requirements standpoint. And BTW, all those games use DirectX 9, not 11 (except Civ5, which has two versions)

Unfortunately if you must get the most out of your hardware for gaming, DirectX is your only option until games start coming out supporting OpenGL 4.

Could you elaborate?

----------

Interesting. Ive heard there is also a 5% bottleneck on the PCI. Is that 60% CPU bottleneck on top of the PCI bottleneck?

This doesn't make any sense. Please read up on what a bottleneck means, e.g.here: http://en.wikipedia.org/wiki/Bottleneck

 

[fhenry]

i start a thread with bench to find answers : https://forums.macrumors.com/threads/1685478/

if you have bench to post, you are welcome !

 

[branana]

Only if the programmer is terrible

A programmer can't do much when the CPU can't feed enough rendering commands to the GPU.

No, its simply because all these games are a joke from the GPU requirements standpoint. And BTW, all those games use DirectX 9, not 11 (except Civ5, which has two versions)

The games might be jokes in terms of rendering complexity, but 20fps at max resolution/detail and the same 20fps at minimal resolution/detail while GPU is at a 60% workload indicates the bottleneck is somewhere else. Anecdotally, even those DirectX 9 games run much better in Windows.

This doesn't make any sense. Please read up on what a bottleneck means, e.g.here: http://en.wikipedia.org/wiki/Bottleneck

I understand what bottlenecking means. My understanding of OpenGL 3 under OSX for games is that it is pitiful compared to other modern operating systems. I am using OpenGL Driver profiler.

- Looking at a 120% CPU usage on the i7 970's 3.2ghz core tells me that the games are not taking much advantage of multicore. The clock speed is going to be the main determining factor here.
- At the same time OpenGL Driver profiler is showing the GTX 680 only uses about 500m out of possibly max of 950m core utilization.
- Comparing the GPU utilization against the CPU utilization while running Xcom or Dota 2, while the game is running below the intended performance level says that those games are CPU bottlenecked.
- Comparing against those games running under the same conditions in Windows, my blame automatically points to the graphics subsystem differing between Windows and OSX. There are a variety of documentation out there indicating that OpenGL 3 on OSX is subpar compared to OpenGL 4 and OpenGL in general on other platforms and definitely way behind DirectX when it comes to gaming and making use of the latest gaming video cards.

 

[leman]

I understand what bottlenecking means.

Oh, I have no doubt about that — I was responding to cryingrobot here

My understanding of OpenGL 3 under OSX for games is that it is pitiful compared to other modern operating systems. I am using OpenGL Driver profiler.

[...]

- Comparing against those games running under the same conditions in Windows, my blame automatically points to the graphics subsystem differing between Windows and OSX. There are a variety of documentation out there indicating that OpenGL 3 on OSX is subpar compared to OpenGL 4 and OpenGL in general on other platforms and definitely way behind DirectX when it comes to gaming and making use of the latest gaming video cards.

I absolutely agree that OS X graphics subsystem is subpar in comparison to Windows. I disagree though that its an API issue. AFAIK, the only thing that GL4 brings to the table over GL3 for better batching is ARB_draw_indirect with friends and tessellation. Tessellation is neat but its not exactly a draw call overhead reducing technique - more a quality improving one. And indirect drawing is only useful with certain algorithms which I am sure none of those games use. I simply don't see how you can improve the batching behaviour of games like X-Com and Civ with GL4 over GL3.3

P.S. X-Com runs at steady > 30 fps on OS X with my rMBP. I find it highly suspicious that your 680 is so much slower — the i-920 can't be that bad.

P.P.S. The GL Profiler hooks into the application, hence it can reduce its performance. Depending on how you use it of course. Collecting traces will make any application CPU-limited. Simple fps monitoring should has only minimal impact, but using GL Profiler for that is a pita. Not to mention that with OS X you should run games in full screen, as this allows the system to optimise the performance.

 

[Guigue]

Thanks for the good discussions guys!

I'm starting to think my specific problem is driver related. I put the graphics at the lowest possible settings, found a place where it would jerk when I panned over it (not even a big environnement, a closed room) and included is a screen shot of my CPU history. None of the CPUs are maxed out.

So I still have room on the CPU, the GPU is obviously capable of rendering a game at 640x480, and it still bugs down on specific game elements. After reading some terrible benchmarks on NVIDIA 6xx cards in Mavericks, I'm thinking the drivers are just plain bad.

I'll see if I can install 10.8 on another partition someday and try this out. Boot Camp is not an option yet, on a Mac Pro 1.1 with the original optical drive removed, it's a real pain in the butt.

In the mean time, I'll let my friend do the heavy killing in busy parts of the game...

 

[branana]

Oh, I have no doubt about that — I was responding to cryingrobot here



I absolutely agree that OS X graphics subsystem is subpar in comparison to Windows. I disagree though that its an API issue. AFAIK, the only thing that GL4 brings to the table over GL3 for better batching is ARB_draw_indirect with friends and tessellation. Tessellation is neat but its not exactly a draw call overhead reducing technique - more a quality improving one. And indirect drawing is only useful with certain algorithms which I am sure none of those games use. I simply don't see how you can improve the batching behaviour of games like X-Com and Civ with GL4 over GL3.3

P.S. X-Com runs at steady > 30 fps on OS X with my rMBP. I find it highly suspicious that your 680 is so much slower — the i-920 can't be that bad.

P.P.S. The GL Profiler hooks into the application, hence it can reduce its performance. Depending on how you use it of course. Collecting traces will make any application CPU-limited. Simple fps monitoring should has only minimal impact, but using GL Profiler for that is a pita. Not to mention that with OS X you should run games in full screen, as this allows the system to optimise the performance.

True about trace performance impact. I couldn't find another thing that would show simple overlay for framerate. I'm sure the tracing and the fact there's a second 1080p desktop showing is impacting my fps. I also find my poor framerate suspect.

If you wouldn't mind can you share your framerate numbers on the first mission of Xcom without moving any of your guys at all. Running high details and a few resolutions?

I have a first gen rMBP as well. Xcom didn't run super well on it. It might have gotten better since Feral has worked on it some. But I'd say the i7 in the rMBP is in fact boosting to pretty high clock and although similar architecture to the i7 970, is still a good 5-10% faster clock for clock. I wouldn't be surprised if it gets better fps.

----------

Thanks for the good discussions guys!

I'm starting to think my specific problem is driver related. I put the graphics at the lowest possible settings, found a place where it would jerk when I panned over it (not even a big environnement, a closed room) and included is a screen shot of my CPU history. None of the CPUs are maxed out.

So I still have room on the CPU, the GPU is obviously capable of rendering a game at 640x480, and it still bugs down on specific game elements. After reading some terrible benchmarks on NVIDIA 6xx cards in Mavericks, I'm thinking the drivers are just plain bad.

I'll see if I can install 10.8 on another partition someday and try this out. Boot Camp is not an option yet, on a Mac Pro 1.1 with the original optical drive removed, it's a real pain in the butt.

In the mean time, I'll let my friend do the heavy killing in busy parts of the game...

What game is that? Borderlands 2?

That looks like all 4 cores are evenly used? Are they all being used by the game? If so, that's a pretty well CPU optimized game!

----------

Unfortunately, it is a problem with the openGL drivers that Apple have not updated in a long long time. Because OS X does not use directX like in Windows, it relies on oGL, which should be good enough when properly coded for. Many games now, when they are ported over, have very rudimentary translation techniques, like Cider or a wineskin. Borderlands 2 is somewhat recoded, but it's still not "native." It was written to use directX libraries. Mac performance in games are worse for this reason, despite what people say about Macs not having good hardware. It's the outdated and unoptimised graphics libraries and terribly ported games themselves that create these issues.

I'm running a Hackintosh with a 670 and 3570K, and believe me, for BL2, performance is ♥♥♥♥ty on both Mac and PC sides. BL2 is just very terribly optimised on PC to begin with, and Mac performance is even worse. I suggest playing all your games on your Bootcamp partition, not just BL2. Mac gaming performance is unlikely to get better as long as Apple does not update their graphics libraries and drivers, and as long as games are developed for PC and terribly ported to OS X.

Taken from http://forums.steampowered.com/forums/showthread.php?t=3124412

 

[Guigue]

What game is that? Borderlands 2?

That looks like all 4 cores are evenly used? Are they all being used by the game? If so, that's a pretty well CPU optimized game!

Yes, Borderlands 2. Indeed it seems all 4 cores are used evenly. I made sure it was the only program using my CPUs in a meaningful way, so it's all BL2

 

[branana]

Yes, Borderlands 2. Indeed it seems all 4 cores are used evenly. I made sure it was the only program using my CPUs in a meaningful way, so it's all BL2

I think you are just out of luck, this is as far as that Mac Pro 1,1's CPU and the state of gaming on Macs goes.

 

[goMac]

The 1,1 has a slower PCI-E bus than the 2008 that slows it's GPU performance. This first started showing up with the Radeon 5870 upgrade and could be what you're seeing.

No real way around except getting a new machine.

 

[Guigue]

The 1,1 has a slower PCI-E bus than the 2008 that slows it's GPU performance. This first started showing up with the Radeon 5870 upgrade and could be what you're seeing.

No real way around except getting a new machine.

The weird thing is I get slowdowns no matter the settings. At 640x480 with everything off or low, I don't think the GTX 660, even on a slower PCI-E bus, it at maximum capacity

 

[Asgorath]

Thanks for the good discussions guys!

I'm starting to think my specific problem is driver related. I put the graphics at the lowest possible settings, found a place where it would jerk when I panned over it (not even a big environnement, a closed room) and included is a screen shot of my CPU history. None of the CPUs are maxed out.

So I still have room on the CPU, the GPU is obviously capable of rendering a game at 640x480, and it still bugs down on specific game elements. After reading some terrible benchmarks on NVIDIA 6xx cards in Mavericks, I'm thinking the drivers are just plain bad.

I'll see if I can install 10.8 on another partition someday and try this out. Boot Camp is not an option yet, on a Mac Pro 1.1 with the original optical drive removed, it's a real pain in the butt.

In the mean time, I'll let my friend do the heavy killing in busy parts of the game...

The OS is time slicing the application across all available cores, so you won't see a single core hitting 100% unless all of them do. One thread running at 100% of one core will show up as 25% across all 4 cores, for example.

The MacPro1,1 from 2006 is now approaching 8 years old, using CPU technology from 9+ years ago. So, to answer your question: yes, your CPUs cannot keep up with the much more modern GPU and are absolutely the bottleneck in your system. Clock speeds have gone from 2.66GHz to closer to 4GHz on a modern CPU, and you're missing generation upon generation of architectural and process improvements that will mean a modern CPU is more than twice as fast as the CPU you have.

Here's a great comparison for reference:

http://www.anandtech.com/show/7603/mac-pro-review-late-2013/5

I recently upgraded from a 3.2GHz Quad-Core Nehalem in a 2010 Mac Pro to a 3.5GHz Core i7-4770K (Haswell) and was shocked at how much performance in games like Starcraft 2 improved. World of Warcraft went from a stuttering mess in any busy scene to buttery smooth with a modern CPU.

It really might be time to put the old MacPro1,1 to rest and get an up-to-date system. You might be surprised at just how much faster they are.

 

[Frong]

...Clock speeds have gone from 2.66GHz to closer to 4GHz on a modern CPU..

My Mac Pro Rev1.1 has Xeon at 3Ghz, two dual core ie 4 core. I doubt much gaming configuration use quad core at 3Ghz.

But yeah now quad core at 3Ghz is starting to become old.

 

[Asgorath]

My Mac Pro Rev1.1 has Xeon at 3Ghz, two dual core ie 4 core. I doubt much gaming configuration use quad core at 3Ghz.

But yeah now quad core at 3Ghz is starting to become old.

My point is that a modern 4GHz CPU isn't going to be only 33% faster than your old MacPro1,1 (from 3GHz -> 4GHz alone), it's going to be more than twice as fast due to architectural and process improvements, increases in RAM speed, and so on. Your Woodcrest CPUs are at least five generations old at this point.

 

[EHuntington]

I don't want to be a downer here, but gaming in Mac OS is a bummer. Every game is going to run worse than windows. In my experience, it's not really worth it to try. I have a Mac Pro 1,1 with a Radeon 7950 and playing anything in Mac OS is a stuttery mess. I appreciate Aspyr for porting games to the Mac, but I feel like they're hampered by slow drivers and a bad OpenGL implementation. Running games in OpenGL through windows yields much faster results.

If you don't have the means to run Boot Camp, I think you'll just have to suffer through it. From my experience, try using 10.8.5 and download Nvidia's drivers. Supposedly there's a 15% performance penalty for using 10.9. I usually had to turn off AA as well. I wish you luck, and hope that maybe 10.9.3 will bring updated drivers. I feel your pain.

 

[Asgorath]

I have a Mac Pro 1,1 with a Radeon 7950 and playing anything in Mac OS is a stuttery mess. I appreciate Aspyr for porting games to the Mac, but I feel like they're hampered by slow drivers and a bad OpenGL implementation.

While there might be some penalty from the fact that Apple's OpenGL implementation is behind DirectX or OpenGL on Windows, your CPU is just way too slow to power such a modern GPU. I'm running WoW at 2560x1600 at Ultra settings and 4xAA at 100+ FPS on my Core i7-4770K (Haswell) system, while I could barely break 40 with my 2010 Mac Pro with a 3.2GHz Nehalem processor.

 

[EHuntington]

While there might be some penalty from the fact that Apple's OpenGL implementation is behind DirectX or OpenGL on Windows, your CPU is just way too slow to power such a modern GPU. I'm running WoW at 2560x1600 at Ultra settings and 4xAA at 100+ FPS on my Core i7-4770K (Haswell) system, while I could barely break 40 with my 2010 Mac Pro with a 3.2GHz Nehalem processor.

You're completely right. The architecture is far behind, and that's a huge reason for the deficiency in gaming performance. I just think, despite its age, the Mac Pro 1,1/2,1 should be capable of more than it is under OS X. I mean, using Boot Camp I can play Skyrim at 60fps (with AA set to 2x.) I can play Borderlands 2 at 60fps. I know your i7 can do much, much more, but the old 1,1 can produce very playable results under Windows with Direct X or even OpenGL.

 

[Frong]

My point is that a modern 4GHz CPU isn't going to be only 33% faster than your old MacPro1,1 (from 3GHz -> 4GHz alone), it's going to be more than twice as fast due to architectural and process improvements, increases in RAM speed, and so on. Your Woodcrest CPUs are at least five generations old at this point.

First point is you are forgetting the two processors thing, putting it quite in advance in term of number of core and Ghz. The second point is the CPU power rush forward has been tempered a lot those last years. The third point is my GPU is still the bottleneck despite I upgraded it and alas couldn't made the best upgrade without risk so I sticked to Radeon HD 5770.

Yeah now this one is an old beast too. But it's still the GPU that is drag down the whole (lower resolution better performance proving it in multiple games).

 

[cryingrobot]

I spent a couple of hours on the 1,1 running a 660 playing BL2 on Mavericks with everything on highest settings. It started out a bit ragged but within 5-10 minutes it ran much more smoothly. Not sure why. Also, there was one game crash which basically never happened before.

This was on an old 1280 resolution display but I did originally test it on a 1920 resolution and it was performing about the same. It was actually not as smooth as the imac on 2560 resolution IMO but still very decent and a big improvement from the 4870 I had earlier.