Potato upgrade, Part 1 (GPU)

[David Mann]

In a previous thread (see below here), I asked about upgrading the CPU on my old Alienware X51 R2. I'm going to get to that eventually, but before that, I'd like to tell you about the GPU upgrade that I just did. I'd also like to thank everyone who contributed to the other thread for the lively and enlightening discussion.

How many CPU cores do I need?

My gaming computer is an old Alienware x51 r2. I've been slowly upgrading it (16 GB RAM, GTX 970 4 GB is on the way). I'm thinking about upgrading the CPU from a dual-core to a quad-core (i3-4130 to i7-4790). It seems like the i7-4790 is only slightly better at single-thread benchmarks. So...

www.racedepartment.com

As a bit of background, I was looking to spend a few hundred bucks to make this computer last several more years. I got the biggest A/C adapter they make for that computer (330 W) for about $60 to support the upgrades. The new GPU came in at about $100, which leaves me more than enough in my budget to pick up a CPU upgrade later. I'm going to go from the i3-4130 to the i7-4790, which is almost the fastest CPU I can get for this computer (the price premium for the i7-4790K isn't worth it to me). I've already maxed out the system RAM at 16 GB.

I'm presenting here some test results before and after the GPU upgrade. The games I tested are Dirt Rally 2.0 and ACC, which are the newest driving games that I play.

I got an EVGA GTX 970 4 GB, part number 04G-P4-2972-KR. This is a low-end 970 (slower clock), which is fine because I was worried about heat and power consumption. This replaces the GTX 645 1 GB that came with the computer. Here are the before and after stats for my computer (important notes are the H87 chipset and 16 GB of RAM):

Here are some before and after benchmarks. I'm interested in power usage (measured at the wall with a Kill-A-Watt meter), FPS (measured with FRAPS), CPU and GPU processor usage, GPU memory usage, and CPU and GPU temperatures. Room temperature was about 69 °F / 21 °C for all tests.

First is a Furmark run with the CPU Burn and GPU Stress tests both running. Test settings were 4 CPU threads, 2560x1080 resolution, and 2xMSAA.

	GTX 645	GTX 970
FPS	12	50
Power	120 W	235 W
CPU Usage	100%	100%
CPU Temperature	66 °C	66 °C
GPU Usage	100%	100%
GPU Temperature	77 °C	82 °C
GPU Memory	??	??

These results are promising all around. The GPU upgrade provides much more performance at reasonable temperatures, with plenty of power to spare for the future CPU upgrade.



The next test was in Dirt Rally 2.0. I drove for about one minute on a stage in Argentina with dusk/rain conditions. All graphics settings were at minimum except for Low night lighting (vs. Off). I also tried the graphics set to High with the 970.

	GTX 645	GTX 970	GTX 970 (High)
Minimum FPS	35	122	42
Mean FPS	41.7	145.1	59.3
Maximum FPS	50	168	70
Power	??	180 W	185 W
CPU Usage	30%	65%	50%
CPU Temperature	60 °C	68 °C	68 °C
GPU Usage	100%	100%	100%
GPU Temperature	69 °C	81 °C	80 °C
GPU Memory	1.0 GB (max)	3.5 GB	4.0 GB (max)

The numbers tell the story here. The experience is pretty poor with the GTX 645, but really good with the 970. I will say that there are some stages that had much lower frame rates with the 645, typically those stages with lots of trees at night. I am looking forward to a much better experience now. As a side note, I just used Task Manager to get the Usage and Memory stats. The CPU has Hyperthreading, so full usage of the two physical cores reads as 50% CPU usage, and four threads running full blast would show up as 100%. The CPU is effectively running at full capacity at a reading of 50% or above, so the game is now fully utilizing the CPU and GPU.



The other game I tested was ACC, which is very much CPU-limited on my computer. I tested by driving one lap around the Nurburgring at night in stormy weather, with 15 AI opponents. This was NOT a replay test, but actual driving to test real conditions. Graphics were set to Low (I also tested the 970 at Medium and High).

	GTX 645	GTX 970	GTX 970 (Medium)	GTX 970 (High)
Minimum FPS	18	30	25	29
Mean FPS	26.1	46.8	39.6	35.7
Maximum FPS	40	58	70	47
Power	100 W	175 W	190 W	190 W
CPU Usage	90%	100%	100%	100%
CPU Temperature	68 °C	71 °C	71 °C	71 °C
GPU Usage	100%	30%	50%	80%
GPU Temperature	77 °C	80 °C	80 °C	80 °C
GPU Memory	1.0 GB (max)	3.6 GB	3.4 GB	3.0 GB

It looks like having sufficient video memory is important, but beyond that the game is CPU-limited for me. The future CPU upgrade should help this a lot.



Just for kicks, I tested the GTX 970 with Tomb Raider (2013). This is the newest game I own that has a built-in benchmark. I was able to run it at Ultimate settings with 71.5 mean FPS. When I cranked up everything to the maximum, I got 34.3 mean FPS (the main difference is that Ultimate uses FXAA, while the highest setting for anti-aliasing is 4xMSAA). At the highest settings, this game is GPU-limited, while it is more balanced at lower settings. Even at Ultimate, the game looks fantastic for the most part.

Thanks for reading. I hope this will help some folks out there who can't afford the latest and greatest computer hardware. It's possible to get good results with decent second-hand hardware. I'll do this test again when I upgrade the CPU, and I'm looking forward to the results, especially in ACC.

Update: Part 2 of the upgrade (CPU) is documented here:

Potato upgrade, Part 2 (CPU)

After asking about how many CPU cores I need here and posting the results of my GPU upgrade here, I've finally upgraded my CPU. The upgrade is from a dual-core i3-4130 to a quad-core i7-4790 in an old Dell Alienware X51 R2. The computer is now officially upgraded from Potato status to Sweet...

www.racedepartment.com

 

[maelstrom]

That 80°C GPU temp is quite hot. I'd look at something like Afterburner and see about setting a custom fan curve to try and cool it a little more under load.

 

[RasmusP]

Great write up! Nice!

I don't see any issues with thr gpu going up to 80°c. GPUs can pretty much run beyond 100°c without taking too much damage.
All my GPUs were and are still running at about 80°c and they're all fine.
CPUs can take quite some Temps too. Intels auto shutdown at about 100°c. Lots of laptop CPUs are hitting 90 degrees and more and it's kinda "normal" for these.

All this "degradation" thingy in the gaming industry comes from keeping your maximum overclocking stable for years.
My i7 2600k needed a little bump in voltage after running at a nice sweetspot oc of 4.4 GHz for multiple years.
In a few years it might need another bump, which will also increase temperatures of course but it's not like the chip would be dying or anything

 

[maelstrom]

I don't see any issues with thr gpu going up to 80°c. GPUs can pretty much run beyond 100°c without taking too much damage.

But why allow the temp to be this high if you can easily avoid it?

 

[David Mann]

Great write up! Nice!

I don't see any issues with thr gpu going up to 80°c. GPUs can pretty much run beyond 100°c without taking too much damage.
All my GPUs were and are still running at about 80°c and they're all fine.
CPUs can take quite some Temps too. Intels auto shutdown at about 100°c. Lots of laptop CPUs are hitting 90 degrees and more and it's kinda "normal" for these.

All this "degradation" thingy in the gaming industry comes from keeping your maximum overclocking stable for years.
My i7 2600k needed a little bump in voltage after running at a nice sweetspot oc of 4.4 GHz for multiple years.
In a few years it might need another bump, which will also increase temperatures of course but it's not like the chip would be dying or anything

Yeah, I'm not worried about temperatures. I'm not overclocking anything, and the GPU seems to want to run at about 80 °C.

 

[RasmusP]

But why allow the temp to be this high if you can easily avoid it?

For me it's because of noise.

I've done some research about it all and this is what I ended up with:

In acc both CPU and gpu are used pretty much to the maximum of all games I play.
My CPU reaches 75°c and my gpu up to 83°c.
These are the sweetspots. Allowing the fans to go to full speed would mean 71 and 79 degrees.
But lowering the noise quite a lot only means a few degrees more.
The parts can take the temperatures without issues and the warmth in my room isn't affected by the temperature of the parts anyway.
My room gets heated by the power consumption.

Keeping the Temps inside the case only means heating it through radiation of the case instead of blowing the hot air into the room so I decided to focus in the noise level.

I don't like to use headphones for longer periods of time but I also don't like to put my speakers so loud that I wouldn't hear my gpu fans anymore.

I have a gtx 1070 msi gaming 8G. Not the most silent fans but pretty good and a sound I don't find too annoying.

Hope that explains it, hehe.

 

[maelstrom]

@RasmusP

There's a good video here from JayzTwoCents which covers some good, free monitoring/testing apps:

It's worth watching the whole video but there are a couple of points which are relevant for this thread. At around 5:00 on he shows how to see the maximum/default throttling temps for your GPU. His GPU throttled, by default, at 83°C. Could that be why yours runs at that max temp? That looks to me like a great reason to have a custom fan curve which keeps your GPU temp lower.

More useful info at about 9:40 on what may be causing a GPU to not reach maximum load (not always the CPU).

 

[RasmusP]

@RasmusP

There's a good video here from JayzTwoCents which covers some good, free monitoring/testing apps:

It's worth watching the whole video but there are a couple of points which are relevant for this thread. At around 5:00 on he shows how to see the maximum/default throttling temps for your GPU. His GPU throttled, by default, at 83°C. Could that be why yours runs at that max temp? That looks to me like a great reason to have a custom fan curve which keeps your GPU temp lower.

More useful info at about 9:40 on what may be causing a GPU to not reach maximum load (not always the CPU).

Nice video and nice input!
I already knew everything from the video but it's a really well done video!

When I'm racing I enable my custom msi afterburner profile which I dialed in nicely. There's no limit showing in GPU-Z when I enable it but in the standard profile it says "VRel" and the GPU won't go beyond 78°C.
VRel means the GPU-BIOS locks my voltage so it won't go higher. If I play around the sliders it will change to power limit and later on temperature limit if I force the fans to a lower speed.

My temperature stays at 83°c because I did the following:
1. Choose the most heat intensive GPU stress test I could find -> Furmark Furry Donut with 8x AA
2. Force fans to lowest speed possible (30%) until the Temperature hits 80°C
3. Slowly raise the fans but let the GPU hit 85°C
4. Raise the fans until the temperature goes down again
5. Lock the temperature at 83°C
6. Create a custom curve that has a plateau (hysteresis) at that fan speed to maintain a static noise floor while gaming
7. Create a bump in the fan curve to 100% fan speed beyond 84°C to not risk anything!

My fan curve:
I know the curve doesn't show the plateau but combined with the hysteresis I get a few sweetspots for most games where the fan speed and therefore the noise floor will become static.

For the CPU I did the same. Luckily with my old hardware I can use the good old "SpeedFan". Newer hardware mostly comes with a nice tuning tool though.

My CPU curve:
You can see the plateaus a lot better here. I need to have them more pronounced here due to the CPU fluctuating a lot more than the GPU! A little drop in load means 8°C less while a short burst to 100% means an instant jump.

 

[Durge Driven]

3900X with Kraken X62 never seen such high temps
I run only 4.2Ghz @ 1.25v for sims runs way cooler
Using only single 1440p and little DSR so no need to push system
I have enough frames in rF2 for best physics and CPU under-clocked

 

[David Mann]

I played DR2.0 with HWinfo64 running last night. The maximum GPU temperature was 81 °C, and the maximum temperature flag on the GPU was on at some point. I think that means that the GPU throttled back because of high temperature. The GPU fans only ever got up to 50%. I'm going to try using the EVGA utility and whatever else I can find to speed up the fans on the GPU and elsewhere in the system, to see if I can keep the GPU temperature under 80 °C.

 

[RasmusP]

I played DR2.0 with HWinfo64 running last night. The maximum GPU temperature was 81 °C, and the maximum temperature flag on the GPU was on at some point. I think that means that the GPU throttled back because of high temperature. The GPU fans only ever got up to 50%. I'm going to try using the EVGA utility and whatever else I can find to speed up the fans on the GPU and elsewhere in the system, to see if I can keep the GPU temperature under 80 °C.

I can recommend msi afterburner. It might be a bit overkill with the settings though but in the main screen of afterburner you can raise the temperature target. Simply raise it a few degrees and check again.
In afterburner you can also monitor the clockspeeds and even write them into a history file so you can check after your gaming session what did what

Imo all other gpu tools just try to make bad copies of afterburner.
It also works in symbiosis with Riva tuner statistics server (rtss).
Rtss links into the rendering chain and shows you whatever you want as an overlay while gaming.
So you can see everything in real time.
Also hands down the best fps limiter on the planet

 

[Trebormoore84]

In a previous thread (see below here), I asked about upgrading the CPU on my old Alienware X51 R2. I'm going to get to that eventually, but before that, I'd like to tell you about the GPU upgrade that I just did. I'd also like to thank everyone who contributed to the other thread for the lively and enlightening discussion.

Is this the case you have?: https://hothardware.com/reviews/dis...infused-alienware-x51-r2-sff-gaming-pc?page=3

I played DR2.0 with HWinfo64 running last night. The maximum GPU temperature was 81 °C, and the maximum temperature flag on the GPU was on at some point. I think that means that the GPU throttled back because of high temperature. The GPU fans only ever got up to 50%. I'm going to try using the EVGA utility and whatever else I can find to speed up the fans on the GPU and elsewhere in the system, to see if I can keep the GPU temperature under 80 °C.

This is all correct way of doing things in terms of management, however I think the main reason behind the GPU temps being higher than the original card is the type of cooling on the GPU. The original card looks like it was a blower style design that sucks air in and blows out the back of the case as part of the case circulation, I'm sure in such a tiny case Alienware will have carefully selected parts to get the temps to work. Whereas the new 970 you have is a normal double fan type which dumps heat into the case and relies on good case ventilation. This is quite a difficult thing to do in such a small case however.

Ramping the fans on the GPU up will help, but that heat needs to be taken out the case to be sustainable. Maybe try fixing the RPM to the CPU (which looks like it doubles as the case extract fan?) and case intake fans to the highest level you can cope with and see if that helps. You could even try turning the intake around and make it an extract as an experiment. Otherwise you might need to look into adding (modding the case) additional small fans to extract heat from the GPU area. Leaving the side of the case off, or and i'm sorry I know you don't want to hear this but replacing the GPU with a blower style model will most likely resolve the issue.

 

[David Mann]

I can recommend msi afterburner. It might be a bit overkill with the settings though but in the main screen of afterburner you can raise the temperature target. Simply raise it a few degrees and check again.
In afterburner you can also monitor the clockspeeds and even write them into a history file so you can check after your gaming session what did what

Imo all other gpu tools just try to make bad copies of afterburner.
It also works in symbiosis with Riva tuner statistics server (rtss).
Rtss links into the rendering chain and shows you whatever you want as an overlay while gaming.
So you can see everything in real time.
Also hands down the best fps limiter on the planet

I played around with MSI Afterburner to run the GPU fan faster and keep its temperature down to about 75 °C. It keeps the GPU off the temperature limit, but I haven't noticed a performance difference.

Is this the case you have?: https://hothardware.com/reviews/dis...infused-alienware-x51-r2-sff-gaming-pc?page=3



This is all correct way of doing things in terms of management, however I think the main reason behind the GPU temps being higher than the original card is the type of cooling on the GPU. The original card looks like it was a blower style design that sucks air in and blows out the back of the case as part of the case circulation, I'm sure in such a tiny case Alienware will have carefully selected parts to get the temps to work. Whereas the new 970 you have is a normal double fan type which dumps heat into the case and relies on good case ventilation. This is quite a difficult thing to do in such a small case however.

Ramping the fans on the GPU up will help, but that heat needs to be taken out the case to be sustainable. Maybe try fixing the RPM to the CPU (which looks like it doubles as the case extract fan?) and case intake fans to the highest level you can cope with and see if that helps. You could even try turning the intake around and make it an extract as an experiment. Otherwise you might need to look into adding (modding the case) additional small fans to extract heat from the GPU area. Leaving the side of the case off, or and i'm sorry I know you don't want to hear this but replacing the GPU with a blower style model will most likely resolve the issue.

Yes, that is my case, although I had a different GPU. The GTX 645 that I had was a tiny single-slot card that just had a single small fan on it. And I'm not sure, but I think my GTX 970 does suck in some air from the faceplate.

I haven't messed around with the fan curves other than for the GPU fan. Listening and looking at traces in Speedfan, I noticed during testing and gameplay that the CPU and system fans do ramp up when things get hotter. So, I think I'm in a good place right now.

By the way, i7-4790K's are more expensive than I thought on eBay. I have a bid in for one now, but if I lose that I'll let the overclockers have those and just get an i7-4790. Only 10% clock speed difference at boost for about an 80% price difference.

 

[David Mann]

CPU upgrade is posted. Thanks again for reading and commenting. I hope to get some good comments on Part 2 as well.

Potato upgrade, Part 2 (CPU)

After asking about how many CPU cores I need here and posting the results of my GPU upgrade here, I've finally upgraded my CPU. The upgrade is from a dual-core i3-4130 to a quad-core i7-4790 in an old Dell Alienware X51 R2. The computer is now officially upgraded from Potato status to Sweet...

www.racedepartment.com