Old PC upgrade suggestions - Updated to RTX2060

[agent71]

Thanks for info.

I am probably pushing the limit of the existing 530W PSU but so far, assuming I'm reading responses correctly, I'm not seeing any artificial throttling due to power even if I uncap the power for the 2060.

Although the 2060 is showing as "Performance State" = P2 rather than P0? Seems like there's room to improve the 2060.

The 106 though will not go beyond 75W. nvidia-smi doesn't allow.

I'll play with the slots and see what mb can do. It's only a 1 x PCIE 3.0 x16, 1 x PCIE 2.0 x16 board. With a GPU in the PCIE 2.0 slot it drops to 2.0 x4 speed. The P106 is in the PCIE 2.0 slot.

300K+ would be nice from the 106 but I've got a feeling it's more a 1050 Ti + than close to 1060 perf. Could be wrong? Have to admit at this point P106 seems a false economy. GTX 1650 Super better $/PPD/watt.

Code: Select all

@folding01:~$ nvidia-smi
Fri Jun  5 19:55:04 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.64.00    Driver Version: 440.64.00    CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce RTX 2060    On   | 00000000:01:00.0 Off |                  N/A |
| 82%   74C    P2   162W / 200W |    479MiB /  5934MiB |     99%      Default |
+-------------------------------+----------------------+----------------------+
|   1  P106-090            On   | 00000000:02:00.0 Off |                  N/A |
| 49%   50C    P0    56W /  75W |    236MiB /  3021MiB |     99%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      5524      G   /usr/lib/xorg/Xorg                            12MiB |
|    0      6756      C   ...org/v7/lin/64bit/Core_22.fah/FahCore_22   455MiB |
|    1      5524      G   /usr/lib/xorg/Xorg                             6MiB |
|    1      5772      C   ...org/v7/lin/64bit/Core_22.fah/FahCore_22   219MiB |
+-----------------------------------------------------------------------------+
@folding01:~$

Code: Select all

@folding01:~$ nvidia-smi -q -d PERFORMANCE

==============NVSMI LOG==============

Timestamp                           : Fri Jun  5 19:54:30 2020
Driver Version                      : 440.64.00
CUDA Version                        : 10.2

Attached GPUs                       : 2
GPU 00000000:01:00.0
    Performance State               : P2
    Clocks Throttle Reasons
        Idle                        : Not Active
        Applications Clocks Setting : Not Active
        SW Power Cap                : Not Active
        HW Slowdown                 : Not Active
            HW Thermal Slowdown     : Not Active
            HW Power Brake Slowdown : Not Active
        Sync Boost                  : Not Active
        SW Thermal Slowdown         : Not Active
        Display Clock Setting       : Not Active

GPU 00000000:02:00.0
    Performance State               : P0
    Clocks Throttle Reasons
        Idle                        : Not Active
        Applications Clocks Setting : Not Active
        SW Power Cap                : Not Active
        HW Slowdown                 : Not Active
            HW Thermal Slowdown     : Not Active
            HW Power Brake Slowdown : Not Active
        Sync Boost                  : Not Active
        SW Thermal Slowdown         : Not Active
        Display Clock Setting       : Not Active

@folding01:~$

[PantherX]

...Although the 2060 is showing as "Performance State" = P2 rather than P0? Seems like there's room to improve the 2060...

IIRC, this is a feature of Nvidia's Drivers (https://www.reddit.com/r/RenderToken/co ... th_nvidia/)

[MeeLee]

P0 state can only be achieved under liquid cooling, or, it is the initial boost your GPU can offer you, if it's below 45C (cold); and should not be enabled.
You could run pretty close to P0 state speeds, through overclocking a P2 state, as long as you cool your GPU well.

It appears your 2060 is running at 168W, but set to 200W.
It will be safer to set the cap to 170W manually by typing:

Code: Select all

sudo nvidia-smi -i 0 -pl 170

and capping the max frequency you deem the GPU to be working correctly on.

Code: Select all

sudo nvidia-smi -i 0 -lgc 300,1935

-i 0 is setting the parameters for the first GPU ('-i 1' should be your second GPU).
-pl 170 is limiting the GPU to 170W, to protect your PSU from being overloaded on a high atom count WU. It would even be better to set this to 150 or 140, or 130 in some situations. The drop in GPU frequency you'll notice from limiting the power consumption, you can easily adjust in nvidia x server (only accessible via GUI) through overclocking. But be gentle with the overclocks, FAH is very sensitive to overclocking gpus. So use sparingly.
-lgc 300,1935 is limiting the max frequency to 1935Mhz. Usually older GPUs are unsafe to run much beyond 1935 or 1995Mhz (depending on the silicon, and temperture). They could run into errors, their capacitors will make noise (coil whining, which has nothing to do with coils btw), or give crashes or freezes in worst case scenario (especially when overclocking incorrectly).

Check your version by typing 'lsb_release -r' in terminal if you need to.
To enable overclocking in Linux, you can safely on 18.04 or 16.04 do:

Code: Select all

sudo nvidia-xconfig --enable-all-gpus
sudo nvidia-xconfig --cool-bits=28

then reboot.
Then Nvidia-X server will have a manual slider for fan curve, as well as GPU and VRAM overclocking options
(be aware, this may break your GPU if you overclock too high).

On versions 19 and up (the newest version is 20.04) you can only enable overclocking on the primary GPU, unless you don't mind a constantly crashing desktop...
That means only line "sudo nvidia-xconfig --cool-bits=28" will work. The other will crash the desktop. You can still somewhat use it though...

[HaloJones]

To enable overclocking in Linux, you can safely on 18.04 or 16.04 do:

Code: Select all

sudo nvidia-xconfig --enable-all-gpus
sudo nvidia-xconfig --cool-bits=28

then reboot.
Then Nvidia-X server will have a manual slider for fan curve, as well as GPU and VRAM overclocking options
(be aware, this may break your GPU if you overclock too high).

On versions 19 and up (the newest version is 20.04) you can only enable overclocking on the primary GPU, unless you don't mind a constantly crashing desktop...
That means only line "sudo nvidia-xconfig --cool-bits=28" will work. The other will crash the desktop. You can still somewhat use it though...

Where did you get this about overclocking a secondary gpu will crash the desktop?

[MeeLee]

To enable overclocking in Linux, you can safely on 18.04 or 16.04 do:

Code: Select all

sudo nvidia-xconfig --enable-all-gpus
sudo nvidia-xconfig --cool-bits=28

then reboot.
Then Nvidia-X server will have a manual slider for fan curve, as well as GPU and VRAM overclocking options
(be aware, this may break your GPU if you overclock too high).

On versions 19 and up (the newest version is 20.04) you can only enable overclocking on the primary GPU, unless you don't mind a constantly crashing desktop...
That means only line "sudo nvidia-xconfig --cool-bits=28" will work. The other will crash the desktop. You can still somewhat use it though...

Where did you get this about overclocking a secondary gpu will crash the desktop?

Try it.
I just haven't gotten around undoing this. I run 18.04 which doesn't have the crashing issue.
Also, you need to install the nvidia drivers from the .run files for multi GPU overclocking.
Using the provided .deb files in the repositories also crashes the desktop on the second GPU (even on 18.04).

[agent71]

Code: Select all

@folding01:~$ cat /var/log/Xorg.0.log | grep -i link
[     5.860] (II) NVIDIA(0): Detected PCI Express Link width: 16X
[     5.864] (II) NVIDIA(1): Detected PCI Express Link width: 4X
@folding01:~$

MB doesn't allow me to set slot speeds only PCIE Gen values. So I tried that but no real success.

Forcing 2060 (gpu 0) to PCIE 2.0 doesn't appear to have affected it's folding perf much. -3% PPD reduction but that has only resulted in a +3% PPD on the P106. With 2060 giving more PPD I'm down >3% overall. Although +/-3% could be caused by reboot/FAH restart/WU etc.

So I think, as we say in England, I'm p***ing in the wind trying to get more out of this.

I'm also not convinced about the P106-90. They might be cheap at 50-60 euros/GBP/$ and low wattage at 75W but they only return 200-250K PPD.

3 x used P106-90 = 225W, £180 and 3 x slots for 600-750K PPD. Re-sale of GPU not great either assuming anyone will want them in x years time.

1 x new GTX 1650 Super = 110W, £155, 1 x slot for <> 750K PPD.

Everyone has their budget and different goals but for me the P106-90 feels a false economy and a GTX 1650 Super better value for money. Might look at one once the PC rebuilt into new case. Have a Lian Li PC-O11 Air coming for this machine. Slightly overkill... but interested in the case for main/gaming PC should this one impress.

This has definitely spiralled from my initial plans!!! Although I'm enjoying it and hopefully it's my way of making a small positive difference in the world.

[MeeLee]

Code: Select all

@folding01:~$ cat /var/log/Xorg.0.log | grep -i link
[     5.860] (II) NVIDIA(0): Detected PCI Express Link width: 16X
[     5.864] (II) NVIDIA(1): Detected PCI Express Link width: 4X
@folding01:~$

MB doesn't allow me to set slot speeds only PCIE Gen values. So I tried that but no real success.

Forcing 2060 (gpu 0) to PCIE 2.0 doesn't appear to have affected it's folding perf much. -3% PPD reduction but that has only resulted in a +3% PPD on the P106. With 2060 giving more PPD I'm down >3% overall. Although +/-3% could be caused by reboot/FAH restart/WU etc.

So I think, as we say in England, I'm p***ing in the wind trying to get more out of this.

I'm also not convinced about the P106-90. They might be cheap at 50-60 euros/GBP/$ and low wattage at 75W but they only return 200-250K PPD.

3 x used P106-90 = 225W, £180 and 3 x slots for 600-750K PPD. Re-sale of GPU not great either assuming anyone will want them in x years time.

1 x new GTX 1650 Super = 110W, £155, 1 x slot for <> 750K PPD.

Everyone has their budget and different goals but for me the P106-90 feels a false economy and a GTX 1650 Super better value for money. Might look at one once the PC rebuilt into new case. Have a Lian Li PC-O11 Air coming for this machine. Slightly overkill... but interested in the case for main/gaming PC should this one impress.

This has definitely spiralled from my initial plans!!! Although I'm enjoying it and hopefully it's my way of making a small positive difference in the world.

If you're looking at performance per watt per dollar, the best buy is actually a 2060KO. (a 2060 Super if you're into overclocking, as the KOs have very limited overclocking ability).
But the 2060 KO a price point of about a 2060 Super, and a performance of a 2070 Super, it's really the best buy.
It's about twice the cost of a 1650, but it also gets about 2,5x the performance.

Especially with Nvidia having a turnover of one new or upgraded GPU every year, it makes little sense to get the most expensive models, unless you:
a- Can afford them,
b- Can afford the extra electric cost,
c- Want fastest PPD,
d- Have a wattage treshold you can not modify (eg: 1 socket 1450W).

A 2080Ti would take about 12 years to pay itself back in $$ to a 2060KO; while a 2060KO would pay itself back immediately compared to 2x GTX1650s, and it's running cost is much lower too.

[HaloJones]

If you're looking at performance per watt per dollar, the best buy is actually a 2060KO. (a 2060 Super if you're into overclocking, as the KOs have very limited overclocking ability).
But the 2060 KO a price point of about a 2060 Super, and a performance of a 2070 Super, it's really the best buy.
It's about twice the cost of a 1650, but it also gets about 2,5x the performance.

the 2060KO does not have the performance in FAH as a 2070Super.

https://docs.google.com/spreadsheets/d/ ... =323814829

based on the above submitted results, the 2060KO averages 1.1m ppd while a 2070S averages 1.6m ppd.

the report by GamersNexus into the EVGA 2060KO and its performance in Blender has never been repeated in FAH.

[MeeLee]

Not sure if the spreadsheet is correct.
A lot of the numbers are core21.
The KO should actually be closer to a 2080 from what I've heard (1,4M PPD on Core 21, 2,1M PPD on core 22).

[agent71]

Those averages for the 2060KO are on par with my current 2060 which will deliver anything between 1.0-1.3M PPD.

The KO appears to be a 2060 for near 2060 Super money but with no gain in performance...

Based on PPD/Watt/$/£ the 1650 Super is looking pretty compelling. It looks likely it'll work within the limits of the existing PSU too. It's low watts, low cost, but well above average PPD. The obvious answer is slam another 2060 in the second slot but due to 100% WFH right now I'm saving £150 per month in coffees. A £150 1650 Super is easy to justify going to science rather than Starbucks

[HaloJones]

Not sure if the spreadsheet is correct.
A lot of the numbers are core21.
The KO should actually be closer to a 2080 from what I've heard (1,4M PPD on Core 21, 2,1M PPD on core 22).

all 5 results returned for the 2060KO in that spreadsheet are for Core22.

We've had long discussions over this 2060KO before. It looked promising at first but then the guy with both vanilla 2060 and 2060ko saw that actually there was no difference.

If you could provide a source for what you heard, I'd be very interested to read it.