[SMP Client] Why is my PPD Low?

This guide will cover why you may be experiencing lower-than-expected PPD (Points Per Day) on your SMP client.

Here is the list of potential issues that may be slowing down your SMP client. I'll cover each of them in more detail below:

1. Bonuses not being calculated
2. An ATI GPU client
3. Multiple GPU clients
4. Gaming While Folding
5. SMP Client doesn't have a SMP WU
6. Other programs hogging the CPU
7. Old WU
8. Low-PPD WU

Bonuses not being calculated
1). One of the most common issues behind the question "Why is my SMP client so slow" is that bonuses are not being calculated. The current version of the SMP client is heavily dependent on bonus points. The work units have a low base-point value (most of them are worth slightly under 500 points to start with). Stanford then credits you additional points for a quick return. One of the most popular Folding@home client monitors, FAHmon, does not calculate these bonuses. As such, it only calculates the base points of the WU into your PPD estimate. To see how much of a difference this makes, let's look at HFM (client monitoring software that does calculate bonuses) and Fahmon on my Core i7 @ 3.8ghz.
Without bonuses, I'm getting ~2,000 PPD



With bonuses, on the other hand, I'm getting slightly over 15,000 PPD



Please note, just because you are using HFM does not mean that bonuses are being calculated. If you are unsure, open HFM.NET and go to Edit->Preferences->Options and make sure that "Calculate Bonus Credit and PPD" is selected



An ATI GPU client
2). By default, the ATi GPU client will use an entire core on your computer. Clearly, this has the potential to vastly cut into your PPD. However, there is a solution. First of all, apply the registry file containing the environmental variables in this post and reboot. If your PPD still suffers, follow the rest of this guide. It probably won't be needed, but may help you strike a better balance between SMP and GPU PPD.

Multiple GPU clients
3). Next we have what at first may seem like a paradox: Folding@home slowing itself down. Now, this isn't actually a paradox. This can occur when you are running multiple Folding@home clients (see #2 above). However, while 1 nVidia client typically has very little impact on SMP PPD, 2 or more can make a large difference. If you are disappointed by the PPD of your SMP client, try closing the GPU client(s) and let the CPU client fold a few percent (I like to give it at least 5%). If this is the issue, your PPD should go back up to expected values. Unfortunately, there's really nothing that can be done about this.

Running 2 nVidia GPU2 clients on my i7 drops the CPU PPD about 2k. This is worth it, however, because between the pair, I get ~10k from the GPU clients. Adding more GPUs will slow down the CPU, it's just a matter of how much. As such, if you're already folding on a fast CPU, such as the i7 used above, it's probably not even worth it to fold on a very slow GPU, such as a Geforce GT210. The hit to the CPU PPD will likely be more than the GPU can do. This is especially significant with ATI GPUs, nothing under a Radeon HD4850 or HD5750 is likely going to be worth folding on due to the hit it will make on the CPU, even with the environmental variables above.

Gaming While Folding
4). This is probably less likely than most of the other issues I'm covering here, but I still think that it deserves consideration. While it is perfectly possible to leave the SMP client running while gaming (and I usually do), the game will slow down the SMP client. Just like the GPUs slowing it down, this isn't debatable. It's fact. You probably won't be checking your HFM setup while you're gaming, but if you decide that you must, this is why things will be low.

SMP Client doesn't have a SMP WU
5). In contrast to the previous issue, this is by far one of the most common issues. If the SMP client is configured incorrectly (no shortcut with the -smp flag, or if the standard Unicore client is used with the flag) you are going to get one of the unicore work units. These will only use 1 core (or thread) of your processor, so if you have anything faster than a Pentium D it's not worth folding them. You're wasting the true processing power of your CPU. This is what you do not want to see:



Under "Client", you want to see "SMP", not "Standard". Under "Core" you want to see "GRO-A3". Under "Core ID" you want to see A3. Anything else is wrong. If you see anything else, delete your 'work' folder and queue.dat (located in the SMP folder) and follow Zodac's excellent SMP guide

Other programs hogging the CPU
6). Background programs, we all run them. Some of us more than others. If you suspect that other programs that are using the CPU may be slowing down your folding, open up Task Manager and on the "Processes" tab you should see that FahCore_a3.exe is using 90%+ of your CPU (ideally 95-99%). If it isn't, you likely have other programs that are stealing the CPU cycles. If you can close them, do so. You want to see something similar to the following:



(User Name column blanked to protect my identity)

Old WU
7). The final issue is that of the "old WU". The bonus points that you receive for completing a work unit are based on how long it takes you to return the WU to Stanford University's Collection Servers, starting the second the WU was downloaded. As long as you get it returned by the preferred deadline, you get a bonus. Even if it's only 1 minute before the preferred deadline.

On my i7 860, I can get most SMP WUs done in about 5.5 hours, and I get about 3.5k points for them. Say that half way through the WU, I had to shut down my computer for 6 hours. As far as Stanford is concerned, by the time I get the WU returned, it will have taken me 12 hours to fold it. Thus, instead of getting the 3.5k points, I would get about 2.3k. So, if your computer has been off or not folding for some period since the WU was downloaded, you will get fewer points, and thus lower PPD. How much lower depends entirely on how long it was off for. A way to counter this, and time your Folding for the most PPD can be found here.

For most SMP WUs, you have 3 days to return the WU to receive a bonus (this is called the Preferred Deadline; more time on P6701, P6702, P6703, and P6040 WUs). If you do not get it returned by then, you get no bonus. However, as long as it is returned by the Final Deadline (6 days on most WUs), you still get the base points.

Low-PPD WU
8). Stanford has released a couple work units that are assigned to SMP clients and give lower PPD than the others. If you have a P6701, P6702, or P7603 (you can check in the "Project" tab of HFM.NET) and your PPD is below what you are expecting, this is probably why. These work units are larger (they take a bit over twice the time as most other WUs on my i7 and Core 2 Duo), but don't give correspondingly many points. While my i7 can get up to 17k PPD on most SMP WUs (particularly the P6015s), with the P6700-series WUs I get between 11 and 12k PPD without GPU clients. These WUs always give lower PPD, but there's nothing that can really be done about it.
Edited by mmx+ - 8/23/10 at 11:41am

You mean 6701/6702?
I'll add the P670Xs in a few minutes

1) This is correct

2) As in number of SMP WUs submitted? If so, that's completely wrong. Bonuses are the same, regardless if you've submitted 10 WUs with a passkey or 10k

I'll see what I can do about incorporating a bigadv section this weekend, it might be better suited to go in the SMP guide
I'll also see what could be done about bonuses

You're only running a standard unicore WU. Did you create the shortcut with the -smp flag? This is vitally important. And you downloaded the SMP client, not the unicore one, right?

Delete the 'work' folder and queue.dat (in the SMP folder). If you follow Z's guide, there's the part about making the shortcut with the flag. Try that