I am trying this one right now:

-sbs 1024 -period_iterations_num 1 -spike_fft_thresh 2048 -high_perf -tune 1 2 1 16 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -cpu_lock

I have put that command line in, and it seems to be working pretty well. I think the GPU is almost as fast as when I tried running just 2 CPU tasks and 1 GPU task.

---

Seti@home classic: 1,456 results, 1.613 years CPU time

So this is my current command line for my 2200G:

-sbs 1024 -period_iterations_num 4 -spike_fft_thresh 2048 -high_perf -tune 1 2 1 16 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -cpu_lock

The reason period iterations is 4 is because I would get BSOD if that value was 1, 2, and 3. It has been stable at 4 for about a day now. I have had this value at 5 in the past for much longer periods of time, so I know it is stable at there. I'm wondering if this is the best I can make the stock application run, or if there are better settings to implement. I've read through the readme file, and it seems to be good at explaining what each command line is, but I am not familiar with the definitions of some of these terms, or the math/science behind them. So, since I don't know what it all means, I'm not sure I understand how to dial in these command line values other than to "mash buttons" and to hope for the best. Is there somewhere else with some more in detail reading so I can figure out what to do, or does anyone else have some further suggestions?

---

Seti@home classic: 1,456 results, 1.613 years CPU time

Okay, once I get back within range of my APU I will copy that CL and see if I can get it to process below 30 minutes reliably.

The cpu is on a different project so the RAC will not popup much. :)

Tom

---

A proud member of the OFA (Old Farts Association).

Anyone know the RAC for this card?

---

Anyone know the RAC for this card?

Which AMD card? I looked and you have several of them.
If you are running Windows you can use gpu-z utility to access their database and get a relative estimate of the performance of that card and compare it to the RAC's produced by other cards.

Shaggy has a report he runs that you can multiple the numbers by 24 (since he is displaying / hour) and get a pretty idea of the RAC a specific card can generate.

Tom

---

A proud member of the OFA (Old Farts Association).

Anyone know the RAC for this card?

Which AMD card? I looked and you have several of them.
If you are running Windows you can use gpu-z utility to access their database and get a relative estimate of the performance of that card and compare it to the RAC's produced by other cards.

Shaggy has a report he runs that you can multiple the numbers by 24 (since he is displaying / hour) and get a pretty idea of the RAC a specific card can generate.

Tom

Vega 64

---

...not forgetting that Shaggie's figures are for the stock applications running under Windows. This in the main due to all the issues with "optimised" applications where users may be running more than one task per GPU, or running some very "peculiar" application - both of which report as "Anonymous platform"; plus all the varieties of other operating systems have different performance characteristics that would make his life a bit of a nightmare.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

So I have finally been able to do a little more research about command lines. First, there have been responses in various threads that I feel are relevant, so I'll link them here:

Brief discussion about command line changes, per Kissagogo27.

Links to a Lunatics post by Raistmer about how to tweak performance with the command line.

Kissagogo27, thank you for the posts! I haven't had time to read through and attempt to understand any of this until now. I have a couple(!) of questions:

1. I've read through the Lunatics post, and its nice to have an explanation of how these different options work. I've only read through it a couple of times so I don't think I fully understand it. However, does anyone know if the info Raistmer posted applies to the anonymous application, or does that work for the stock application as well?

2. Raistmer states on the Lunatics forum that tt's default value is 60, but in the readme file for the application it says 15. Does anyone know if the discrepancy is an error, the difference between the Lunatics app and the stock app, or something else?

3. Raistmer says if the max value is significantly higher than the mean value to increase period_iterations_num. What is a "significant" difference? For example, this task has a difference of (211.1 max - 68.25 mean) 142.85 ms difference. I would assume that is significant. However, in this task over here, the first line has a difference of (102.3 max - 83.93 mean) 18.37 ms.

4. My sbs value is 1024. As I posted over in an APU memory post, I was wondering if I should assign more memory to the APU. I am thinking it may be worth changing a setting in bios to assign 4 GB to the GPU and increase sbs to 2048 or possibly higher. Is there anything in the stderr output that would indicate if this sbs value is too low or not? In other words, I am wondering if there is a way to see if I am choking the GPU by not having enough memory, or to tell if I have more memory allocated for the GPU than necessary.

5. As a follow-up to the last question, the Lunatics post mentions that pref_wg_size and pref_wg_num_per_cu would affect performance in tandem with sbs. However, I don't think I have seen anyone mention these switches before. Should these switches be added and if so, any suggestions for the Vega 8?

6. Between the readme sections and posts online, there is talk about GPU lagging. How do I tell if the GPU is lagging? Is it simply the display slowing down when doing simple Windows tasks, or is there something else I should be looking for?

---

Seti@home classic: 1,456 results, 1.613 years CPU time

6. Between the readme sections and posts online, there is talk about GPU lagging. How do I tell if the GPU is lagging? Is it simply the display slowing down when doing simple Windows tasks, or is there something else I should be looking for?

Gpu lagging is actually referring to too aggressive settings in the parameters, that cause "laggy" mouse movements and very slow keyboard inputs. What I see is when I type in an Editor, it can takes several seconds for any of my keyboard input to show up. Very annoying. Dial back the aggressive settings to some lower values until any "lagginess" disappears or gets below your tolerance threshold.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

So I have finally been able to do a little more research about command lines. First, there have been responses in various threads that I feel are relevant, so I'll link them here:
----edit----

4. My sbs value is 1024. As I posted over in an APU memory post, I was wondering if I should assign more memory to the APU. I am thinking it may be worth changing a setting in bios to assign 4 GB to the GPU and increase sbs to 2048 or possibly higher. Is there anything in the stderr output that would indicate if this sbs value is too low or not? In other words, I am wondering if there is a way to see if I am choking the GPU by not having enough memory, or to tell if I have more memory allocated for the GPU than necessary.

5. As a follow-up to the last question, the Lunatics post mentions that pref_wg_size and pref_wg_num_per_cu would affect performance in tandem with sbs. However, I don't think I have seen anyone mention these switches before. Should these switches be added and if so, any suggestions for the Vega 8?
---edit---

I am going to hazard a guess here. My Amd 2400g gpu is listed as being right next to a gtx 1030 in terms of performance on Gpu-Z's website/database.
This means you are not going to get even close to the performance of say a gtx 750ti out of it.

Your ram is bottle-necked by the fact it is DDR4 ram not video-card ram (I think). At least at one time one of the reasons that video card ram was a lot more expensive is you could read/write "at the same time"? which jumped the performance a lot compared to the standard ram of the day. And may still.

Once you get enough ram in there to allow the app to run without paging parts of the program out to the HD (eg. virtual storage) then I doubt unless you want to run 2 tasks at once, that you need more space.

All that said, Viper has pointed out that we have to bow to experimentation.

Presuming you have enough ram on the MB, there should be no hardware issue with running an APU ram space of 4GB and jumping your -sbs way up there. And then let it set at those settings for a couple of three weeks to try to get rid of our problem(s) with variable data helping us to jump to conclusions that may/maynot be warranted.

I would love to see a "real" Ryzen 3000 apu with the next gen Vega/Navi whatever gpu. But you have to consider the market. APU systems are not aimed at the HEDT. They are really aimed at the bottom of the gamer's market and anyone else who needs a basic office machine.

I like the through reviews that Anandtech has about the Ryzen 5 2200g/2400g apus. https://www.anandtech.com/show/12621/memory-scaling-zen-vega-apu-2200g-2400g-ryzen

While they are not Boinc/Seti specific they do explore the limits of this product.

Tom

Tom

---

A proud member of the OFA (Old Farts Association).

Gpu lagging is actually referring to too aggressive settings in the parameters, that cause "laggy" mouse movements and very slow keyboard inputs. What I see is when I type in an Editor, it can takes several seconds for any of my keyboard input to show up. Very annoying. Dial back the aggressive settings to some lower values until any "lagginess" disappears or gets below your tolerance threshold.

Yup, that is what I have experienced in the past with other command lines. Thank you for confirming, Keith.

Your ram is bottle-necked by the fact it is DDR4 ram not video-card ram (I think). At least at one time one of the reasons that video card ram was a lot more expensive is you could read/write "at the same time"? which jumped the performance a lot compared to the standard ram of the day. And may still.

(snip)

Presuming you have enough ram on the MB, there should be no hardware issue with running an APU ram space of 4GB and jumping your -sbs way up there. And then let it set at those settings for a couple of three weeks to try to get rid of our problem(s) with variable data helping us to jump to conclusions that may/maynot be warranted.

Yes, the DDR4 memory will be one of the bottlenecks (not sure if it the primary one or not), and I know the APU will never be as good as a discrete GPU, even from a few generations ago. My goal is to see if I have left any processing power on the table for the hardware I have. I do have 16GB of total system memory, so allocating 4 GB of it to the GPU definitely won't hurt memory usage elsewhere. I doubt it would ever use more than that, and honestly I would be surprised if it even used that amount.

To clarify my questions #4 and #5, it appears that the sbs is the amount of memory for one buffer. I am guessing the sbs cannot equal or exceed the total usable video card memory, whether it is a discrete card or not. I would also guess that the maximum allowable buffer size cannot be too big to hurt performance, but that is just speculation on my part without any research.

---

Seti@home classic: 1,456 results, 1.613 years CPU time

If you are speaking of using the standard SoG OpenCL app, you also face another restriction. You can't use more than 26% of the available VRAM on a discrete graphics card. So, 2048MB is the absolute maximum for a standard VRAM of 8GB on a 1070 for example. I have no idea what the restriction might be on a APU though that is accessing system memory for its VRAM requirements. I found that 2048 really didn't speed things up that much and settled on the good compromise of -sbs 1024 for my command line.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Interesting. Now I am wondering why I have about 1.7 GB of my video memory dedicated. I know there is a difference between being allocated and actually used, but I am not running anything else graphically intense while crunching.

---

Seti@home classic: 1,456 results, 1.613 years CPU time

Interesting. Now I am wondering why I have about 1.7 GB of my video memory dedicated. I know there is a difference between being allocated and actually used, but I am not running anything else graphically intense while crunching.

If you are talking about your 2200G host, then that seems exactly correct since it reports 7205MB of memory. 26% of that is your 1700MB of detected video memory or close enough. The API of OpenCL limits the available memory to 26%. At least the AMD API actually reports all the memory on card and not like the maximum 4GB of memory that the Nvidia API reports. There are constant posts aking why Nvidia cards don't show all their memory. Too long a story to delve into.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I have a Vega56 and it turned out that my overclocked Ryzen 1800x CPU is more efficient under Seti than my Vega56 undervolted and using optimized ATI app (and command line parameters) under Windows. If you don't care about efficiency (I mean performance/watts) than give it a try, it can do about 20k RAC I guess. But with constant 180-200W power consumption. Not a good deal.

---

Interesting. Now I am wondering why I have about 1.7 GB of my video memory dedicated. I know there is a difference between being allocated and actually used, but I am not running anything else graphically intense while crunching.

If you are talking about your 2200G host, then that seems exactly correct since it reports 7205MB of memory. 26% of that is your 1700MB of detected video memory or close enough. The API of OpenCL limits the available memory to 26%. At least the AMD API actually reports all the memory on card and not like the maximum 4GB of memory that the Nvidia API reports. There are constant posts aking why Nvidia cards don't show all their memory. Too long a story to delve into.

I think I'm more curious now why the APU reports 7205 MB of memory. I have 16 GB total for the system, and I am pretty sure 2 GB is assigned in the bios. At the very least HWinfo reports I have ~14 GB of used and available memory.

---

Seti@home classic: 1,456 results, 1.613 years CPU time

I think I'm more curious now why the APU reports 7205 MB of memory. I have 16 GB total for the system, and I am pretty sure 2 GB is assigned in the bios.

The 2GB reserved in the BIOS means it is for Video use only, it can't be used for system memory. However the rest of the system memory is available for use as video memory if it is required, up to a maximum of 7.2GB.

---

Grant
Darwin NT

Hello.
Sorry for resuming a slightly old thread.
Thank you so very much for starting this thread, however. Very interesting reading.
I had ran this project for a while on my rx570, however all of these commandline options are a little daunting and I see no mention of Polaris GPUs. I was a little discouraged at my RAC at the time but would like to try again.
A few questions:
1. What, exactly, do these numbers/options mean and how do they seem to impact performance?
2. Am I to assume that these options should be split in half e.g. 512 = 256, 64 = 32, if I want to try different, smaller values, or change them?
3. I do have Lunatics installed, so am I to look in a different directory for these specific options? There is a read me referenced but I am unable to find it. Maybe I need more coffee (or beer!) ;)
Referenced system is an AMD Ryzen 1800x running windows 10 and aforementioned rx570. It does have the slightest of overclocks to 1350 mhz over the default 1244.
thank you for any help and I do apologize if I missed something blatantly obvious pointed out in this thread already.

Read the read me first.
Its located in seti folder projects/docs.

---

With each crime and every kindness we birth our future.