Einstein@Home | Aborting task h1... : exceeded elapsed time limit...

Yes! I see your host has some FGRP5 and GW CPU tasks. They should run fine and your Ryzen is very suitable for that.

Running CPU tasks along with GPU might slow down the GPU tasks some amount, but I don't have personal experience with Ryzens on that. Also running 100% of the cores might introduce some instability (which might not mean reboot or such at all, but could manifest itself thru invalid validations, perhaps for the GPU tasks). But that is to be seen and could be fixed with lowering the allowed core amount to 75 % for example in Boinc.

Boinc scheduler will consider what kind of task it will send to that host when both CPU and GPU work is accepted. It should settle eventually into situation where it will feed the host enough both types of work so that neither type will run empty. That is if there won't be problem with completing tasks before deadlines. I believe the best configuration for your host would be to run only GW CPU tasks (for the CPU) while it is running FGRP5 work for the GPU. I'm not sure about this (can't remember), but I think that running FGRP5 tasks for CPU + FGRPB1G tasks for GPU makes the scheduler sort of wander, because completion times for tasks are constantly so much different between CPU tasks and GPU tasks. Here in Einstein this system works in a way that within a search type (Gamma-ray... in that scenario) the scheduler can't separate those CPU and GPU things too well. But it should find a working balance after a while.

The easiest way to do that is by visiting again https://einsteinathome.org/account/prefs/project . There is 'GPU utilization factor of FGRP apps' and default value is 1.00

1.00 means the whole GPU is reserved for one task. If you change that to 0.50 it will allow two FGRPB1G tasks to "fit" in the GPU at a time, as long as it has enough free VRAM. If I recall right 2GB should be enough for three tasks. Setting 0.33 would allow three tasks. Then click save changes. This won't lead to instant change for the running tasks. it will require at least one new task to be downloaded until they start to run concurrently.

The latter.  The 1 CPU + 0.5 GPU description you mention should already have budgeted for 2 CPU threads to be available to support 2 GPU tasks.

To determine why you don't have two GPU tasks running, you need to give more details about how you are trying to operate.  Here are some questions.

1. Are you running CPU tasks for projects other than Einstein?
2. Are you running GPU tasks for projects other than Einstein?
3. Your machine is listed as 8C/16T.  How many of the 16 threads have a running CPU task of any description?
4. Are any of those running CPU tasks under deadline pressure?  BOINC would run such tasks first.
5. What are your work cache settings - how many days for each of the two values?
6. To allow concurrent GPU tasks, are you using GPU utilization factor or the app_config.xml mechanism?
7. If it's the former (as was recommended), has new work downloaded since you last changed the setting?

You mentioned that your machine was new.  I'm a little surprised that a new machine came with a low end and 'old architecture' GPU.  It's relatively unsuited to GPU crunching and I very much doubt that it would show any gain in performance by running concurrent GPU tasks.  The validated results that show so far show an elapsed time ranging from around 1.5 hours to almost 2.5 hours.  That amount of variation indicates a machine under real stress since the gamma-ray pulsar GPU tasks are known to be very uniform in crunch time.

To put that into perspective, a modern low to mid-range GPU should be able to crunch these tasks in around 20 mins or so (and rather less if you get into the mid-range).  As an example, I have Polaris series (RX 460 which really are low end) which take about 23 mins singly or around 20 mins per task when running two at a time.  I purchased these in early 2017, almost 3 years ago.  Your architecture goes back a lot earlier than that.

Rather than trying to get 2 concurrent tasks crunching, you should first work out why the huge variation when running singly.  I would guess you may be running too many individual CPU tasks.  I saw one bit of evidence that supports this.  A single validated CPU task where the run time was significantly longer than the CPU time and both times were rather longer than what you would expect for your quite capable machine.  A total elapsed time getting up towards 2 days seems rather excessive.

I would suggest a simple test.  Use the controls in BOINC manager to suspend all the waiting CPU tasks (including all tasks for other projects) so that you just have the current running tasks and the waiting Einstein GPU tasks.  Start suspending the currently running CPU tasks one by one and see if a second GPU task springs into life.  The very first CPU task suspended should allow a 2nd GPU task to start.  If it does, then BOINC was favouring the CPU tasks for some reason.  If it doesn't there is likely something misconfigured somewhere.

You could probably extend the time you leave CPU tasks suspended to allow you to measure any increase in the rate of progress of GPU tasks.  That would help you work out the most efficient operating conditions.  Since the rate of progress of GPU tasks should be quite uniform, even 'back of the envelope' type calculations as the task progresses should be able to give you some idea as to whether you can do better than the times you are already seeing.