By the way, I think there was one interesting thing when I created another Google instance and attached it to Albert. It looked like project server didn't initially recognize that new instance correctly. What I saw at the project account page at first was a mixture of name of the new instance + credits of the old one. Old instance wasn't shown at all.
Isn't it normally that a new machine gets a light start and it is sent only a few tasks to begin with? This second instance got about 14 tasks right away (like it had been a veteran) and begun crunching at full power.
So I contacted the first instance and forced it to perform a project update. Then I made that same for the new instance. After all that I reloaded Albert account page and now the new instance was shown correctly (with zero credit). Everything else was also correctly displayed.
Server had cancelled 12 of the initial tasks for the new instance, because they didn't obviously belong to it so quickly. It had been left with 4 tasks to begin with... and now I see it has got one more. It's gradually given a full workload.
Ps. I had planned to launch a Windows instance for comparison, but it was more than double the price of an instance with Ubuntu image.
Windows Server 2012 R2: $407.30 per month ... and that's mainly because of the usage fee* ($233.60 / month)
* Image usage fee is charged by Microsoft and billed by Google
Ubuntu 16.04 LTS: $172.10 per month
So a Windows instance would'd run only about three weeks. I got greedy and chose Ubuntu again, because it will be more "efficient". Please forgive me.
I tried several free trials on cloud services and it worked pretty good
* Google Cloud
* MS Azure (I don't have free credit anymore, so I can't test it, but I'm interested to hear from anyone who tested or will test the GPU option in Azure)
* digital ocean
* IBM Bluemix (this interface is confusing, but with little bit of effort you can set it up)
* softlayer (cancelling your account is difficult and the interface is also confusing. But the helpdesk helped me with cancelling the account. So no problem here. They even let my vps running for weeks after cancelling)
* VMware service was too complicated. I didn't get it working.
* vpsie
* vultr
* cloudsigma was also too difficult to setup (they also use VMWare I think. The problem is I wasn't able to make a network connection. But this is interesting because they have GPU support).
I didn't have any credit card problems with the above mentioned services. But if you test it, you should be cautious of course. It is important you search for reviews of the services you want to test to see what the reaction are of previous users
I decided not to test the following: Oracle, HP, fujitsu, linode
This Linux desktop host is running 9 tasks concurrently at the moment and they seem to take ~42k secs to complete, but with 8 tasks it was ~38k. No GPU work.
I see the details for Linux host says "Cache 12288 KiB", but for all similar Windows hosts it says "256 KiB". I wonder why.
The Intel website suggests 12Mb is correct. You might find odd things such as running 7 in parallel completed say 56 tasks faster than running 8 in parallel.
By the way, I think there was
)
By the way, I think there was one interesting thing when I created another Google instance and attached it to Albert. It looked like project server didn't initially recognize that new instance correctly. What I saw at the project account page at first was a mixture of name of the new instance + credits of the old one. Old instance wasn't shown at all.
Isn't it normally that a new machine gets a light start and it is sent only a few tasks to begin with? This second instance got about 14 tasks right away (like it had been a veteran) and begun crunching at full power.
So I contacted the first instance and forced it to perform a project update. Then I made that same for the new instance. After all that I reloaded Albert account page and now the new instance was shown correctly (with zero credit). Everything else was also correctly displayed.
Server had cancelled 12 of the initial tasks for the new instance, because they didn't obviously belong to it so quickly. It had been left with 4 tasks to begin with... and now I see it has got one more. It's gradually given a full workload.
Ps. I had planned to launch a Windows instance for comparison, but it was more than double the price of an instance with Ubuntu image.
Windows Server 2012 R2: $407.30 per month
... and that's mainly because of the usage fee* ($233.60 / month)
* Image usage fee is charged by Microsoft and billed by Google
Ubuntu 16.04 LTS: $172.10 per month
So a Windows instance would'd run only about three weeks. I got greedy and chose Ubuntu again, because it will be more "efficient". Please forgive me.
Hi,I tried several free
)
Hi,
I tried several free trials on cloud services and it worked pretty good
* Google Cloud
* MS Azure (I don't have free credit anymore, so I can't test it, but I'm interested to hear from anyone who tested or will test the GPU option in Azure)
* digital ocean
* IBM Bluemix (this interface is confusing, but with little bit of effort you can set it up)
* softlayer (cancelling your account is difficult and the interface is also confusing. But the helpdesk helped me with cancelling the account. So no problem here. They even let my vps running for weeks after cancelling)
* VMware service was too complicated. I didn't get it working.
* vpsie
* vultr
* cloudsigma was also too difficult to setup (they also use VMWare I think. The problem is I wasn't able to make a network connection. But this is interesting because they have GPU support).
I didn't have any credit card problems with the above mentioned services. But if you test it, you should be cautious of course. It is important you search for reviews of the services you want to test to see what the reaction are of previous users
I decided not to test the following: Oracle, HP, fujitsu, linode
Richie_9 wrote: It's this
)
The Intel website suggests 12Mb is correct. You might find odd things such as running 7 in parallel completed say 56 tasks faster than running 8 in parallel.
You might find help with the PCM tools from Intel - in this CPU only cruncher thread which i used to investigate a similar problem.
The PCM tools show what resource is being bottlenecked, if some cores are running slow etc.
Good luck