xtalker
xtalker copied to clipboard
Spycsh
Trying to get my hands on Sapphire SPR CPUs, which Cloud Instance to choose from? (M7i?)
Hello I am trying to test this repo with the mentioned CPU,
After some research I discovered that the Saphhire SPR might be called M7 ? or M7I ?
Could you confirm? If not, then which one should I choose? Is there a smaller CPU than can do the same? or alternatives? (bigger and smaller) And finally, where else could I rent it? Thanks
Edit: I am also noticing there are MANY M7i instances types, ones with more or less vCPU and Memory, I don't know which to choose from?? edit2: source = https://calculator.aws/#/addService/ec2-enhancement?nc2=h_ql_pr_calc
Hi @Spycsh thanks for the answer, 10 dollar a hour, is too much for me! Question: If a borrowed this CPU, do you think I could use it to run MULTIPLE "xtalker" processes at once? I need to make sure my money is worh it, or is it only ONE xtalker possible with one m7i.48xlarge? I need to find ways to multiply even more than x10 the speed you mentioned, maybe by having parallel xtalkers running if its possible? Maybe you can explore this
@AIhasArrived ,
If a borrowed this CPU, do you think I could use it to run MULTIPLE "xtalker" processes at once?
sure you can. xtalker basically bind different cores to processing different image + audio frames. I totally agree that for personal user, SPR is too expensive. What I suggest that you might do is to rent a relatively lightweight CPU and see if it works. And then you can try to apply parallelism. Basically the more cores you have, the more parallel instances you can have and the more speedup you can obtain.
Both m7i.48xlarge and m7i.xlarge have the same CPU "Intel(R) Xeon(R) Platinum 8488C". What makes this work on the 48x instances but not the smaller one, considering they have the same CPU? Yes, there are more CPUs in one than the other but running with just 2 CPUs on the 48x large speeds things up by two while on the smaller box this is not the case - the speed stays constant. Thanks
To partially answer my own question: From https://aws.amazon.com/blogs/aws/new-seventh-generation-general-purpose-amazon-ec2-instances-m7i-flex-and-m7i/ "The Intel QAT, Intel IAA, and Intel DSA accelerators will be available on the m7i.metal-24xl and m7i.metal-48xl instances." So I assume it is one of those: QAT, IAA, or DSA which is only enabled on the entire box and not the smaller ones. I am thinking it is DSA (as opposed to QAT or IAA) that is required for these to gain efficiency by running parallel.
To expand on this further: You need to get a whole metal Saphire Rapids server to get the speed ups talked about here. For my use case using the AWS r7iz.metal-16x gave me results in 92 seconds and used half of it's CPUs. OVH offers servers of this size for about $570 / month. AWS is much more expensive if you plan on keeping it always on. The other options is a GPU - I tested AWS' g3s.xlarge and it gave the best bang for the buck but was twice as slow as running on half of the 16x metal box.