AWS Ice Lake Comparison: Benchmarks and Insights

Key Takeaways

what's up everybody wanted to really quickly make a video walking through some of the facts and details behind the Intel c6i ice Lake processor and how it operates on AWS specifically ec2 instances so in this video we're going to walk through some of the facts behind uh the ice Lake chips performance we'll kind of compare that performance to other GPU instances to better understand what the relationship between speed and cost looks like and then we will finish with a quick summary of use cases in which the ice lake is a great thing to pick so with that started a little history right so the ice Lake chip also known as the c6i is one of Intel's latest chips it's very powerful and in this video we're going to be comparing the 2x large instance in our comparisons with the GPU units so really quickly for those not familiar with AWS there's a formatting so everything to the left of the period is the name of the processing unit so in this case right so the c6i that's our processing unit and everything to the right of the period is the size of the instance so there's large x-large 2x large Etc and the the bigger the instant size you get the more resources you have so for example we are focusing on the 2XL large sorry 2x large which has access to eight vcpus at virtual CPU if you recall ec2 instances are a remote server that is provided to you by the AWS service so the idea is the 2x large gets you eight virtual CPU units and it also gets you 16 gigabytes which is very similar to gigabytes they're just mathematically altered to match reality right so basically 16 gigabytes in memory which is a really powerful useful size you can do a lot of things with this much resources specifically though let's see how performance performing computer vision inferences Stacks up for the CPU units versus the GP units which we have down here so this column this is our isolate column and the most interesting things to be paying attention to are first this one second is this one and then third is this one which we're going to talk about instant control order now so what is this row right here this row is our FPS this is the frames per second at which our Processing Unit can process uh images and get computer vision results right so a process is you give us an image we perform computer vision on it and we tell you what's in that image right and the inference results are I detected a car I detected a person whatever they may be so in this case I'm comparing across the instances there are all 2x large and the only thing that changes is the processing unit and the GP unit in each one so let's walk through some of these results first and foremost we'll talk about speeds you will notice that of the four instances compared here that the ice lake is the slowest of the four which is okay because in a moment we're going to be talking about price which is equally important and the idea is that as you switch over to gpus right so from this particular CPU unit which is the ice Lake to the T4 Nvidia graphics card the speed does increase on the GPU and as we increase the size of the GPU it also increases in speed but more importantly is speed increases but it provides diminishing returns in relation to the price so for example between the lowest and the highest is about a factor of 10x right so the ice Lake on demand pricing is 34 cents for the V100 which is our most powerful tested uh graphics card in this experiment uh the price difference is about a factor of 10 whereas the speed difference is a factor of about two right so um that's always something to keep in mind is that speed typically does not scale with price and with that said we can get a analysis and overall analysis of the speed to cost ratio down here in this final row so for the ice Lake what you get for the price point to speed ratio is a rating of 152. we get this rating by just taking the FPS and dividing it by its cost and you'll notice that the ratio of Effectiveness or Price efficiency or however you want to think about it tends to decrease or slightly increased I guess in this case right it goes from 94 to 95 and then back down to 40 with price but the important thing here is that the trend kind of goes down into the right like so so what this tells us is that if you're looking for an instance that has really great processing power at a reasonable price the ice Lake offers you more than enough features and functions and speed to get the majority of use cases done for an effective price as you run into more Niche sectors in which you have to really get specifically High complex video streams you know multi-parallel video streams going on you might potentially need more FPS but in a general purpose setting we're confident with saying that the ice Lake chip does offer a better experience in terms of ease of use price and scalability and the long-term run of things and uh if you want to learn more about the ec2 c6i instance it's very easy to get started all you have to do is essentially grab one of these instances and spin it up on AWS right so here I have a AWS ec2 c6i 2x large instance running once I have this instance running it's just a virtual machine so I could remote into it I can get my application running on it I can set up AWS batch I could turn it into an endpoint whatever I want to do with it I can serve customers at a really effective balance of speed to price here and um if the you know too many customers start to interact all I have to do is spin up more instances and the more ice Lake instances I spin up the more cost effective I get so as I scale I save money and I serve customers I hope this video has been useful to everybody and if anybody has any questions or comments please leave it in the blog post tied to this or the YouTube channel and uh I'm looking forward to seeing those responses have a good one guys see ya