how many gigabits per second does a CPU need

[The_Promised_End]

I was wondering if we can almost use a cpu without bottlenecks at 40 gigabits per second then what would we need if we were running a cpu not a gpu

[Ezzy-525]

[The_Promised_End]

5 minutes ago, Ezzy-525 said:

lol

[tp95112]

Um what?

[Enderman]

Did you mean to say GPU?

Who said anything about 40Gbps?

 

[The_Promised_End]

5 minutes ago, Enderman said:

Did you mean to say GPU?

Who said anything about 40Gbps?

 

thunderbolt egpus linus did a video on this recently

[ARikozuM]

I have no idea what you're asking... 

[The_Promised_End]

5 minutes ago, ARikozuM said:

I have no idea what you're asking... 

basically how many gigabits per second does a cpu use under full load

[Cebolonha]

Trying really hard to understand what IS the question.
 

"I was wondering if we can almost use a cpu without bottlenecks at 40 gigabits per second then what would we need if we were running a cpu not a gpu"

Let's break it down:

"if we can almost use a cpu without bottlenecks at 40 gigabits per second" - Assumes 40gb/s is near max throughput of CPU - Which CPU? Where does this information come from? I don't know much about hardware, so... let's assume that to be true. Is this applicable to CPUs? Don't we measure operations/clock cycle instead? Doesn't I/O depend on other stuff? I understand we can, like, "sum a list of numbers up to 30GB/s in a modern CPU", but... what are you even doing?

"then what would we need" - Need for what? Is 40gb/s not enough for you? What do we need? That's sort of a deep question. Or undefined.

"if we were running a cpu not a gpu" - Wait, what? I assumed you were running a CPU to begin with. Where does this GPU come from? 

[Lurick]

2 minutes ago, The_Promised_End said:

basically how many gigabits per second does a cpu use under full load

You don't measure CPU power in bits per second.

You measure it in Gigaflops, Petaflops, or Teraflops

[Jurrunio]

I think this guy is so new to technology he doesn't even know the differences between a CPU, or a GPU, and he's now asking whether Thunderbolt 3 can give enough bandwidth for an external GPU

[The_Promised_End]

5 minutes ago, Cebolonha said:

Trying really hard to understand what IS the question.
 

"I was wondering if we can almost use a cpu without bottlenecks at 40 gigabits per second then what would we need if we were running a cpu not a gpu"

Let's break it down:

"if we can almost use a cpu without bottlenecks at 40 gigabits per second" - Assumes 40gb/s is near max throughput of CPU - Which CPU? Where does this information come from? I don't know much about hardware, so... let's assume that to be true. Is this applicable to CPUs? Don't we measure operations/clock cycle instead? Doesn't I/O depend on other stuff? I understand we can, like, "sum a list of numbers up to 30GB/s in a modern CPU", but... what are you even doing?

"then what would we need" - Need for what? Is 40gb/s not enough for you? What do we need? That's sort of a deep question. Or undefined.

"if we were running a cpu not a gpu" - Wait, what? I assumed you were running a CPU to begin with. Where does this GPU come from? 

lol what i ment is this we can use a gpu almost with out bottle necks at a speed of 40 gigabits per second could we do this with a cpu and how much more bandwidth would it need

[BuckGup]

OP is asking what's the bandwidth the CPU uses over say the motherboard

[The_Promised_End]

5 minutes ago, Jurrunio said:

I think this guy is so new to technology he doesn't even know the differences between a CPU, or a GPU, and he's now asking whether Thunderbolt 3 can give enough bandwidth for an external GPU

no no no just am not articulating well sorry

[The_Promised_End]

5 minutes ago, BuckGup said:

OP is asking what's the bandwidth the CPU uses over say the motherboard

yes thanks

[Jurrunio]

1 minute ago, The_Promised_End said:

lol what i ment is this we can use a gpu almost with out bottle necks at a speed of 40 gigabits per second could we do this with a cpu and how much more bandwidth would it need

You can't use a CPU without a motherboard (at least no one has tried to make it external), so this question is meaningless

[The_Promised_End]

5 minutes ago, Jurrunio said:

You can't use a CPU without a motherboard (at least no one has tried to make it external), so this question is meaningless

yes but do you know the answer

[Cebolonha]

Just now, The_Promised_End said:

lol what i ment is this we can use a gpu almost with out bottle necks at a speed of 40 gigabits per second could we do this with a cpu and how much more bandwidth would it need

The thing is, we don't measure CPU and GPU speed in terms of transfer speed or I/O. We measure in operations per clock cycle and frequency, and FLOPS/sec, but the latter is only "floating points per second", which is only one kind of operations among others, but is very important anyway. 

[The_Promised_End]

6 minutes ago, Cebolonha said:

The thing is, we don't measure CPU and GPU speed in terms of transfer speed or I/O. We measure in operations per clock cycle and frequency, and FLOPS/sec, but the latter is only "floating points per second", which is only one kind of operations among others, but is very important anyway. 

not the speed but how much data it is pulling in and out of the mobo socket

[Lurick]

The most powerful consumer grade CPU on the market today that you can buy, has a maximum theoretical bandwidth on the Memory Controller of 76.8GB/s (GigaBytes) which equates to 614.4Gb/s (Gigabits)

 

https://en.wikichip.org/wiki/intel/core_i7ee/i7-6950x

 

As I said before though, you do not measure a CPU's power in gigabits, that's not how it works.

[ac130kire]

I think I understand what you mean. You are wondering if you could add another CPU in an external solution. Unfortunately bandwidth isn't really the main problem here. It mainly has to do with the interface that is being used. Other people mentioned Thunderbolt 3 so I will go off of that. over PC Thunderbolt 3 communicates directly to the CPU over PCIe lanes and these are not meant for this type of CPU to CPU communication.  Additionally there would be bandwidth problems with even memory.  RAM needs a ton of bandwidth way more than 40Gbps and needs to be a very short distance from the CPU in order to minimize latency and not waste clock cycles while accessing it.  On the problem of latency, the CPU's need to be a small distance away from each other in oder to keep latency down.

 

I'm sure there is more that I'm forgetting to mention, but this is the gist of it.

[Enderman]

24 minutes ago, The_Promised_End said:

thunderbolt egpus linus did a video on this recently

You said "can almost use a cpu without bottlenecks" in your post.

No GPU needs 40Gbps.

Also how much bandwidth a CPU uses completely depends on the workload.

[Cebolonha]

26 minutes ago, The_Promised_End said:

not the speed but how much data it is pulling in and out of the mobo socket

You can give the CPU just 64bits of data and make the CPU work under 100% load for days. That's why this doesn't matter much. And the result of said process might be just 1 byte of data. The max throughput of an algorithm like this one is almost meaningless.

Take, for example, Folding@home. They give you some parameters and a program on which you should donate CPU time, but they aren't doing a whole lot of I/O, if any.

I guess you can measure how much data a process is moving around inside the CPU, but that depends on the operations being executed: some are faster than others. 

Now, achieving max throughput using the least CPU possible IS what you want, and that is really important for web servers and web frameworks. You generally want to saturate your network card before achieving 100% cpu load. 

Edited June 4, 2017 by Cebolonha
gremmar

[The_Promised_End]

23 minutes ago, Cebolonha said:

You can give the CPU just 64bits of data and make the CPU work under 100% load for days. That's why this doesn't matter much. And the result of said process might be just 1 byte of data. The max throughput of an algorithm like this one is almost meaningless.

Take, for example, Folding@home. They give you some parameters and a program on which you should donate CPU time, but they aren't doing a whole lot of I/O, if any.

I guess you can measure how much data a process is moving around inside the CPU, but that depends on the operations being executed: some are faster than others. 

Now, achieving max throughput using the least CPU possible IS what you want, and that is really important for web servers and web frameworks. You generally want to saturate your network card before achieving 100% cpu load. 

i dont really understand but thanks for trying to help if you could try to explain to an 8 year old

[Cebolonha]

4 minutes ago, The_Promised_End said:

i dont really understand but thanks for trying to help if you could try to explain to an 8 year old

You can have a really really long cooking recipe that uses only 1 or 2 ingredients. The dish takes hours and hours to get ready to eat. 

You can also have a really short recipe that uses 15 ingredients. Just mix them together and it's done in 5 minutes.

Recipe is the program, ingredient is the data your program uses.