Fun discussion topics

[runt23]

Over the last few days some questions have popped into my head. Mostly about cpu's processing ability. Here is the list

How many "flops" can the modern cpu obtain. Why isnt there freeware to measure how many ours can do, like those supercomputers can.

How fast would a super computer do a superpi 1m?

Why dont video card companies, such as nvidia etc create a board that can "swap" conponents. Such as upgrade memory etc?

Here is a hard one, I understand programming "basics" such as programming to utalize multiple cores. However, I dont program I just understand the principles....

So....Is there a way to create a dual core processer, that utalizes "somehow" both cores for a program that is made for just one? I bet there are ways, but why havent they been created?

Why arent CPU's larger? This is probably an obvious question to others, but for some reasons it would be logical for heat, and to fit "more" on there. So do they shoot for smallest when creating?

Why doesnt Tomshardware do tests....Well, what I mean is testing on myths and complaints. Such as "My AW8D motherboard doesnt work with OCZ memory" complaints we have day to day. And that seem to echo through out forum topics. It would be interesting to see some reviews of "not working" things. Maybe it would push for some better company backing in customer service on these items.

How low would I have to clock X CPU in order for it to run WITHOUT a cooling fan?

Testing benchmarks with different drivers for video cards. Maybe also testing programs with different video card bios.

A lot here, but just some thoughts =)

Have fun

 

[1Tanker]

Why doesnt Tomshardware do tests....Well, what I mean is testing on myths and complaints. Such as "My AW8D motherboard doesnt work with OCZ memory" complaints we have day to day. And that seem to echo through out forum topics. It would be interesting to see some reviews of "not working" things. Maybe it would push for some better company backing in customer service on these items.

How low would I have to clock X CPU in order for it to run WITHOUT a cooling fan?

Testing benchmarks with different drivers for video cards. Maybe also testing programs with different video card bios.

A lot here, but just some thoughts =)

Have fun

Well, i hate to have to break it to you, but Tom's runs this site to make money. They can't afford to buy every mobo/RAM/HD that are incompatable for people, and they would have to hire testers to run these tests. If they did this, they would have to run more advertisements/banner ads to pay for this, and judging by threads like this, people wouldn't like that.

http://forumz.tomshardware.com/hardware/modules.php?name=Forums&file=viewtopic&t=192131&highlight=stop+spam

 

[voxel]

How many "flops" can the modern cpu obtain. Why isnt there freeware to measure how many ours can do, like those supercomputers can.

FLOPs measurement isn't very useful for consumer CPUs. If you vectorize code (SSE) the FLOPs numbers can be very high, but very few PC apps use SSE fully. Integer calculations per second is probably a better metric for daily use.

How fast would a super computer do a superpi 1m?

If the algorithm can be vectorized - maybe in one CPU cycle! If it couldn't, then probably as fast as one CPU in the supercomputer could process it.
Apples vs. Oranges question. Supercomputers are for vectorizable algorithms that can be parallelized.

Why dont video card companies, such as nvidia etc create a board that can "swap" conponents. Such as upgrade memory etc?

Cause they keep changing the memory type all the time! Memory tech gets faster and faster...

Here is a hard one, I understand programming "basics" such as programming to utalize multiple cores. However, I dont program I just understand the principles....

Algorithms need to be parallelized (i.e threads or multiple operations in one cycle) to take advantage of multi-cores.

So....Is there a way to create a dual core processer, that utalizes "somehow" both cores for a program that is made for just one? I bet there are ways, but why havent they been created?

Engineers have tried - with some success. The problem is that code + data is interdependent: If A then B Else C. So trying to auto-schedule stuff kinda of works, but not always. Branch prediction is pretty standard in modern CPUs (but removed in the Itanium and Cell). The CPU guesses which path B or C and executes that code BEFORE "If A" finishes. But if the CPU guesses wrong it has to throw away the old results. Read about Superpipeline and Superscalar architectures.

Why arent CPU's larger? This is probably an obvious question to others, but for some reasons it would be logical for heat, and to fit "more" on there. So do they shoot for smallest when creating?

Probably related to yields, etc. I'm totally guessing this. Larger chips - more entropy - possibility of errors.

 

[Pain]

A lot of times the manufacturers already have a qualified hardware list that says what memory is tested to work with the the board. People should probably start with that. :!:

 

[nategarst]

A lot of times the manufacturers already have a qualified hardware list that says what memory is tested to work with the the board. People should probably start with that. :!:

LOL, why do that. Starting with the manual would involve reading.

 

[yourmothersanastronaut]

Hate to break it to you, but if it solves a lot of headaches, reading the manual is a good thing. Whether it's on a page or a computer screen, it's still text.

Corsair works great with my Asus. OCZ doesn't work with Asus AM2 boards - every review I've read combining the two said they didn't work.

I guess it's a good thing I prefer Corsair to OCZ...