Need Help Finding a Book about the Nitty-Gritty of Computer Hardware

[Kitten_Taco]

I'm a computer hobbyist, and I agreed to help someone build a computer. However, after scanning some sites, I realized that when I look at a product, such as RAM, I have no idea what most of the words in the item title (let alone the specs) are referring to. For example, "G.SKILL Ares Series 16GB (2 x 8GB) 240-Pin DDR3 SDRAM DDR3 2400 (PC3 19200)" . Obviously, I understand that its 2 8 gig sticks of RAM, and that its DDR3, but, what does the SDRAM DDR3 2400 (PC3 19200) mean? This is a reoccurring theme across the board with hardware. I feel like I'm missing out on buying high quality hardware because I don't understand the difference in clock speeds between to GPUs, or what difference latency makes between RAM. By knowing that these pieces of information are there, and not knowing what they mean, I feel like I'm really not "doing this right", despite having built PCs before. So, what are some books (or websites) that can help explain these things to me?
Thanks,
Kitten_Taco

 

[Kewlx25]

Wiki is a great place to start.

"DDR3 2400 (PC3 19200)" - This is just a way to say "DDR3 1866" is equivalent to "PC3 14900". Like saying "CRT (Cathode Ray Tube)".

DDR3 = PC3
1866 is the frequency = 14900 is the bandwidth
Frequency and bandwidth are directly related. A 10% increase in frequency will result in a 10% increase in bandwidth and visa versa.

Technically, 1866mhz is the external frequency on the DDR3 interface. The actual memory is running a at a much slower frequency but an inversely wider data path to make up the difference. An example would be a 100mhz chip that is 16bits wide can move 100mil * 16bits worth of data, or 1.6gigabits/s. The external bus may run at 800mhz but only be 2 bits wide, which is also 1.6gigabits/sec.

This is why DDR3 seems to have a higher latency than DDR1. The external frequency has gone up a lot , but the internal has increased very little. While DDR1 may have had 3 cycle latency, the bus was not "double or quad pumped". One cycle meant one bit. DDR3 does something like 4 bits per cycle, so it moves more bits. Internally the ram chip may still have 3 cycle latency, but because of more external cycles worth of data being moved, it gets listed as stuff like 9 cycle latency.

Latency means just that, how long. A 3 cycle latency means it takes 3 clock cycles. The ramifications on the other hand, which is what I assume you really mean by "what does it mean", depends on what you're doing. You can hide latency by increasing the number of requests in the same way 9 women can make it look like a baby is born every month when it really takes 9 months from start to finish.

When it comes to GPUs, you can't compare frequency. They are in a problem domain where throughput is king. All your care about is how many numbers can be crunched in a given time. Just look at benchmarks.

Power consumption increases linearly with frequency, but voltage must be increased with the sin of the frequency to maintain stability with increased frequency, and power consumption increases with the square of the voltage. In other words, power consumption increases dramatically as frequency goes up. Of course the scaling is not perfect and there is a floor and ceiling.