Why is N1's linpack score so much higher than droid?

[excalion]

I thought linpack was supposed to measure raw processor speed, which is capped? So why does N1 bench so much better than a droid if both are at 1ghz? It is because of intermediaries between the raw unprocessed data and the processor? By intermediaries, I mean stuff like JIT. So is the N1 just better engineered to utilize the 1ghz than the droid?

This isn't a bragging rights question, I'm just curious why it works the way it does.


Answer:

Your question was answered already...it comes down to processor architecture. Qualcomm's Snapdragon platform and more specifically the Scorpion application processor, while being related to TI's Omap Arm v7 series, has enhancements made by Qualcomm. The part of the cpu that handles the SIMD instructions has a wider pipeline, 128 bits vs 64 bits in TI's Omap. Scorpion also has a deeper pipeline to better handle all that data which I'd assume offsets some of the performance benefits a little bit.

As far as the difference between the two benchmarks, they are written to benchmark two different things. Linpack can run almost entirely within the SIMD/NEON portion of the cpu thereby showing off the enhancements made by Qualcomm. Quadrant stresses the entire core showing off total system performance showing that only in certain situations will Snapdragon outperform any other Arm v7 based core.

 

[compywiz]

Sort of guessing here, but I'd say part of the reason is because the clock speed isn't everything. A 1ghz ARM processor just doesn't have the same output as a 1ghz Snapdragon. Just because they are both at 1Ghz doesn't mean they will perform equally.

 

[jl8851]

You must be comparing an n1 on 2.2 to a droid on 2.1

OMAP processor on the X blows away the snapdragon


EDIT: Thought you were talking about the X, nevermind

 

[excalion]

Sort of guessing here, but I'd say part of the reason is because the clock speed isn't everything. A 1ghz ARM processor just doesn't have the same output as a 1ghz Snapdragon. Just because they are both at 1Ghz doesn't mean they will perform equally.

Yea that's understandable.

What I don't understand is why a 1.25ghz droid can score about the same as an N1 on quadrant processor benchmarks, but only get ~19 mflops when N1 gets 35+.

________________quadrant(processor score)______________________linpack
droid(2.2)________________5788______________________________~19 mflops

N1(2.2)_________________~5500(source)_______________________35+ mflops


This is what I don't get.

 

[SwiftLegend]

Different hardware means different results. That's like comparing a stock Intel CPU to a stock ATI, the Intel will most likely win.

 

[excalion]

Different hardware means different results. That's like comparing a stock Intel CPU to a stock ATI, the Intel will most likely win.

I'm not asking IF different hardware means different results. I'm asking WHY different hardware means different results when they are both at the same speed? As I understand it, the electrons running through the processor chip turns many "nodes" inside the processor on and off. The speed at which these electrons flow through the chip is the clock speed(X mhz).

So if both chips are having their nodes turned on and off at the same rate/speed, why is one faster than the other? The only possible conclusion I can see lies in the difference between the efficiency of an "intermediary" between the raw data and the processing unit. Does such an intermediary exist? If it possible to access this portion of the processor through conventional means? Why is the droid's intermediary running at half the efficiency of N1's?(According to linpack) These are the questions I'm asking.

 

[lockdownx1x]

Two Words: CPU Architecture. GHz mean nothing.

 

[SwiftLegend]

Two Words: CPU Architecture. GHz mean nothing.

Exactly. Like I said, different hardware means different results, asking why is kinda >.>. Again with the computer analogy, a 1.2GHz Intel in a Mac vs. a 1.2GHz ATI in a PC, the Mac will surely win. Different companies build their products different, thats what causes competition.
A Snapdragon is a higher grade cpu unlike the ARM. If I remember correctly, the Droid is supposed to run at 550 mhz, and we are overclocking it to 1 GHz, you're soft-modding the CPU. You really think a 550 MHz processor forced to run at 1 GHz can compare to an originally 1 GHz processor?

 

[takeshi]

The only possible conclusion I can see lies in the difference between the efficiency of an "intermediary" between the raw data and the processing unit.

Faulty assumption. Processor architecture absolutely matters. Clock speed is just clock speed.

You really think a 550 MHz processor forced to run at 1 GHz can compare to an originally 1 GHz processor?

There's also a faulty assumption there. It's certainly possible.

 

[excalion]

Two Words: CPU Architecture. GHz mean nothing.

Exactly. Like I said, different hardware means different results, asking why is kinda >.>. Again with the computer analogy, a 1.2GHz Intel in a Mac vs. a 1.2GHz ATI in a PC, the Mac will surely win. Different companies build their products different, thats what causes competition.
A Snapdragon is a higher grade cpu unlike the ARM. If I remember correctly, the Droid is supposed to run at 550 mhz, and we are overclocking it to 1 GHz, you're soft-modding the CPU. You really think a 550 MHz processor forced to run at 1 GHz can compare to an originally 1 GHz processor?

Two things.
1. The reasons why droids are clocked at 550mhz instead of 1ghz are many, but the limitations of the chip itself are NOT one of them.

2. Quadrant benchmarks both chips to be about the same. So yes, a 550mhz OC'd chip CAN run just as well as an originally 1ghz chip.

I appreciate you trying to help, but seriously though, if I just wanted some very vague and basic answers I'd just google it. I'm looking for an answer from an electrical engineering perspective. Are there any tech gurus out there who can answer this question with elaborate details?

The only possible conclusion I can see lies in the difference between the efficiency of an "intermediary" between the raw data and the processing unit.

Faulty assumption. Processor architecture absolutely matters. Clock speed is just clock speed.

So why can't the CPU architecture be the intermediary between the nodes being turned on and off by the electrons, and the raw unprocessed data? How is my assumption a faulty one?

 

[HolyGrail]

People keep measuring processors by their mhz, which isn't accurate. You have to measure processors by their clock cycles. The Snapdragon has a higher clock cycle.

Clock cycle = Instructions per cycle. Clock speed = Cycles per second (hz). Its both of them being multiplied. A core 2 duo at 3gz is going to be slower then a I7 at the same speed. The number of instructions per second for a processor can be derived by multiplying the instructions per cycle and the clock speed (measured in cycles per second or Hertz [Hz]) of the processor in question. The number of instructions per second (Clock Cycles) is an approximate indicator of the likely performance of the processor.

 

[excalion]

People keep measuring processors by their mhz, which isn't accurate. You have to measure processors by their clock cycles. The Snapdragon has a higher clock cycle.

Clock cycle = Instructions per cycle. Clock speed = Cycles per second (hz). Its both of them being multiplied. A core 2 duo at 3gz is going to be slower then a I7 at the same speed. The number of instructions per second for a processor can be derived by multiplying the instructions per cycle and the clock speed (measured in cycles per second or Hertz [Hz]) of the processor in question. The number of instructions per second (Clock Cycles) is an approximate indicator of the likely performance of the processor.

I see, thank you for your explanation, but this still doesn't explain why quadrant benchmarks both chips to be about the same. If the N1 chip is faster overall(2x faster according to linpack), why is quadrant's CPU test saying they're about the same?

 

[Gimic26]

Your question was answered already...it comes down to processor architecture. Qualcomm's Snapdragon platform and more specifically the Scorpion application processor, while being related to TI's Omap Arm v7 series, has enhancements made by Qualcomm. The part of the cpu that handles the SIMD instructions has a wider pipeline, 128 bits vs 64 bits in TI's Omap. Scorpion also has a deeper pipeline to better handle all that data which I'd assume offsets some of the performance benefits a little bit.

As far as the difference between the two benchmarks, they are written to benchmark two different things. Linpack can run almost entirely within the SIMD/NEON portion of the cpu thereby showing off the enhancements made by Qualcomm. Quadrant stresses the entire core showing off total system performance showing that only in certain situations will Snapdragon outperform any other Arm v7 based core.

 

[ewells2420]

The only possible conclusion I can see lies in the difference between the efficiency of an "intermediary" between the raw data and the processing unit.

Faulty assumption. Processor architecture absolutely matters. Clock speed is just clock speed.

You really think a 550 MHz processor forced to run at 1 GHz can compare to an originally 1 GHz processor?

There's also a faulty assumption there. It's certainly possible.

heh yeah overclocking doesnt really mean much. It will increase processing power and results, however it doesnt make it efficient. Most cpus are made on basically a sheet ..... due to differences in micro-archatecture (I cant spell) and manufacturing on one sheet of cpus .... not all are created equal.

They are tested and labeled according to the output they can match set by the company. So if you are selling Q6600 cpus that run 2.4gighz stock with a multiplier of say 8. Then all chips must be able to run 2.4 gighz FSBx8 to reach 2.4gighz at a set voltage to ensure long life. The process of overclocking is just forcing that chip to run faster by pumping more voltage into it, until it runs stable. This in no way makes it inferior to a chip 'designed' to run faster or as fast as a OCd CPU.

All other chips are then marketed as a step slower. Thats why you have say a E6600 and a Q6600. Although thats a bad example as the first is dual corp and second is quad. But I think you get the point.

So basically what it comes down to is how well that CPU can process a set of instructions like SSE (I believe thats one of em, dont remember). Which then tells that CPU how to handle data.

Oh yeah .... and the N1 runs 2.2 which is by far superior to 2.1 when it comes to processing data. So the N1 looks a lot better right now. Just wait until 2.2 comes out for the droid or or X.

 

[excalion]

Your question was answered already...it comes down to processor architecture. Qualcomm's Snapdragon platform and more specifically the Scorpion application processor, while being related to TI's Omap Arm v7 series, has enhancements made by Qualcomm. The part of the cpu that handles the SIMD instructions has a wider pipeline, 128 bits vs 64 bits in TI's Omap. Scorpion also has a deeper pipeline to better handle all that data which I'd assume offsets some of the performance benefits a little bit.

As far as the difference between the two benchmarks, they are written to benchmark two different things. Linpack can run almost entirely within the SIMD/NEON portion of the cpu thereby showing off the enhancements made by Qualcomm. Quadrant stresses the entire core showing off total system performance showing that only in certain situations will Snapdragon outperform any other Arm v7 based core.

I don't know what you mean by my question was answered already. The only response that truly answered my question was yours. Gonna post it in OP, thanks.

Also, new question: Since the N1 has a wider pipeline, does this mean the N1 would take longer to run processes that need to access master memory?