What makes Intel CPU faster ?

[Mir96TA]

OMG u get a bonk aigo for free card for me missing the C2D part.



you know i honestly think a X6 could keep up with a C2D.

I have no clue..

Err... the PhII is a lot faster in that "benchmark" ...

Higher numbers are better :-D

See I though Intel would be slower, but results are so different
Intel 21.4392 (Winner)
AMD 19.8934 (Loser)
Even with basic root elementary adavantage does not help to have a
better results then Intel
I am not a fan boy, I just wondering what one CPU manufacture have on other, when they can be newer and basic advantage but still can't get a winning number :\

 

[Martimus]

The scores are very close, so it makes it more difficult to say the exact reasons. I will go through a few things that C2D has advantages over Thuban in FP calculations (that I know of):

Thuban has 3 dedicated FP pipelines with 14 entry queue each. One is a 128 bit FADD, one is a 128 bit FMUL, and one is a 128 bit FMISC.

The Core 2 Duo has a more flexible 32 entry, with 2 FPU units, one for 128 bit FADD, and one 128 bit FMUL FDIV. Although all three ports can perform FP moves.

To be honest, I don't see huge differences at the top level, so what Tuxdave said is likely right. Intel just had a more refined FPU than AMD, even though it is technically an older design. The other part is that the program uses an intel compiler, which is optimized for Intel processors, which likely gives the C2D processors an edge.

I went through the known architecture differences, and didn't see anything that jumped out at me. If I knew them better, I would probably have a better answer for you, but alas, I only have passing knowledge of the differences between the architectures.

 

[bryanW1995]

I heard that they use tiger blood in their processors.

 

[KIAman]

At the high level, Intel processors do more work per clock than AMD. What are you missing here?

Intel does 4 units of work per clock.
Amd does 3 units of work per clock.

If I made a proc that does 6 comparative units of work per clock, my proc would be faster clock for clock than Intel's.

 

[biostud]

They add cold solution to their chips

 

[TuxDave]

I heard that they use tiger blood in their processors.

Whoever leaked that information to you is going to get soooo fired. Now EVERYONE is going to use tiger blood.

 

[Idontcare]

Sorry I was little late to do a second Post
Here is a Intel CPU E3110 (E8400) running at 3.1 Gig
Here is a Pic

You did not set the memory size to be the same for both benches. It matters (data and discussion is in the Linx thread here somewhere).

Another thing you would prolly find to be pretty cool is the instruction latency tool that is part of Everest.

Run it on both of your chips and you'll see in the output that just because a processor supports a particular instruction does not mean it can execute it to completion within the same number of clockcycles as another processor.

 

[IntelUser2000]

1. Intel says on the page that it'll score better on their own processors
2. Probably there is a small design advantage
3. Yep, memory size
4. Also, variability.

 

[Mir96TA]

...
..
.
In correct Post

 

[lamedude]

Not sure if it affects this benchmark but I've only heard less than positive things about AMDs SSE performance compared to Intels.

 

[jsedlak]

the AMD got 60GFlops the C2D got 20GFlops... what's the problem?

Intel gets more gflops per core?

 

[blckgrffn]

They add cold solution to their chips

This.

 

[Mir96TA]

1. Intel says on the page that it'll score better on their own processors3. Yep, memory size
.

You did not set the memory size to be the same for both benches. It matters (data and discussion is in the Linx thread here somewhere).

This is with excat memory size, Core speed, and Core numbers

 

[Mir96TA]

Intel gets more gflops per core?

With slower memory speed, and Kind (Intel running on DDR2) and old CPU and MB Chipset
An older Intel CPU with Older Chipset and a pervious slower running Memory still comes out 11% faster ?
That is like heck of a perfoamnce:hmm:

 

[jiffylube1024]

Intel's architecture has higher IPC (instructions per clock) than AMD's.

Here's Anand's Phenom II review where he starts talking about the performance-per-MHz discrepancy:
http://www.anandtech.com/show/2702/4

AMD's L3 caches run a bit slower than Intel's, for one. Their "uncore" (part of the CPU die that's not a CPU core) - including the L3 cache runs at 1.8 GHz or 2 GHz. I believe Intel's L3 memory is full speed. More info from the Phenom II review:

http://www.anandtech.com/show/2702/8

AMD's Phenom II is a redesigned version of their Phenom I, which is an older architecture than the Core i series. AMD needs to maintain AM2+ backwards compatibility on the CPU's, and needs to have both a DDR2 and DDR3 memory controller.

 

[Mir96TA]

AMD's L3 caches run a bit slower than Intel's, for one. Their "uncore" (part of the CPU die that's not a CPU core) - including the L3 cache runs at 1.8 GHz or 2 GHz. I believe Intel's L3 memory is full speed. .

No L3 Cache on Intel

 

[Idontcare]

I think plenty of folks have explained the discrepancy already...it comes down to instruction latency, cache latency, etc. I.e. specific nuances in the implemented microarchitectures to support the ISA as needed to run the given program.

 

[Tuna-Fish]

Plenty of people have posted what are essentially the correct answers -- however, I'll try to put in a form that's easier to understand.

The "back end", the part that actually does the computation, is essentially the same on both processors. However, the speed of a processor isn't that much about the execution units these days -- a 64-bit FPMUL is like 80k transistors, and there are 800M (!) transistors in a Penryn core 2 duo processor. Most of those transistors are a part of the "front end", and their job is to get an uninterrupted stream of instructions and data to the execution units. The short and sweet version is that Intel is better at that, and their execution units can thus spend some 10% more of their time working as opposed to being stalled either on instructions or data and having nothing to do. This is mostly because of better branch prediction and better data prefetchers.

If you want more through explanations, go look in the various in-depth reviews by anandtech, realworldtech, or others.

 

[Mir96TA]

I think this is what I am understanding,
Having newer CPU/Chipset, and having faster memory with more bandwidth plus with more Instruction set and cache does equate Quicker CPU.
How CPU compute its logical instruction set to crunch data, makes a Quick CPU.
Having a Faster CPU does not equate a Quicker CPU

 

[veri745]

As IntelUser explained LinX is a synthetic FP benchmark, and is almost entirely dependent on the FP performance of the CPU. As your results show, on a per-core basis. The Core2 and PhII architectures are very close in FP performance, and C2 has a slight (3-4%) advantage.

You'd get the same results on a AM2+ PhII x4 920 that came out 3 years ago, and on an Athlon II.

What the PhII x6 brought to the table is more cores and faster clockspeeds. Once you take those advantages away, you're obvious going to have a more even fight.

 

[Cogman]

There are several things.

1. They are different architectures. As a result, AMD does some things faster than intel does. Workload has always been an important factor when saying CPU x is faster than CPU y. (yes, I do realize that Intel is generally faster.)

2. Intel has a much bigger budget for R&D. While that doesn't always equal better quality (see Xscale) it really helps. Intel has lots of money it can spend on whatever.

3. As a result of #2, I don't think AMD/Global Foundries has ever beaten Intel to a process node. In fact, they usually lag by almost a year. Smaller fab = faster CPUs.

4. Did I mention that they are different architectures? Intel just does things differently from AMD. We can't really go into too much detail because we don't have a whole lot. I haven't seen what their block level schematics look like. Yes, there is a wealth of information out there on what they are doing or what they did do, but that still doesn't give a complete picture.