Why Intel Core i7 2700K and 3960X can beating AMD FX-8150?

[frenzyvanrafi]

I was wondering why Intel Core i7 2700K and Core i7 3960X can beating AMD FX-8150, though it has 8-core but why it can be defeated by Hexa-core and Quad-core. I was thinking if the processor has more core, they had more performance, but if Hexa-core/Quad-core can defeating Peta-core from AMD, what factors led to Core i7 3960X 6-core and Core i7 2700K 4-core can defeating AMD FX-8150? Is the instruction set is different, cutting edge technology, many of thread, or a nice architecture?

Here's the proof:

Spoiler

But if they don't have money for extreme overclocking, they choose AMD FX-8150. But for high wealth serious gamer (but wealthy not younger, but young wealthy also) choose Core i7 3960X.

 

[amuffin]

PASSMARK IS HORRIBLE!
Anyways, even the core i5-2500k outperforms the fx-8150. This is because the bulldozer chip isn't a "real" 8 core processor. It is actually many half cores mashed together to simulate an 8 core processor, like hyperthreading. As for performance, sandybridge is a much better architecture than Zambezi. Read these below.

Look at these:
http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043.html
http://www.anandtech.com/show/5091/intel-core-i7-3960x-sandy-bridge-e-review-keeping-the-high-end-alive

 

[quilciri]

PASSMARK IS HORRIBLE!
Anyways, even the core i5-2500k outperforms the fx-8150. This is because the bulldozer chip isn't a "real" 8 core processor. It is actually many half cores mashed together to simulate an 8 core processor, like hyperthreading. As for performance, sandybridge is a much better architecture than Zambezi. Read these below.

Look at these:
http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043.html
http://www.anandtech.com/show/5091/intel-core-i7-3960x-sandy-bridge-e-review-keeping-the-high-end-alive

That's good enough for government work, but not quite accurate.

The 8150 does have 8 actual, physical cores. These are arranged in 2 core "modules" that each share a single L3 cache and a single floating point processor between the 2 cores. Each core has it's own L1 and L2 cache, and it's own integer processor.

The largest reason that the I-series outperforms it so badly is because each individual I-series core has a longer pipeline and executes much more per clock-cycle than each core on an fx chip.

p.s. windows 7 also handles core assignment on the fx chips pretty poorly. Ideally, you would assign work so that each floating point processor and L3 cache is used before loading the 2nd core on each module, but right now, win7 uses each core in a single module before loading the next one, so unless the app is using 3 or more cores, 1 fp processor and L3 cache is being wasted.

 

[amuffin]

Better IPC too Sandybridge clock for clock is 15% faster, in gaming the i3-2120 and 2100 outperform the fx-8120. Even the sandybridge pentiums do too: http://www.tomshardware.com/reviews/gaming-fx-pentium-apu-benchmark,3120.html

 

[quilciri]

While amd has nothing in the lineup to compete with a 2500k or better, if you are willing to overclock the AMD lineup does slightly outperform similarly priced intel chips, but *only* if you take them to the highest stable OC.

Why intel multiplier locked the entire i3 series is beyond me, they'd have shut AMD out of the market if there were i3-k and pentium G-k chips.

 

[Pinhedd]

Intel i7 processors are capable of executing more than twice as many Dhrystone instructions per second per core as AMD Bulldozer processors.

While Bulldozer may have 8 cores, the fixed point capabilities of each core are significantly impaired compared to i7 or even previous generation AMD processors. The floating point unit is also shared between two processors, but this has no impact on the Dhrystone IPS because Dhrystone contains no floating point instructions.

Even on a purely fixed point basis, each Bulldozer core is vastly inferior to each SandyBridge core and even most Nehalem cores.

 

[loneninja]

That's good enough for government work, but not quite accurate.

The 8150 does have 8 actual, physical cores. These are arranged in 2 core "modules" that each share a single L3 cache and a single floating point processor between the 2 cores. Each core has it's own L1 and L2 cache, and it's own integer processor.

The largest reason that the I-series outperforms it so badly is because each individual I-series core has a longer pipeline and executes much more per clock-cycle than each core on an fx chip.

p.s. windows 7 also handles core assignment on the fx chips pretty poorly. Ideally, you would assign work so that each floating point processor and L3 cache is used before loading the 2nd core on each module, but right now, win7 uses each core in a single module before loading the next one, so unless the app is using 3 or more cores, 1 fp processor and L3 cache is being wasted.

You have some information way off. The L3 cache is shared between all modules/cores, while each module shares L2 cache between 2 cores, L1 cache is the only thing dedicated to an individual core. Also FX has a longer Pipeline than Sandy Bridge/Ivy Bridge/Phenom II, and a number of other processor designs, this is part of the reason IPC is lower. A longer pipeline lowers IPC but increases clock speed potential. Intels Pentium 4 had a much longer pipeline than Pentium 3 or Core did.

 

[amuffin]

Eh, if OP is looking at passmark he wouldn't understand^^

 

[InvalidError]

A longer pipeline lowers IPC but increases clock speed potential.

It isn't necessarily the case, you can achieve the same IPC on a 30 stages pipeline as you can on a 10 stages pipeline but you have to manage 3X as many in-flight instructions on the longer pipeline which is much more complex to manage. Keeping a longer pipeline filled is further complicated by pipeline stalls when instructions have inter-dependencies and nothing to feed into the pipeline in the meantime, which is where multi-threaded cores come into play to provide a separate instruction flow to help fill pipelines with and increase overall IPC.

 

[jaguarskx]

PassMark basically sucks because if you compare actual application benchmarks amongst various CPUs you'll notice the ranking of FX CPUs will be very different.

Basically each core in an Intel CPU can process more instructions than each AMD FX core. Therefore, the FX CPU cores would have to be clocked faster to process the same amount of instructions that Intel CPU cores can. It's kinda like a person who can read 40 pages per hour vs. a person who can only read 30 pages per hour.

Regarding the number of cores... yeah the FX has up to 8 cores while Intel Core i7-2600k only has 4 cores. The problem is not very many programs can use more than 4 cores, that's especially true for games. Can programs and games be designed to use more than 4 cores? Yes, but doing so requires more development time which in the end means you need to wait longer for those programs and games to come out, plus they will probably cost more to buy because it costs more to make.

 

[quilciri]

You have some information way off. The L3 cache is shared between all modules/cores, while each module shares L2 cache between 2 cores, L1 cache is the only thing dedicated to an individual core. Also FX has a longer Pipeline than Sandy Bridge/Ivy Bridge/Phenom II, and a number of other processor designs, this is part of the reason IPC is lower. A longer pipeline lowers IPC but increases clock speed potential. Intels Pentium 4 had a much longer pipeline than Pentium 3 or Core did.

No, it's a L3 cache per module. I was wrong about the L2 cache, though.
http://www.guru3d.com/article/amd-fx-8150-processor-review/2

and sorry, meant to say shorter pipeline there.

 

[frenzyvanrafi]

Better IPC too Sandybridge clock for clock is 15% faster, in gaming the i3-2120 and 2100 outperform the fx-8120. Even the sandybridge pentiums do too: http://www.tomshardware.com/reviews/gaming-fx-pentium-apu-benchmark,3120.html

Dual Core can against 8 Core? Wow... It doesn't sense

 

[Keanu Reeves]

I feel like this thread is getting derailed, and for the OP, in layman's terms, each core in the i-series processors is able to execute tasks more efficiently and quickly because of the way they are built. You can have a house with 8 windows and a house with 4 windows, but if the 8 windows are dirty, you can't tell someone what's going on outside. On the other hand, if you have 4 clean windows, it's much easier to tell what is going on. That's the way I explained it to my friend and he understood it a bit better, so hopefully you will too. It's all in the execution, my friend, not the cores.

 

[lee_m_616]

I've been wondering the same thing myself, in regards to 6 vs 8 cores, as I'm looking into building my own machine for gaming, programming, editing, all the standard stuff someone does on a beastly machine, I think you guys should have a look at this,
http://www.knowbytes.com/home/articles/processors/intel-core-i7-3960x-versus-amd-fx-8150

The guy writing the article makes some valid points in regards to price vs performance.
The only real limiting factor on the AMD processor is that it can only take half the RAM of the Intel. Anything else can be beefed up by external components.
In reality, what would the average end user be doing with 64GB of RAM anyway?
I know my post is late, but you should have a look anyway.

Cheers