Richland CPUs - Are AMD CPUs the "best bang for the buck"?

[VirtualLarry]

I own a couple of Thubans and a LLano. They're ok, I guess, for people like me that prefer "cheap" CPUs.

 

[Accord99]

I think this is true but at what cost premium? Who cares if it takes double the cores to perform the same task that an Intel can perform in the same time span. If the AMD quad-core chip is cheaper than the equivalent performing Intel dual-core CPU?

But they're not equivalent, they would only be equivalent with 4 or more CPU intensive threads. In many tasks, double the cores won't allow the AMD quad-core to complete the task at the same time as the Intel dual-core.

Or to put it in car terms, would I buy the AMD Bulldozer minivan that can carry 8 people at a time but regardless of how many people I actually carry, it can only go slow? Or would I buy the Intel Hyperthreaded family sedan that when carrying <=4 people is 60-70% faster than the AMD minivan carrying <=4 people, and the family sedan can still carry 8 people to a destination faster than the AMD minivan and also gets twice the mileage.

Given a 10-15% Intel price premium on a typical rig, I say the Intel is the better for the buck at least for me.

 

[AtenRa]

But they're not equivalent, they would only be equivalent with 4 or more CPU intensive threads. In many tasks, double the cores won't allow the AMD quad-core to complete the task at the same time as the Intel dual-core.

For a single task maybe no, what about 4 tasks ??

Or to put it in car terms, would I buy the AMD Bulldozer minivan that can carry 8 people at a time but regardless of how many people I actually carry, it can only go slow? Or would I buy the Intel Hyperthreaded family sedan that when carrying <=4 people is 60-70% faster than the AMD minivan carrying <=4 people, and the family sedan can still carry 8 people to a destination faster than the AMD minivan and also gets twice the mileage.

Given a 10-15% Intel price premium on a typical rig, I say the Intel is the better for the buck at least for me.

Actually your paradigm can illustrate the difference between AMD FX and Intel Core i CPUs Difference.

Lets assume that Intel can carry 4 people in one hour 100km away.
Lets assume that AMD can carry 8 people in two hours 100km away.

From the above Intel is 100% or 2x faster covering 100km with 4 people. But how long it will take Intel to transport 8 people 100km away ???

It will take Intel one hour to transport 4 people 100km, then it will also take one hour to return back 100km and load another 4 people. And finally it will take another hour to transport the last 4 people 100km. The total time for Intel to transport 8 people 100km away will be 3 hours.
AMD will be 33,3% faster transporting 8 People 100km away than Intel although Intel is 2x faster at transporting 4 people 100km.

That is called Throughput. AMD can carry more people at the same time 100km away or in Computer terms, AMD 8 Core can finish more jobs at a given time than Quad core Intel.

 

[BallaTheFeared]

Lets assume that Intel can carry 4 people in one hour 100km away.
Lets assume that AMD can carry 8 people in two hours 100km away.

Your analogy is bad because you aren't accounting for the time it takes for Intel to return making AMD look better than just on par which is what they'd be if comparing the i5 in this situation.

A better analogy would be...

It takes Intel up to two hours to carry eight people 100km, however they can carry four or less people 100km in an hour.

It takes AMD up to two hours to carry eight people 100km, however they can carry less people but without any speed benefit so carrying four people will still take AMD two hours.

 

[AtenRa]

Your analogy is bad because you aren't accounting for the time it takes for Intel to return making AMD look better than just on par which is what they'd be if comparing the i5 in this situation.

A better analogy would be...

It takes Intel up to two hours to carry eight people 100km, however they can carry four or less people 100km in an hour.

It takes AMD up to two hours to carry eight people 100km, however they can carry less people but without any speed benefit so carrying four people will still take AMD two hours.

Thank God we have invented Start Trek's transporters and we instantaneous transport the car back 100km to reload another 4 people. Beam me Up Scotty

ps: Can you show me any review or even data that support your claim that Core i5 will do the same amount of jobs at the same time as 8 core FX??

 

[BallaTheFeared]

ps: Can you show me any review or even data that support your claim that Core i5 will do the same amount of jobs at the same time as 8 core FX??

What do you mean? You can run eight instances down the i5, it just won't be any faster than doing four and then four again.

 

[Accord99]

For a single task maybe no, what about 4 tasks ??

I already answered that before.

Actually your paradigm can illustrate the difference between AMD FX and Intel Core i CPUs Difference.

Lets assume that Intel can carry 4 people in one hour 100km away.
Lets assume that AMD can carry 8 people in two hours 100km away.

From the above Intel is 100% or 2x faster covering 100km with 4 people. But how long it will take Intel to transport 8 people 100km away ???

It will take Intel one hour to transport 4 people 100km, then it will also take one hour to return back 100km and load another 4 people. And finally it will take another hour to transport the last 4 people 100km. The total time for Intel to transport 8 people 100km away will be 3 hours.

Except the Intel family sedan can miraculously and instantly transform into 8 seater, at the expense of top-speed (this extends to all Intel and AMD "cars" in general). But with Hyperthreading, it's reduced top-speed is higher than normal. So it too can transport 8 people in two hours 100km away. Or to be more accurate with benchmarks:

the Hyperthreaded Intel can carry 4 people 100km away in 75 minutes, and it can also carry 8 people 100km away in 120 minutes.

AMD will be 33,3% faster transporting 8 People 100km away than Intel although Intel is 2x faster at transporting 4 people 100km.

That is called Throughput. AMD can carry more people at the same time 100km away or in Computer terms, AMD 8 Core can finish more jobs at a given time than Quad core Intel.

But Intel Hyperthreaded cores have the same throughput as an AMD module (if we are allow for a modest clock speed advantage to AMD). That's AMD problem with Bulldozer, it doesn't have any throughput advantage at two threads versus a Intel Hyperthreaded core, which means it's also far behind with only one thread.

 

[AtenRa]

What do you mean? You can run eight instances down the i5, it just won't be any faster than doing four and then four again.

In order for the Core i5 (4 threads) to be as fast as an 8 Core FX (8 threads) in Throughput, it will have to be 2x faster (for simplicity i dont count CPU Core scaling, i assume 100% scaling).

I can tell you that Core i5 Integer performance is not 2x faster than FX. Even single thread performance in Cinebench is not 2x faster than FX.

There is a difference in performance computing a single application with 4 or 8 threads against 8x applications simultaneously with 4 or 8 threads.

 

[BallaTheFeared]

In order for the Core i5 (4 threads) to be as fast as an 8 Core FX (8 threads) in Throughput, it will have to be 2x faster (for simplicity i dont count CPU Core scaling, i assume 100% scaling).

I can tell you that Core i5 Integer performance is not 2x faster than FX. Even single thread performance in Cinebench is not 2x faster than FX.

There is a difference in performance computing a single application with 4 or 8 threads against 8x applications simultaneously with 4 or 8 threads.

It actually doesn't because of the modular design not only is the per core bad, but the per core scaling is also leaving a lot to be desired. There is a performance penalty on FX for running two different workloads down the same module.

The MT scaling on FX is abysmal, which is why with eight cores it's competing with the i5 in MT in the first place.

If you were to run eight different workloads the performance loss on the FX will be greater than it is running a MT program.

 

[AtenRa]

I already answered that before.


Except the Intel family sedan can miraculously and instantly transform into 8 seater, at the expense of top-speed (this extends to all Intel and AMD "cars" in general). But with Hyperthreading, it's reduced top-speed is higher than normal. So it too can transport 8 people in two hours 100km away. Or to be more accurate with benchmarks:

the Hyperthreaded Intel can carry 4 people 100km away in 75 minutes, and it can also carry 8 people 100km away in 120 minutes.


But Intel Hyperthreaded cores have the same throughput as an AMD module (if we are allow for a modest clock speed advantage to AMD). That's AMD problem with Bulldozer, it doesn't have any throughput advantage at two threads versus a Intel Hyperthreaded core, which means it's also far behind with only one thread.

We are talking about 4 cores (threads) vs 8 cores (threads).
But yes, intel at 8 threads have almost the same throughput as AMD 8 threads at almost the same clock speed.

 

[AtenRa]

It actually doesn't because of the modular design not only is the per core bad, but the per core scaling is also leaving a lot to be desired. There is a performance penalty on FX for running two different workloads down the same module.

The MT scaling on FX is abysmal, which is why with eight cores it's competing with the i5 in MT in the first place.

If you were to run eight different workloads the performance loss on the FX will be greater than it is running a MT program.

You are completely wrong about that.

4x Socket AMD Opteron 6386 SE @ 2.8GHz (Base), 64 Cores 140W TDP = Score 1060

4x Socket Intel XEON E5-4650 @ 2.7GHz (Base), 64 Threads 130W TDP = Score 1180

Throughput performance is very very close between the two.

 

[ShintaiDK]

You are completely wrong about that.

4x Socket AMD Opteron 6386 SE @ 2.8GHz (Base), 64 Cores 140W TDP = Score 1060

4x Socket Intel XEON E5-4650 @ 2.7GHz (Base), 64 Threads 130W TDP = Score 1180

Throughput performance is very very close between the two.

32 Intel cores faster than 64 AMD cores. Are you sure its what you wanted to show?

 

[AtenRa]

32 Intel cores faster than 64 AMD cores. Are you sure its what you wanted to show?

32 AMD Modules 64 Threads
32 Intel Cores 64 Threads

Yeap, that's exactly what I wanted to show.

 

[BallaTheFeared]

32 AMD Modules 64 Threads
32 Intel Cores 64 Threads

Yeap, that's exactly what I wanted to show.

If only they were actually that similar, we wouldn't even be having this conversation.

Both of you can drop the smarmy act
-ViRGE

 

[Lonbjerg]

32 AMD Modules 64 Threads
32 Intel Cores 64 Threads

Yeap, that's exactly what I wanted to show.

You might wanna tell AMD about that "name" change, just to keep things nice and orderly:

http://www.amd.com/us/products/desktop/processors/amdfx/Pages/amdfx.aspx

Take your PC&#8217;s megatasking abilities to extreme levels with the first native 8-core desktop processor...

http://www.amd.com/us/products/serv...ries-platform/pages/6000-series-platform.aspx

Unleash unprecedented performance with our brand new architecture with up to 16 cores

Hell even ther spec page says "core" not "module"

http://products.amd.com/en-us/OpteronCPUDetail.aspx?id=814&f1=AMD+Opteron%E2%84%A2+6300+Series+Processor&f2=&f3=Yes&f4=&f5=1000&f6=G34&f7=C0&f8=32nm&f9=&f10=6400&f11=&

Core count: 16

Nice try though

 

[inf64]

SIMD capabilities wise, AtenRa is correct. 32FP(2 way SMT) units Vs 32FP(2 way SMT) units, 64 threads vs 64 threads. Or some of you expected that each FP unit in AMD core should somehow magically be much faster ? If so then the answer is no. It's painfully obvious that at similar clockspeeds and thread count, the 2 compared products offer similar throughput. There is a question of power efficiency though, but that is another matter. The point here was performance.

 

[AtenRa]

If only they were actually that similar, we wouldn't even be having this conversation.

The conversation was about Intel's Quad Core (Core i5) vs AMD FX 8 core in Throughput Performance. There is no way the Intel Quad Core i5 will have same Throughput Performance than AMDs FX 8 core CPU.
I told you that for that to happen, the Intel Quad core should have 2x the single thread performance to reach the same Throughput performance of AMD 8 Core CPU (assuming perfect 100% Core scaling).

Oh and by the way, AMD could call them BANANAs instead of Cores and still what counts at the end its the performance not the name.

 

[BallaTheFeared]

The conversation was about Intel's Quad Core (Core i5) vs AMD FX 8 core in Throughput Performance.

Indeed it was, so why you telling me to produce 8 job load reviews then telling me I'm wrong while linking a spec benchmark?

My i5 is as fast as the 8350 in just about everything threaded. It isn't quite twice as fast in single threaded generally, but I covered that... The modular tax.


You want to stick with Intger and legacy code that's fine, we won't dive into FMA3 and AVX2 where the i5 easily trounces the 8350 clock for clock in real world workloads like Handbrake. However you can't claim it's the same as HT because it isn't, that's because there is a integer core for each "core" eight integer cores in a 8350, four flex 256 bit FPU cores which can split to 8 126 bit FPU cores.

There is no similarity to AMDs deign and Intels, which is why the i7 maintains it's ST performance against the i5 and while the 8350 pales in comparison.

 

[AtenRa]

Indeed it was, so why you telling me to produce 8 job load reviews then telling me I'm working while linking a spec benchmark?

The SPECint_rate_base2006 in the Link I provided measures Integer Throughput.

http://www.spec.org/cpu2006/Docs/readme1st.html#Q15

Q15. What is the difference between a "rate" and a "speed" metric?

There are several different ways to measure computer performance. One way is to measure how fast the computer completes a single task; this is a speed measure. Another way is to measure how many tasks a computer can accomplish in a certain amount of time; this is called a throughput, capacity or rate measure.

• The SPECspeed metrics (e.g., SPECint2006) are used for comparing the ability of a computer to complete single tasks.
• The SPECrate metrics (e.g., SPECint_rate2006) measure the throughput or rate of a machine carrying out a number of tasks.

For the SPECrate metrics, multiple copies of the benchmarks are run simultaneously. Typically, the number of copies is the same as the number of CPUs on the machine, but this is not a requirement. For example, it would be perfectly acceptable to run 63 copies of the benchmarks on a 64-CPU machine (thereby leaving one CPU free to handle system overhead).

 

[ShintaiDK]

The conversation was about Intel's Quad Core (Core i5) vs AMD FX 8 core in Throughput Performance. There is no way the Intel Quad Core i5 will have same Throughput Performance than AMDs FX 8 core CPU.
I told you that for that to happen, the Intel Quad core should have 2x the single thread performance to reach the same Throughput performance of AMD 8 Core CPU (assuming perfect 100% Core scaling).

Oh and by the way, AMD could call them BANANAs instead of Cores and still what counts at the end its the performance not the name.

I have bad news for you...again.
While there is no 8350, there is a 8150.
http://www.spec.org/cpu2006/results/res2011q4/cpu2006-20111121-18937.txt

And there is a 2500, 3570 and a 4670:
http://www.spec.org/cpu2006/results/res2013q3/cpu2006-20130701-25995.txt
http://www.spec.org/cpu2006/results/res2012q3/cpu2006-20120702-23377.txt
http://www.spec.org/cpu2006/results/res2011q3/cpu2006-20110718-17551.txt

Specint_rate, base and peak:
FX8150 106/115
2500 127/134
3570 146/156
4670 173/179

Not much "throughput" performance there for the FX. No surprise since server benchmarks shows the same.

 

[BallaTheFeared]

The SPECint_rate_base2006 in the Link I provided measures Integer Throughput.

And it does this by running eight different benchmarks at the same time to produce the "same amount of jobs at the same time as 8 core FX??" requirement you laid out previously?

I didn't think so.

 

[AtenRa]

I have bad news for you...again.
While there is no 8350, there is a 8150.
http://www.spec.org/cpu2006/results/res2011q4/cpu2006-20111121-18937.txt

And there is a 2500, 3570 and a 4670:
http://www.spec.org/cpu2006/results/res2013q3/cpu2006-20130701-25995.txt
http://www.spec.org/cpu2006/results/res2012q3/cpu2006-20120702-23377.txt
http://www.spec.org/cpu2006/results/res2011q3/cpu2006-20110718-17551.txt

Specint_rate, base and peak:
FX8150 106/115
2500 127/134
3570 146/156
4670 173/179

Not much "throughput" performance there for the FX. No surprise since server benchmarks shows the same.

Those are on Windows 7 SP1 without the Bulldozer Patches. Can you find any SandyBridge 4 Core XEONs to compare against 8 core Bulldozer Opterons (62xx) in Linux ??

 

[ShintaiDK]

Those are on Windows 7 SP1 without the Bulldozer Patches. Can you find any SandyBridge 4 Core XEONs to compare against 8 core Bulldozer Opterons (62xx) in Linux ??

I am sure you can look yourself if you wish to change the parameters.

Also I assume the topic of the thread is geared towards Windows for that matter.

If you are lucky, you might be able to get a 140W FX8350 to perform like a 95W i5 2500 with 8 threads in throughput.

 

[Sweepr]

So you need to pick 2-year old SB CPUs and run Specint specifically on Linux in order to make Piledriver look better? Got it.

 

[sm625]

Richland is not a good value because you are paying an integration penalty. For the same price as an A10-6800K, you can buy a celeron and a discrete graphics card and get better gaming performance and better performance in general usage, web surfing, javascript, etc. There are only a few cases where an A10-6800K would outperform a celeron + discrete, but those corner cases arent worth what you give up.

FX-6300 however can be a very good value. Its 6 cores might be worth the loss in single threaded performance compared to an i3 of the same price range. That depends on what you are doing.