Sign in with
Sign up | Sign in
Your question

Cores/Hyperthreaded AMD/Intel

Last response: in CPUs
Share
January 7, 2013 6:43:32 PM

I've done some research but have not found a definitive answer on this as of yet.

Intel has six cores that are hyper-threaded (3970X), therefore 12 threads. Now on those 12 threads, can 2 threads be ran simultaneously?

Therefore obtaining a score like 12 threads * 2 threads each = 24 * 3.5ghz = 84

vs

AMD 1 thread per core, so a dual opteron 6272 would be 32 * 2.1ghz = 67.2



a c 169 à CPUs
a b å Intel
a b À AMD
January 7, 2013 7:07:32 PM

reiddar said:
I've done some research but have not found a definitive answer on this as of yet.

Intel has six cores that are hyper-threaded (3970X), therefore 12 threads. Now on those 12 threads, can 2 threads be ran simultaneously?

Therefore obtaining a score like 12 threads * 2 threads each = 24 * 3.5ghz = 84

vs

AMD 1 thread per core, so a dual opteron 6272 would be 32 * 2.1ghz = 67.2


Hyperthreading improves performance by splitting a single core's resources between two functionally independent thread contexts. This allows for two threads to be executed simultaneously with the resources being allocated dynamically between them.

Hyperthreading exploits inherent inefficiencies in program code as well as the superscalar architecture of Intel microprocessors to allow for an execution core's resources to be used more effectively. It does not change the capabilities of that core, simply how it can be used by the operating system. This is built on the idea that if a microprocessor executes one thread context at a time, physical resources may go unused.

For example, a processor may be able to generate X addresses per clock cycle, perform Y arithmetic operations per clock cycle, and perform Z floating point operations per clock cycle. If a particular thread does not use all of X, Y, and Z during each clock cycle, then a separate thread may be able to make up the difference. Hyperthreading thus schedules a second thread to more effectively use all available resources.

A 3.5Ghz Ivybridge i5 without hyperthreading has the exact same execution capabilities per clock cycle as a 3.5Ghz Ivybridge i7 with hyperthreading. Thus, a highly optimized piece of code will show no gains by using a Hyperthreaded processor over a non-Hyperthreaded processor. However, most code is not optimized to that degree.

Overall gains are typically in the 5%-15% depending on the application and family.

So, enabling Hyperthreading is usually a good idea, but it will not double performance. Poorly written applications such as Battlefield 3 can actually have negative gains when Hyperthreading is enabled as they will detect 12 threads and assume that there are 12 cores supporting them, rather than 6. They will then spawn resources to saturate 12 cores (such as physics threads) when only 6 cores worth of resources are actually available.

EDIT: I completely forgot to answer your question.

Yes you can obtain a weighted average of processor capabilities to compare various configurations but Hyperthreading should not be included. The effects of hyperthreading should instead be included in the average IPC (instructions per clock). Current metrics show that Sandybridge/Ivybridge obtain approximately 2.5 times the instructions per clock cycle as Bulldozer/Piledriver. So, we can approximate performance as follows:

Number of processor packages * number of cores per processor * average number of instructions per clock cycle * number of cycles per second

For the 3970x that would be:

1 * 6 * 9 (average for Sandybridge) * 3.5 billion = 189 billion operations per second

For the dual opteron 6272 that would be

2 * 16 * 3.7 (average for Bulldozer) * 2.1 billion = 248.64 billion operations per second.

This is of course completely neglecting turbo modes and memory efficiency.
m
0
l
a c 471 à CPUs
a c 115 å Intel
a c 118 À AMD
January 7, 2013 8:45:15 PM

The real question is what is the purpose of the PC?

If you are building a gaming rig, then just be aware that the vast majority of games do not use more than 2 cores. That are some games that can use more than 2, but the number is tiny in comparison to those using 2 cores. Battlefield 3 is one of those rare games that I believe can use more than 4 cores (multiplayer only). ARMA might be another one. The question is how effectively can those games use more than 2 core?

Back in 2010 the was one review about CPU core scaling in games that can make use of 4 cores; I believe it was from Techspot, but I cannot find that review anymore. Anywaste, on average going from 2 core to 3 cores resulted in an average of 26% increase, but going from 3 cores to 4 cores only yielded another 6% on average.

Having stated the above and while the vast majority of games only use 2 cores, I generally recommend a quad core CPU because it is more "future proof" than a dual core CPU. However, if you are on a tight budget then I would recommend a dual core.

Games do not use Hyper Threading (HT), and over the years benchmarks have shown that on average HT causes a 2% drop in performance.

If you are using heavily threaded applications... then multiple cores / HT can come in handy... as long as they are designed to use more than 2 cores and/or HT. 3D rendering and video encoding are prime examples.
m
0
l
Related resources
a c 169 à CPUs
a b å Intel
a b À AMD
January 7, 2013 8:49:18 PM

jaguarskx said:
The real question is what is the purpose of the PC?

Games do not use Hyper Threading (HT), and over the years benchmarks have shown that on average HT causes a 2% drop in performance.

If you are using heavily threaded applications... then multiple cores / HT can come in handy... as long as they are designed to use more than 2 cores and/or HT. 3D rendering and video encoding are prime examples.


Applications have minimal control over how they are scheduled. If a game can take advantage of 4 simultaneous threads then it will benefit from having Hyperthreading enabled on an i3 far mores so than on i7. The difference between logical processors and physical cores is transparent to applications unless they specifically go looking for them.
m
0
l
a b à CPUs
January 7, 2013 9:05:22 PM

Also, if HT is enabled when playing games, there will be a detriment to the game's performance, supposing that the game has not been restricted to two threads.

It was between 2-8% performance decrease for each HT thread that the game was on, IIRC.
m
0
l
January 7, 2013 9:31:42 PM

The application is a custom software that is database dependent 70% on SSD drives and 30% CPU (running in parallel across all cores)

Re benchmarking:

Quote:
or the 3970x that would be:


1 * 6 * 9 (average for Sandybridge) * 3.5 billion = 189 billion operations per second


For the dual opteron 6272 that would be


2 * 16 * 3.7 (average for Bulldozer) * 2.1 billion = 248.64 billion operations per second.


This is of course completely neglecting turbo modes and memory efficiency.


Where are the 3.7 and 9 multipliers coming from?


In the ASUS P9X79 WS board, what could I comfortably OC the chip to?

Assuming the above math and with an OC of 4.6, I could be at 248.40 billion operations per second?

The case is a Cooler Master Haf 932 Advanced Full Tower EATX Case with Cooler Master Hyper 212 Evo Direct Touch 4 Heatpipe Heatsink AM2 AM3 LGA1366/1155/1156/2011 120mm
m
0
l
a c 169 à CPUs
a b å Intel
a b À AMD
January 8, 2013 2:51:47 AM

reiddar said:
The application is a custom software that is database dependent 70% on SSD drives and 30% CPU (running in parallel across all cores)

Re benchmarking:

Quote:
or the 3970x that would be:


1 * 6 * 9 (average for Sandybridge) * 3.5 billion = 189 billion operations per second


For the dual opteron 6272 that would be


2 * 16 * 3.7 (average for Bulldozer) * 2.1 billion = 248.64 billion operations per second.


This is of course completely neglecting turbo modes and memory efficiency.


Where are the 3.7 and 9 multipliers coming from?


In the ASUS P9X79 WS board, what could I comfortably OC the chip to?

Assuming the above math and with an OC of 4.6, I could be at 248.40 billion operations per second?

The case is a Cooler Master Haf 932 Advanced Full Tower EATX Case with Cooler Master Hyper 212 Evo Direct Touch 4 Heatpipe Heatsink AM2 AM3 LGA1366/1155/1156/2011 120mm


Ah sorry, I should have sourced those.

The 9 and 3.7 come from the Dhrystone benchmark.

The Asus X79 motherboards are the best X79 boards out there. I use a Rampage IV Extreme. The Sandybridge-E processors are extreme power hogs. When compared to Sandybridge i7 they have half again as many IA-32 cores, twice as many memory lanes, almost twice as much cache, and two and a half times as many PCIe lanes. At 4Ghz on all cores my 3960x pulls over 230 watts. This is far more than most desktop coolers can handle so if you are looking to overclock a Sandybridge-E processor you're practically required to buy a full watercooling loop. With that though you can comfortably overclock it to around 4.5Ghz but I would advise against going higher than that.
m
0
l
January 8, 2013 12:40:48 PM

I've looked at the h100i but I'm concerned about it being too loud.

What about the nh-d14 from Noctua, what I could max out at with that cooler? (guesstimate obviously)
m
0
l
!