iXBT Labs - Computer Hardware in Detail

Platform

Video

Multimedia

Mobile

Other

How CPU Features Affect CPU Performance, Part 3

Intel Core i7, Turbo Boost, Hyper-Threading.

October 12, 2009



<< Previous page

     Next page >>

Having concluded the analysis of AMD Phenom II, we proceed to the latest platform from Intel: Core i7. We must admit that Core i7 is more interesting for a reviewer: as in case of Phenom II, we are going to "play" with the number of cores. However, while Phenom II could offer only various modes of the memory controller, Core i7 gives us two "toys": Turbo Boost and Hyper-Threading. We'll give you a recap (details on the Core i7 architecture are provided in this article devoted to the announcement of this CPU line).

Turbo Boost

This technology allows to overclock selected cores, if the current power consumption of a processor indicates that it's not working at its full capacity. Thus, Turbo Boost should theoretically have a positive effect on the speed of mostly old single-threaded software: "extra" cores may be idle in this very case. You can read about Turbo Boost in the article Intel Core i7 LGA1366 Processors.

Hyper-Threading

We have known this technology since the times of Pentium 4. However, Core 2 Duo/Quad and Pentium Dual Core "forgot" about it, and now Core i7 brings it back to life. Hyper-Threading (HT) allows to emulate two logical cores (detected by an operating system) based on a single physical core by sending commands from two parallel threads to one physical core. The main idea behind it is that some execution units in a core are almost always idle, because there are no "proper" commands for them. But if we execute two threads on a single core, all execution units may be loaded, which will increase the overall efficiency of the system. But there is one tiny "but": Intel had to make some sacrifices to implement this theoretically perfect idea -- not all units can be shared between two virtual cores dynamically. In particular, load/store/reorder buffers are just shared between two virtual cores in half, when HT is enabled. Thus, even if one of virtual cores is idle, technical characteristics of the other virtual core are worse than those of a single physical core, when HT is disabled. And its performance in some cases may be lower for absolutely objective reasons. Intel optimistically has it that there are few such cases. Well, we have a great opportunity to see whether it's actually true.

Tests

We should remind our readers of similar tests conducted over Intel Core i7 920 using our previous version of test procedure (2008). We'll tell you right away: our decision to run the tests one more time using the updated test procedure was absolutely correct. We got different results.

Testbed configuration

  • CPU: Intel Core i7 950
  • Cooler: ASUS Triton 81
  • Motherboard: ASUS P6T SE (Intel X58)
  • Memory: 3 x 2GB Corsair DDR3-1800 in DDR3-1600 mode
  • Graphics card: Palit GeForce GTX 275
  • PSU: Cooler Master Real Power M1000

All tests were run on the same testbed, only BIOS options were subject to changes: we disabled both Turbo Boost and Hyper-Threading at first, then we enabled Turbo Boost, then we disabled Turbo Boost and enabled Hyper-Threading, and finally, we enabled both technologies. Like in the previous parts, in this article we use diagrams with an average score for each group and tables with results obtained in each application. This craving for details is quite explainable: we examine technologies, which may affect the speed of real programs in different ways, and this series of articles was intended to find out these details.

Besides, we traditionally publish a link to a Microsoft Excel spreadsheet for curious readers, which contains all test results in the most detailed form. Besides, it includes an additional tab "Compare" to facilitate their analysis. Just like the tables in the article, it compares the four situations percentagewise. The tables compare all systems with the same configuration -- with Turbo Boost and Hyper-Threading disabled (leftmost column). Tables are colored in the traditional way: bright blue background marks outstanding positive results (in this case we marked performance gains above 10%), red background signals problems: performance drops where we should have performance gains (or at least no changes).

3D visualization



  TB & HT disabled TB enabled HT enabled TB & HT enabled
3ds max ↑* 15.42 16.87 9% 16.43 7% 18.01 17%
Lightwave ↓ 14.29 12.02 19% 12.87 11% 12.57 14%
Maya ↑ 3.85 4.44 15% 4.22 10% 4.58 19%
SolidWorks ↓ 55.33 48.04 15% 57.8 -4% 50.74 9%
Pro/ENGINEER ↓ 1024 978 5% 1023 0% 1020 0%
UGS NX ↑ 3.06 3.07 0% 3.23 6% 3.05 0%
Group Score ↑ 134 148 10% 140 4% 147 10%

* -- The up arrow (↑) marks tests, where the highest results are the best, the down arrow (↓) marks tests, where the best results are the lowest.

Visualization does not respond much even to physical cores, so results of enabled HT are encouraging: at least something. However, we are disappointed by SolidWorks results: having gained 15% from TB, it responded negatively to enabled HT. As a result, this program gets a lower performance gain that it's due in the nominal mode (both TB and HT are enabled): it would have been 15% without HT, and now it's only 9%. 3ds max demonstrates a perfect partnership of technologies: +9% from TB and +7% from HT total +17% to the speed (rounding up added another +1). On the whole, 3D visualization programs feature weak optimizations for multithreading, so they gain more from Turbo Boost. It agrees with our initial assumptions.

3D rendering



  TB & HT disabled TB enabled HT enabled TB & HT enabled
3ds max ↑ 15.6 16.18 4% 17.6 13% 18.69 20%
Lightwave ↓ 94.48 90.87 4% 74.41 27% 71.5 32%
Maya ↑ 02:44 02:35 6% 02:13 23% 02:09 27%
Group Score ↑ 143 150 5% 173 21% 181 27%

Such a perfectly parallel process as rendering expectedly welcomes Hyper-Threading. On the other hand, that's exactly what explains mediocre results from Turbo Boost: when all four cores are loaded, heat release it close to maximum, and it makes no sense to increase the frequency. Lightwave is a champion in squeezing everything from a processor. We don't know whether this article will be read by programmers working on this package, but we'd like to thank them: this program is optimized perfectly.


Write a comment below. No registration needed!


Next page >>

Article navigation:

Page 1: Introduction, tests, part 1

Page 2: Tests, part 2

Page 3: Tests, part 3, conclusions



blog comments powered by Disqus

  Most Popular Reviews More    RSS  

AMD Phenom II X4 955, Phenom II X4 960T, Phenom II X6 1075T, and Intel Pentium G2120, Core i3-3220, Core i5-3330 Processors

Comparing old, cheap solutions from AMD with new, budget offerings from Intel.
February 1, 2013 · Processor Roundups

Inno3D GeForce GTX 670 iChill, Inno3D GeForce GTX 660 Ti Graphics Cards

A couple of mid-range adapters with original cooling systems.
January 30, 2013 · Video cards: NVIDIA GPUs

Creative Sound Blaster X-Fi Surround 5.1

An external X-Fi solution in tests.
September 9, 2008 · Sound Cards

AMD FX-8350 Processor

The first worthwhile Piledriver CPU.
September 11, 2012 · Processors: AMD

Consumed Power, Energy Consumption: Ivy Bridge vs. Sandy Bridge

Trying out the new method.
September 18, 2012 · Processors: Intel
  Latest Reviews More    RSS  

i3DSpeed, September 2013

Retested all graphics cards with the new drivers.
Oct 18, 2013 · 3Digests

i3DSpeed, August 2013

Added new benchmarks: BioShock Infinite and Metro: Last Light.
Sep 06, 2013 · 3Digests

i3DSpeed, July 2013

Added the test results of NVIDIA GeForce GTX 760 and AMD Radeon HD 7730.
Aug 05, 2013 · 3Digests

Gainward GeForce GTX 650 Ti BOOST 2GB Golden Sample Graphics Card

An excellent hybrid of GeForce GTX 650 Ti and GeForce GTX 660.
Jun 24, 2013 · Video cards: NVIDIA GPUs

i3DSpeed, May 2013

Added the test results of NVIDIA GeForce GTX 770/780.
Jun 03, 2013 · 3Digests
  Latest News More    RSS  

Platform  ·  Video  ·  Multimedia  ·  Mobile  ·  Other  ||  About us & Privacy policy  ·  Twitter  ·  Facebook


[an error occurred while processing this directive]

Copyright © Byrds Research & Publishing, Ltd., 1997–2011. All rights reserved.