Our today's article will be devoted to aspects not covered in the reviews of the Intel Core i7 architecture and its first representative Intel Core i7 920. Our readers might remember that we have evaluated performance of Core i7 920 in the nominal mode -- with enabled Turbo Boost and Hyper-Threading -- as it works in users' computers.
However, the urge to research is hard to suppress, so we wonder how much performance is added by new technologies used in these CPUs. In this review we shall analyze Core i7 performance in the "minimalistic" mode (without Turbo Boost and Hyper-Threading), with only Hyper-Threading enabled, and with only Turbo Boost enabled. And, of course, we'll compare the results to the nominal mode (both are enabled).
There are some changes in the testbed configuration: in this series of tests we use 6 GB of DDR3 SDRAM memory from Corsair and the ECS X58B-A motherboard. The other parts haven't changed since our previous Core i7 review. Unfortunately, judging by results, our usage of 6 GB of memory instead of 3 GB in the previous series of tests does not allow to compare them directly. So we re-tested the standard configuration with enabled Turbo Boost and Hyper-Threading with the new memory volume and motherboard.
Testing
Legend:
- All Disabled -- both Turbo Boost and Hyper-Threading are disabled.
- TB Enabled -- only Turbo Boost is enabled.
- HT Enabled -- only Hyper-Threading is enabled.
- All Enabled -- both Turbo Boost and Hyper-Threading are enabled.
- Gain, % -- performance gain of the mode in the previous column (to the left). The gain is relative to the "All Disabled" value.
- Negative gains (performance drops) are marked red.
3D modeling and rendering
| |
All Disabled |
TB Enabled |
Gain, % |
HT Enabled |
Gain, % |
All Enabled |
Gain, % |
| 3ds max |
|
| CPU Render |
9.93 |
10.47 |
5% |
11.11 |
12% |
11.61 |
17% |
| Graphics |
4.26 |
4.35 |
2% |
3.88 |
-9% |
4.41 |
4% |
| Hardware Shaders |
10.38 |
10.82 |
4% |
10.01 |
-4% |
10.57 |
2% |
| Maya |
|
| GFX |
2.83 |
2.89 |
2% |
2.81 |
-1% |
2.89 |
2% |
| CPU |
7.70 |
7.90 |
3% |
7.73 |
0% |
8.18 |
6% |
| Render |
0:04:58 |
0:04:41 |
6% |
0:04:07 |
21% |
0:03:55 |
27% |
| Lightwave |
0:07:28 |
0:06:59 |
7% |
0:07:41 |
-3% |
0:07:43 |
-3% |
| Group Score |
118 |
123 |
4% |
120 |
2% |
127 |
7% |
Turbo Boost naturally yields stable performance gains everywhere: according to its general concept, this function shouldn't lead to performance drops. The situation with Hyper-Threading is more complicated: it provides significant performance gains to rendering (3ds max, Maya), only Lightwave suffers from a small performance drop. It's symptomatic that Hyper-Threading leads to performance drops in graphics (3ds max, Maya). We may assume that the problem is in unoptimized video drivers, which also use processor's functions.
CAD/CAM
| |
All Disabled |
TB Enabled |
Gain, % |
HT Enabled |
Gain, % |
All Enabled |
Gain, % |
| UGS NX |
|
| Total CPU |
3.97 |
4.09 |
3% |
3.90 |
-2% |
4.11 |
4% |
| Total Graphics |
1.95 |
1.97 |
1% |
1.87 |
-4% |
1.97 |
1% |
| Pro/ENGINEER |
|
| CPU Related tasks |
569 |
529 |
8% |
570 |
0% |
535 |
6% |
| Graphics Related tasks |
769 |
743 |
3% |
800 |
-4% |
758 |
1% |
| SolidWorks |
|
| Graphics |
80.33 |
71.78 |
12% |
77.44 |
4% |
69.13 |
16% |
| CPU |
44.17 |
41.52 |
6% |
43.68 |
1% |
40.88 |
8% |
| Group Score |
110 |
116 |
6% |
109 |
-1% |
116 |
6% |
This group of tests did not favor even traditional (physical, not virtual) multiprocessing. So it does not come as a surprise that it treats virtual multiprocessing (Hyper-Threading) no better, even worse actually (Group Score). Turbo Boost demonstrates a high average score -- even higher than in the previous group. It's also natural: the weaker effect of multiprocessing on applications, the higher potential to raise frequency they provide for a single core.
Compiling
| |
All Disabled |
TB Enabled |
Gain, % |
HT Enabled |
Gain, % |
All Enabled |
Gain, % |
| VisualStudio |
0:25:39 |
0:23:38 |
9% |
0:24:15 |
6% |
0:22:36 |
13% |
| Group Score |
116 |
126 |
9% |
123 |
6% |
132 |
13% |
It's one of the few tests that shows performance gains from using any of the two "turbo" functions. Even when used together, they show good efficiency.
Professional photo processing
| |
All Disabled |
TB Enabled |
Gain, % |
HT Enabled |
Gain, % |
All Enabled |
Gain, % |
| Photoshop |
|
| Blur |
0:04:42 |
0:04:28 |
5% |
0:04:04 |
16% |
0:03:50 |
23% |
| Color |
0:01:04 |
0:00:58 |
10% |
0:01:20 |
-20% |
0:01:16 |
-16% |
| Filters |
0:04:26 |
0:04:04 |
9% |
0:04:26 |
0% |
0:04:04 |
9% |
| Light |
0:01:42 |
0:01:34 |
9% |
0:01:38 |
4% |
0:01:30 |
13% |
| Rotate |
0:01:52 |
0:01:48 |
4% |
0:01:48 |
4% |
0:01:42 |
10% |
| Sharp |
0:01:52 |
0:01:46 |
6% |
0:01:46 |
6% |
0:01:40 |
12% |
| Size |
0:00:42 |
0:00:40 |
5% |
0:00:42 |
0% |
0:00:40 |
5% |
| Transform |
0:01:28 |
0:01:22 |
7% |
0:01:26 |
2% |
0:01:20 |
10% |
| Group Score |
134 |
143 |
7% |
135 |
1% |
144 |
8% |
To all appearances, performance drop in the "Color" test (RGB-CMYK-LAB conversion and back) has to do with peculiarities of this very procedure, because the rest of Photoshop tests are quite optimistic about Turbo Boost as well as Hyper-Threading. It's small wonder actually: Adobe Photoshop is one of the top SMP-optimized applications.
Scientific applications
| |
All Disabled |
TB Enabled |
Gain, % |
HT Enabled |
Gain, % |
All Enabled |
Gain, % |
| Maple |
0.0346 |
0.0380 |
10% |
0.0346 |
0% |
0.0376 |
9% |
| Mathematica |
|
| Internal |
4.2240 |
4.6050 |
9% |
3.8650 |
-8% |
4.0950 |
-3% |
| MMA |
1.4325 |
1.5519 |
8% |
1.4233 |
-1% |
1.5484 |
8% |
| MATLAB |
|
| LU |
0.0395 |
0.0377 |
5% |
0.0471 |
-16% |
0.0439 |
-10% |
| FFT |
0.0996 |
0.0961 |
4% |
0.0998 |
0% |
0.0946 |
5% |
| ODE |
0.1655 |
0.1542 |
7% |
0.1676 |
-1% |
0.1546 |
7% |
| Sparse |
0.2259 |
0.2175 |
4% |
0.4514 |
-50% |
0.4238 |
-47% |
| 2D |
0.2434 |
0.2349 |
4% |
0.2469 |
-1% |
0.2368 |
3% |
| 3D |
0.5615 |
0.5517 |
2% |
0.5654 |
-1% |
0.5650 |
-1% |
| Group Score |
113 |
120 |
6% |
101 |
-11% |
108 |
-5% |
To all appearances, overt pessimism about Hyper-Threading in most scientific applications indicates their dependence on libraries, optimized for certain processor architectures. There seems to be no generic solutions to optimization in this field. So practically all new architectures are doomed to low results until someone takes pains to optimize intricate math for them. Note that MATLAB "Sparse" test already demonstrated its unpredictability with regard to new architectures-- last time it showed the same dislike for the new architecture from AMD.
Write a comment below. No registration needed!
|
|
 |
|
|
|
