iXBT Labs - How CPU Features Affect CPU Performance, Part 7 - Page 2: Rendering, computations, bitmaps

3D rendering

	DDR3-800	DDR3-1066		DDR3-1333		DDR3-1600
3ds max ↑	17.17	17.3	1%	18.34	6%	18.69	2%
Lightwave ↓	75.09	74.36	1%	73.05	2%	71.5	2%
Maya ↑	02:13	02:12	1%	02:12	0%	02:09	2%
Group Score ↑	171	173	1%	177	2%	181	2%

	DDR3-1333 3-ch. 6GB	DDR3-1333 2-ch. 4GB		DDR3-1333 2-ch. 6GB
3ds max ↑	18.34	17.88	-3%	17.79	-3%
Lightwave ↓	73.05	73.19	0%	73.18	0%
Maya ↑	02:12	02:12	0%	02:12	0%
Group Score ↑	177	176	-1%	175	-1%

Rendering is not quite sensitive to memory subsystem. This is further confirmed by the results of Core i7.

Scientific computations

	DDR3-800	DDR3-1066		DDR3-1333		DDR3-1600
Maya ↑	9.36	9.8	5%	9.86	1%	10.24	4%
SolidWorks ↓	38.23	38.12	0%	37.83	1%	36.07	5%
Pro/ENGINEER ↓	1548	1513	2%	1496	1%	1473	2%
UGS NX ↓	5.21	5.38	3%	5.55	3%	5.72	3%
MAPLE ↑	0.2023	0.2115	5%	0.2192	4%	0.2227	2%
Mathematica ↑	3.087	3.0892	0%	3.113	1%	3.1403	1%
MATLAB ↓	0.040381	0.040548	0%	0.039324	3%	0.03898	1%
Group Score ↑	135	138	2%	141	2%	144	2%

	DDR3-1333 3-ch. 6GB	DDR3-1333 2-ch. 4GB		DDR3-1333 2-ch. 6GB
Maya ↑	9.86	8.98	-9%	8.79	-11%
SolidWorks ↓	37.83	39.3	-4%	40.66	-7%
Pro/ENGINEER ↓	1496	2005	-25%	1644	-9%
UGS NX ↓	5.55	5.09	-8%	5.09	-8%
MAPLE ↑	0.2192	0.2024	-8%	0.2027	-8%
Mathematica ↑	3.113	2.9639	-5%	2.9287	-6%
MATLAB ↓	0.039324	0.040307	-2%	0.041082	-4%
Group Score ↑	141	128	-9%	130	-8%

Comparing performance at various frequencies, we traditionally see nothing interesting. Just to remind you, we believe the measurement error of ±2% is normal for real application tests. This kind of software was developed to perform certain tasks, not to benchmark performance. So, since there are no other serious reasons, those seemingly abrupt jerks in Maya and SolidWorks can be explained by the accumulation (doubling) of measurement errors when two adjacent results are compared.

The situation with the dual-channel modes is more interesting. Pro/ENGINEER suffers from an insufficient amount of memory, rather than from dual-channel access, in this case as well, so we can disregard its result. But many other applications show much better results on the 4GB machine. We believe this also shows us the performance difference between the dual and single-channel modes. This is surely impressive: disabling one channel reduces performance more than messing with frequencies does.

Bitmap processing

	DDR3-800	DDR3-1066		DDR3-1333		DDR3-1600
ACDSee ↓	04:52	04:48	1%	04:46	1%	04:45	0%
Paint.NET ↓	00:14	00:14	0%	00:14	0%	00:14	0%
PaintShop Pro ↓	08:56	08:54	0%	08:48	1%	08:46	0%
Photoimpact ↓	06:30	06:21	2%	06:13	2%	06:10	1%
Photoshop ↓	05:15	05:07	3%	04:58	3%	04:53	2%
Group Score ↑	149	151	1%	153	1%	154	1%

	DDR3-1333 3-ch. 6GB	DDR3-1333 2-ch. 4GB		DDR3-1333 2-ch. 6GB
ACDSee ↓	04:46	05:36	-15%	04:51	-2%
Paint.NET ↓	00:14	00:14	0%	00:14	0%
PaintShop Pro ↓	08:48	09:34	-8%	09:34	-8%
Photoimpact ↓	06:13	08:37	-28%	07:36	-18%
Photoshop ↓	04:58	06:11	-20%	05:48	-14%
Group Score ↑	153	131	-14%	140	-8%

We can see that memory frequencies have little effect, while many applications do react to memory capacity. The latter has a stronger effect on performance than even the number of channels. It's strange that even such a simple tool as ACDSee (comparing to other programs in this group) suffered that much from the decreased amount of RAM.

Write a comment below. No registration needed!

<< Previous page

Next page >>