Performance tests

Testbed configuration:

• Processors:
  • Intel Core 2 Duo E6600 (2.4 GHz, 1066 MHz FSB)
  • Intel Core 2 Duo E8200 (2.66 GHz, 1333 MHz FSB)
  • Intel Core 2 Extreme QX9770 (3.2 GHz, 1600 MHz FSB)
• Motherboards:
  • MSI P35 Neo Combo (BIOS V1.0B16) on the Intel P35 chipset
  • MSI X48C Platinum (BIOS 7.0b6) on the Intel X48 chipset
  • XFX nForce 790i Ultra 3-Way SLI (BIOS P03) on the NVIDIA nForce 790i Ultra SLI chipset
• Memory:
  • 2 x 1 GB Corsair CM2X1024-9136C5D (DDR2-1142, used as DDR2-800 with 4-4-4-12-2T timings)
  • 2 x 1 GB Corsair CM3X1024-1800С7DIN (DDR3-1800, used as DDR3-1066, DDR3-1333, and DDR3-1600)
• Graphics card: ATI Radeon HD 3870, 512 MB
• HDD: Seagate Barracuda 7200.7 (SATA), 7200 rpm

Software:

• OS and drivers:
  • Windows XP Professional SP2
  • DirectX 9.0c
  • Intel Chipset Drivers 8.3.1.1009
  • NVIDIA Chipset Drivers 9.64
  • ATI Catalyst 8.3
• Test applications:
  • RightMark Memory Analyzer 3.8
  • 7-Zip 4.10b
  • Doom 3 (v1.0.1282)

Preface

In this section we are going to compare typical performance of the new chipsets with a representative of the previously tested group. However, it must be noted that we haven't seen anything interesting in this segment for several years already: memory bandwidth hasn't been lower than FSB throughput since dual-channel DDR access. And DDR2 bandwidth is much higher. Peculiarities of a memory controller in the chipset cannot affect test results in such conditions, even if tests depend much on memory bandwidth, e.g. archiving.

Intel hasn't integrated memory controllers into its processors yet, so our tests still make sense. But differences between chipsets in our tests rarely exceed 5%. In a desperate attempt to obtain interesting results, we'll try to compare the new products in several modes instead of only one standard mode - perhaps we'll find some differences.

It must be noted that in some cases our motherboards did not allow to specify identical timings, although we tried to perform our tests in as similar conditions as possible. For example, P35 was tested with DDR3-1066 at 7-7-7-20, while X48 and nForce 790i Ultra SLI motherboards provided much more interesting and real timings - 5-5-5-16-1T. DDR3-1333 results of the NVIDIA chipset were obtained with 6-6-6-18-1T timings, while X48 was tested at 7-5-5-17-1T. We managed to set identical timings only for DDR3-1600 memory - 7-6-6-17-1T. In this context, we ask you to consider results of Intel P35 with DDR3 memory only as a preliminary ballpark, and regard DDR3-1333 modes as similar.

Test results

We'll traditionally start with a low-level analysis of memory potential in our RightMark Memory Analyzer.

The diagram shows well that Intel P35 and X48 practically don't differ from each other in identical modes. The NVIDIA chipset leads in the synthetic RMMA test. DDR3 brings no advantages to the P35 (don't forget about the increased timings of our testbed in this configuration), while X48 accelerates much with this memory, although it fails to catch up with nForce 790i Ultra SLI.

The layout of forces in our memory write speed test looks almost the same, but the NVIDIA chipset wins only with FSB operating at 1066 MHz: Intel X48 catches up with the 1333 MHz FSB, and then noticeably outperforms its competitor with the 1600 MHz FSB.

Our memory read latency test demonstrates similar tendencies as in the memory read rate test. Summing up, we can expect a little advantage of nForce 790i Ultra SLI in real applications (especially over P35 with DDR3). Faster DDR3 memory is apparently preferable here. And now we'll use a couple of real applications to check our assumptions and evaluate performance gains.

Archiving results confirm our conclusions, but performance gains are expectedly lower: although Intel X48 is slightly outperformed by the NVIDIA chipset, performance differences in equal configurations amount to a couple percents only. Compared to DDR2 (at least with its fastest official modification - DDR2-800), performance gain of nForce 790i Ultra SLI is a tad higher, up to 6%. But it must be noted that only X38/P35 officially work with DDR2, X48 doesn't.

Games show a similar picture. Even relative differences between these chipsets look similar. We've only noticed that the tendency to demonstrate lower performance differences at higher memory and FSB frequencies has led to the formal victory of Intel X48.

Conclusions

Today we examined two chipsets (even three) that shook themselves free of DDR2 and introduced support for top Intel processors with the 1600 MHz FSB. We should admit that these solutions offer similar characteristics, possessing support for all necessary technologies to become top products. NVIDIA nForce 790i Ultra SLI can boast of SLI support and more functional Southbridge; Intel X48 - Turbo Memory support, lower heat release, and lower prices. Each of these chipsets supports its own SPD DDR3 extension, so you must take it into account when you choose memory modules.

Both products perform well - they are not slower than their predecessors, and nForce 790i Ultra SLI is even a tad faster than its competitors. As there have appeared lots of processors with 1333 MHz FSB and in connection with the rollout of the 1600 MHz bus and DDR3 memory, the question of choosing a reasonable minimal memory configuration for each FSB clock rate is again a burning issue. We'll publish a separate article devoted to this topic.

Write a comment below. No registration needed!