![]() |
|
![]() |
|
Nehalem Memory Speed InformationThe memory controller in Intel's new Nehalem processor is a radical departure from previous Intel processor architectures, in that the processor die now contains a memory controller. As a result of this change, much of the system builder community is confused as to the proper memory speed to use for Nehalem processors. This paper provides information to system builders so that the proper memory can be selected.
This
information is extracted from Intel documentation. It is correct,
to the best of our knowledge, as of the time it was written.
|
|
This first generation of Nehalem processors has three channels per socket. When the number of DIMMs installed on all channels is the same, the peak memory bandwidth is achieved. If there are different numbers of DIMMs installed on the various channels, the memory bandwidth will be less.
The SiSoft Sandra benchmark is used to determine the memory bandwidth. The Cinebench benchmark is used to indicate the impact of the changes in memory bandwidth on a memory-intensive application. The benchmarks were performed under 64-bit Windows Vista SP1 with no other patches. The appropriate drivers for the motherboard were installed.
Do not confuse memory bandwidth with real-world application performance.
Once all the CPUs in a system are running with at least two channels, the difference in memory bandwidth usually has no impact on actual application performance. Vastly more important is having enough memory to eliminate paging, and to provide enough memory to cache file I/O.
When dealing with a Uni-Processor system, the bandwidth capabilities are rather limited. The processor clocks for the two tested processors are the same. The differences between them are the UnCore clock and the maximum memory clock. With more than three memory modules, both processors will run the memory at 533MHz. The types memory used in these benchmarks are:
Memory Bandwidth |
||||||||||
| W3520 | X5550 | |||||||||
Unbuffered 533MHz (1) Single-Rank |
Unbuffered 533MHz Dual-Rank |
Unbuffered 533MHz Dual-Rank |
Unbuffered 666MHz Single-Rank |
Unbuffered 666MHz Dual-Rank |
||||||
Memory Modules |
Sandra |
Cinebench |
Sandra |
Cinebench |
Sandra |
Cinebench |
Sandra |
Cinebench |
Sandra |
Cinebench |
1 |
6.84 GB/S |
CB-CPU |
7.23 GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
2 |
12.85 GB/S |
CB-CPU |
13.76 GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
3 |
16.51 GB/S |
CB-CPU |
18.42 GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
4 |
13.35 GB/S |
CB-CPU |
13.72 GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
5 |
15.50 GB/S |
CB-CPU |
16.00 GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
6 |
18.17 GB/S |
CB-CPU |
18.50 GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
GB/S |
CB-CPU |
Notes:
(1) This memory is underclocked 666MHz DDR3 CAS-9 Unbuffered ECC Single-Rank, which runs at CAS-7.
With the exception of the 8-DIMM dual-channel test, the DIMMs were installed as evenly as possible between the two processors, and among the channels of each processor. The first set of benchmarks uses the E5520 processor. The types memory used in these benchmarks are:
Unbuffered 533MHz (1) Single-Rank |
Unbuffered 533MHz Dual-Rank |
|||
Memory Modules |
Sandra |
Cinebench |
Sandra |
Cinebench |
1 |
6.42 GB/S |
22532 CB-CPU |
7.00 GB/S |
21376 CB-CPU |
2 |
14.32 GB/S |
21794 CB-CPU |
15.00 GB/S |
21844 CB-CPU |
3 |
14.58 GB/S |
22089 CB-CPU |
15.32 GB/S |
21560 CB-CPU |
4 |
23.27 GB/S |
21936 CB-CPU |
26.15 GB/S |
21365 CB-CPU |
5 |
24.76 GB/S |
21894 CB-CPU |
26.14 GB/S |
22377 CB-CPU |
6 |
29.27 GB/S |
21326 CB-CPU |
31.19 GB/S |
20751 CB-CPU |
7 |
24.90 GB/S |
20991 CB-CPU |
25.44 GB/S |
21307 CB-CPU |
8 |
25.00 GB/S |
22535 CB-CPU |
25.15 GB/S |
20935 CB-CPU |
9 |
23.51 GB/S |
21806 CB-CPU |
25.40 GB/S |
21687 CB-CPU |
10 |
22.56 GB/S |
20993 CB-CPU |
28.85 GB/S |
21735 CB-CPU |
11 |
22.63 GB/S |
22233 CB-CPU |
28.86 GB/S |
20996 CB-CPU |
12 |
24.20 GB/S |
21704 CB-CPU |
31.20 GB/S |
21185 CB-CPU |
8 (dual-channel) |
20.19 GB/S |
22036 CB-CPU |
25.40 GB/S |
21077 GB-CPU |
Notes:
(1) This memory is underclocked 666MHz DDR3 CAS-9 Unbuffered ECC Single-Rank, which runs at CAS-7.
This set of benchmarks uses the X5550 processor. The types memory used in these benchmarks are:
Unbuffered 533MHz Dual-Rank |
Unbuffered 666MHz Single-Rank |
Unbuffered 666MHz Dual-Rank |
Registered 666MHz Dual-Rank |
|||||
Memory Modules |
Sandra |
Cinebench |
Sandra |
Cinebench |
Sandra |
Cinebench |
Sandra |
Cinebench |
1 |
6.72 GB/S |
26300 CB-CPU |
7.00 GB/S |
26677 CB-CPU |
8.56 GB/S |
26430 CB-CPU |
8.42 GB/S |
26976 CB-CPU |
2 |
14.52 GB/S |
26606 CB-CPU |
12.00 GB/S |
25083 CB-CPU |
18.73 GB/S |
26383 CB-CPU |
18.46 GB/S |
26373 CB-CPU |
3 |
16.00 GB/S |
26009 CB-CPU |
13.27 GB/S |
25258 CB-CPU |
18.88 GB/S |
27060 CB-CPU |
18.77 GB/S |
26945 CB-CPU |
4 |
26.68 GB/S |
25081 CB-CPU |
GB/S |
CB-CPU |
32.12 GB/S |
27068 CB-CPU |
31.26 GB/S |
27269 CB-CPU |
5 |
27.25 GB/S |
26121 CB-CPU |
GB/S |
CB-CPU |
32.32 GB/S |
26429 CB-CPU |
32.20 GB/S |
25848 CB-CPU |
6 |
34.00 GB/S |
26909 CB-CPU |
GB/S |
CB-CPU |
38.38 GB/S |
26695 CB-CPU |
36.37 GB/S |
27382 CB-CPU |
7 |
26.28 GB/S |
24671 CB-CPU |
GB/S |
CB-CPU |
26.27 GB/S |
26915 CB-CPU |
26.21 GB/S |
26119 CB-CPU |
8 |
28.30 GB/S |
25884 CB-CPU |
GB/S |
CB-CPU |
26.40 GB/S |
26182 CB-CPU |
26.63 GB/S |
26199 CB-CPU |
9 |
26.84 GB/S |
25892 CB-CPU |
GB/S |
CB-CPU |
26.60 GB/S |
26758 CB-CPU |
26.22 GB/S |
26105 CB-CPU |
10 |
30.38 GB/S |
25187 CB-CPU |
GB/S |
CB-CPU |
30.66 GB/S |
26497 CB-CPU |
31.00 GB/S |
26502 CB-CPU |
11 |
32.22 GB/S |
26794 CB-CPU |
GB/S |
CB-CPU |
31.86 GB/S |
25993 CB-CPU |
32.40 GB/S |
25493 CB-CPU |
12 |
34.14 GB/S |
26755 CB-CPU |
GB/S |
CB-CPU |
35.58 GB/S |
26582 CB-CPU |
34.65 GB/S |
26508 CB-CPU |
8 (dual-channel) |
25.63 GB/S |
26425 CB-CPU |
GB/S |
CB-CPU |
26.14 GB/S |
27365 CB-CPU |
26.23 GB/S |
25461 CB-CPU |
It can be seen that there's a significant difference in the memory bandwidth between 533MHz dual-rank memory and 666MHz dual-rank memory. It can also be seen that there's a similar difference in memory bandwidth between single-rank and dual-rank memory modules.
| Prev | Index | Next |
If you have any comments or suggestions, please E-mail webmaster@accs.com