bit-tech.net

Intel's Core i7 920, 940 & 965 processors

Lavalys Everest 4.60.1540 Beta Memory Performance

Website: Lavalys

Everest Memory Read Test

  • Core i7 965 (4x3.2GHz, 6.4GHz QPI, SMT enabled)
  • Core i7 965 (4x3.2GHz, 4.8GHz QPI, SMT enabled)
  • Core i7 965 (4x3.2GHz, 6.4GHz QPI, SMT disabled)
  • Core i7 965 (4x3.2GHz, Turbo disabled, SMT enabled)
  • Core i7 940 (4x2.93GHz, 6.4GHz QPI, SMT enabled)
  • Core i7 965 (4x3.2GHz, Dual Channel, SMT enabled)
  • Core i7 920 (4x2.66GHz, 6.4GHz QPI, SMT enabled)
  • Core 2 Extreme QX9770 (4x3.2GHz, 1,600MHz FSB, DDR3)
  • Core 2 Duo E8500 (2x3.16GHz, 1,333MHz FSB, DDR3)
  • Core 2 Quad Q9650 (4x3.0GHz, 1,333MHz FSB, DDR3)
  • Core 2 Quad Q9550 (4x2.83GHz, 1,333MHz FSB, DDR3)
  • AMD Phenom X4 9950 BE (4x2.6GHz, 2.0GHz HTT)
  • Core 2 Quad Q9450 (4x2.66GHz, 1,333MHz FSB, DDR3)
  • Core 2 Extreme QX9770 (4x3.2GHz, 1,600MHz FSB, DDR2)
  • Core i7 965 (4x3.2GHz, Single Channel, SMT enabled)
  • AMD Phenom X4 9850 BE (4x2.5GHz, 2.0GHz HTT)
  • AMD Phenom X4 9750 (4x2.4GHz, 1.8GHz HTT)
  • Core 2 Duo E8500 (2x3.16GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q9650 (4x3.0GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q6700 (4x2.66GHz, 1,066MHz FSB, DDR2)
  • Core 2 Quad Q9550 (4x2.83GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q9450 (4x2.66GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q6700 (4x2.66GHz, 1,066MHz FSB, DDR3)
    • 12931.0
    • 12881.0
    • 12853.0
    • 12847.0
    • 12722.0
    • 12550.0
    • 11364.0
    • 9687.0
    • 8502.0
    • 8352.0
    • 8336.0
    • 8259.0
    • 8248.0
    • 8179.0
    • 8135.0
    • 7974.0
    • 7569.0
    • 7528.0
    • 7487.0
    • 7458.0
    • 7447.0
    • 7402.0
    • 6978.0
0
2500
5000
7500
10000
12500
15000
MB/s (higher is better)
  • MB/s

The read performance of the Core i7s is exceptional - and while triple over dual channel doesn't make much difference in Everest, they are both significantly faster than single channel. There's also little difference between the i7 965 and 940 too, and turning all the advanced features like Turbo mode or SMT, or turning down the QPI to 4.8GT/s doesn't really make any difference. The Core i7 920 at 2.66GHz even has more memory bandwidth than the QX9770 at 3.2GHz with faster 1,600MHz (versus 1,066MHz) DDR3.

Everest Memory Write Test

  • Core i7 965 (4x3.2GHz, 6.4GHz QPI, SMT enabled)
  • Core i7 940 (4x2.93GHz, 6.4GHz QPI, SMT enabled)
  • Core i7 965 (4x3.2GHz, 4.8GHz QPI, SMT enabled)
  • Core i7 965 (4x3.2GHz, 6.4GHz QPI, SMT disabled)
  • Core i7 965 (4x3.2GHz, Dual Channel, SMT enabled)
  • Core i7 965 (4x3.2GHz, Turbo disabled, SMT enabled)
  • Core i7 920 (4x2.66GHz, 6.4GHz QPI, SMT enabled)
  • Core 2 Extreme QX9770 (4x3.2GHz, 1,600MHz FSB, DDR3)
  • Core 2 Extreme QX9770 (4x3.2GHz, 1,600MHz FSB, DDR2)
  • Core i7 965 (4x3.2GHz, Single Channel, SMT enabled)
  • Core 2 Quad Q9650 (4x3.0GHz, 1,333MHz FSB, DDR2)
  • Core 2 Duo E8500 (2x3.16GHz, 1,333MHz FSB, DDR3)
  • Core 2 Quad Q9650 (4x3.0GHz, 1,333MHz FSB, DDR3)
  • Core 2 Quad Q9450 (4x2.66GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q9450 (4x2.66GHz, 1,333MHz FSB, DDR3)
  • Core 2 Quad Q9550 (4x2.83GHz, 1,333MHz FSB, DDR3)
  • Core 2 Duo E8500 (2x3.16GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q9550 (4x2.83GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q6700 (4x2.66GHz, 1,066MHz FSB, DDR2)
  • AMD Phenom X4 9950 BE (4x2.6GHz, 2.0GHz HTT)
  • AMD Phenom X4 9850 BE (4x2.5GHz, 2.0GHz HTT)
  • Core 2 Quad Q6700 (4x2.66GHz, 1,066MHz FSB, DDR3)
  • AMD Phenom X4 9750 (4x2.4GHz, 1.8GHz HTT)
    • 12029.0
    • 11942.0
    • 11942.0
    • 11941.0
    • 11941.0
    • 11940.0
    • 9532.0
    • 8520.0
    • 8502.0
    • 8189.0
    • 7094.0
    • 7089.0
    • 7088.0
    • 7087.0
    • 7083.0
    • 7080.0
    • 7069.0
    • 7068.0
    • 6069.0
    • 5038.0
    • 5036.0
    • 4865.0
    • 4499.0
0
2500
5000
7500
10000
12500
MB/s (higher is better)
  • MB/s

In write performance, again the Core i7s have a mammoth bandwidth - compared to AMD's best, the Core i7 has a 238 percent improvement in raw write bandwidth according to Everest. Even running a single channel with the i7s almost matches the previous generation Intel Penryn CPUs running at dual channel 1,600MHz DDR3, and it's a good 1,000MB/s faster than 1,333MHz dual channel DDR3. Like the read performance - the extra Core i7 enhancements do little to offer extra benefit in raw performance.

Everest Memory Copy Test

  • Core i7 965 (4x3.2GHz, 6.4GHz QPI, SMT enabled)
  • Core i7 965 (4x3.2GHz, 4.8GHz QPI, SMT enabled)
  • Core i7 965 (4x3.2GHz, 6.4GHz QPI, SMT disabled)
  • Core i7 965 (4x3.2GHz, Turbo disabled, SMT enabled)
  • Core i7 940 (4x2.93GHz, 6.4GHz QPI, SMT enabled)
  • Core i7 965 (4x3.2GHz, Dual Channel, SMT enabled)
  • Core i7 920 (4x2.66GHz, 6.4GHz QPI, SMT enabled)
  • Core i7 965 (4x3.2GHz, Single Channel, SMT enabled)
  • Core 2 Extreme QX9770 (4x3.2GHz, 1,600MHz FSB, DDR3)
  • AMD Phenom X4 9950 BE (4x2.6GHz, 2.0GHz HTT)
  • AMD Phenom X4 9850 BE (4x2.5GHz, 2.0GHz HTT)
  • Core 2 Extreme QX9770 (4x3.2GHz, 1,600MHz FSB, DDR2)
  • AMD Phenom X4 9750 (4x2.4GHz, 1.8GHz HTT)
  • Core 2 Quad Q9550 (4x2.83GHz, 1,333MHz FSB, DDR3)
  • Core 2 Duo E8500 (2x3.16GHz, 1,333MHz FSB, DDR3)
  • Core 2 Quad Q9450 (4x2.66GHz, 1,333MHz FSB, DDR3)
  • Core 2 Quad Q9650 (4x3.0GHz, 1,333MHz FSB, DDR3)
  • Core 2 Duo E8500 (2x3.16GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q9550 (4x2.83GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q9650 (4x3.0GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q9450 (4x2.66GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q6700 (4x2.66GHz, 1,066MHz FSB, DDR2)
  • Core 2 Quad Q6700 (4x2.66GHz, 1,066MHz FSB, DDR3)
    • 16720.0
    • 15997.0
    • 15708.0
    • 15708.0
    • 15391.0
    • 14222.0
    • 13187.0
    • 9588.0
    • 8866.0
    • 8602.0
    • 8571.0
    • 8275.0
    • 7681.0
    • 7422.0
    • 7416.0
    • 7375.0
    • 7345.0
    • 7208.0
    • 7181.0
    • 7161.0
    • 7109.0
    • 6747.0
    • 5472.0
0
2500
5000
7500
10000
12500
15000
17500
MB/s (higher is better)
  • MB/s

The intelligent memory controller in the Core i7 means the copy performance is even better than raw read and write by 39 percent. In theory this should certainly benefit gaming in theory, but it depends on the efficiency to navigate from PCI-Express bus, via QPI, into CPU and memory then back out again which is technically longer than the previous MCH approach.

The SMT, QPI and Turbo Mode enhancements seem to make more difference here and the performance of triple channel over dual channel is a much greater at nearly 18 percent, while single channel is 74 percent slower. In comparison to the previous generation, the fastest Penryn running DDR3 is 89 percent slower in raw copy performance. If we want to have a bit of a laugh - comparing the clock for clock Core i7 920 versus Core 2 Quad Q6700 at 2.66GHz with DDR3 memory there is a 241 percent performance difference!

The direct connect mechanism in Intel's Core i7 is clearly a superior technology here, although we hope AMD's DDR3 memory controller will do something special to retain a chance of reining in Intel when it arrives.

Everest Memory Latency Test

  • Core i7 965 (4x3.2GHz, Dual Channel, SMT enabled)
  • Core i7 965 (4x3.2GHz, Single Channel, SMT enabled)
  • Core i7 965 (4x3.2GHz, 6.4GHz QPI, SMT disabled)
  • Core i7 965 (4x3.2GHz, 4.8GHz QPI, SMT enabled)
  • Core i7 965 (4x3.2GHz, 6.4GHz QPI, SMT enabled)
  • Core i7 965 (4x3.2GHz, Turbo disabled, SMT enabled)
  • Core i7 940 (4x2.93GHz, 6.4GHz QPI, SMT enabled)
  • Core i7 920 (4x2.66GHz, 6.4GHz QPI, SMT enabled)
  • AMD Phenom X4 9950 BE (4x2.6GHz, 2.0GHz HTT)
  • AMD Phenom X4 9850 BE (4x2.5GHz, 2.0GHz HTT)
  • Core 2 Extreme QX9770 (4x3.2GHz, 1,600MHz FSB, DDR3)
  • AMD Phenom X4 9750 (4x2.4GHz, 1.8GHz HTT)
  • Core 2 Duo E8500 (2x3.16GHz, 1,333MHz FSB, DDR3)
  • Core 2 Quad Q9650 (4x3.0GHz, 1,333MHz FSB, DDR3)
  • Core 2 Quad Q9550 (4x2.83GHz, 1,333MHz FSB, DDR3)
  • Core 2 Quad Q9450 (4x2.66GHz, 1,333MHz FSB, DDR3)
  • Core 2 Extreme QX9770 (4x3.2GHz, 1,600MHz FSB, DDR2)
  • Core 2 Quad Q6700 (4x2.66GHz, 1,066MHz FSB, DDR3)
  • Core 2 Duo E8500 (2x3.16GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q9650 (4x3.0GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q9550 (4x2.83GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q9450 (4x2.66GHz, 1,333MHz FSB, DDR2)
  • Core 2 Quad Q6700 (4x2.66GHz, 1,066MHz FSB, DDR2)
    • 33.0
    • 34.1
    • 38.9
    • 38.9
    • 39.1
    • 39.1
    • 39.3
    • 43.9
    • 53.4
    • 54.6
    • 55.1
    • 57.9
    • 61.9
    • 62.8
    • 63.0
    • 63.7
    • 66.1
    • 70.1
    • 70.7
    • 70.8
    • 71.5
    • 72.3
    • 75.9
0
10
20
30
40
50
60
70
80
Nanoseconds (lower is better)
  • Nanoseconds

The Core i7 latencies are, of course, far lower than the older Front Side Bus and external northbridge model the previous generation Intel CPUs used which is some 67 percent faster for Core i7 when we compare dual channel DDR3 directly. The dual and single channel setups for Core i7 are slightly faster than triple channel which we expect is due to the drop in efficiency as more channels are added.

The Core i7 940 at 2.93GHz doesn't drop too much to the fastest Core i7 965, however the slower 2.66GHz Core i7 920 drops about four to five nanoseconds to these. The performance enhancements of the Core i7s also don't make much difference here either.

What's surprising is that if we compare the Core i7 dual channel DDR3 1,066MHz at 7-7-7-20 to the AMD Phenoms with DDR2 1,066MHz at 5-5-5-15 - the frequencies and channel width are matched and AMD's memory latencies are lower. However the overall memory latency is considerably less because the CPU speed is lower and the northbridge frequency is lower - 2.0GHz on the AMD versus 2.67GHz on the Intel.