Caches, pipelines and power dissipation
The new Prescott CPU has more cache memory than its Northwood predecessor: both the L1 and the L2 caches are now twice as large as before, at 16KB and 1MB respectively. The Prescott chip also supports SSE3, which includes 13 extra PNI (Prescott New Instructions) commands. However, no applications yet support the new Prescott instructions, so they are currently irrelevant in practice.
Larger L1 and L2 caches should boost Prescott’s speed. The fact that this is mostly not the case -- Prescott is sometimes slower than previous Pentium 4 variants -- is because the chip’s command pipeline has been extended. Think of the pipeline as an assembly-line with several stages: the more stages assembly line has, the faster can it run, in theory. But if the parts (instructions and data) on the assembly-line are not in the correct order, the line must be stopped, corrected, and restarted.
Intel tries to balance this disadvantage with larger caches, which offer faster access to instructions and data than conventional main memory. But larger caches also mean that the power dissipation of the chip rises. The following tables give the results from a variety of PC systems.
| Power dissipation (Watts, with Radeon 7000 GPU) | ||||
| Motherboard |
CPU |
Idle (no load) |
Maximum (full load) |
Cool 'n' Quiet |
| Asus P4C800 | P4 3.2E GHz (Prescott) |
114 |
192 |
n/a |
| Asus P4C800 | P4 3.2 GHz (Northwood) | 76.5 |
144 |
n/a |
| Intel D875PBZ | P4 3.2E GHz (Prescott) |
96.9 |
188 |
n/a |
| Intel D875PBZ | P4 3.2 GHz (Northwood) | 62.7 |
127 |
n/a |
| Asus K8V Deluxe | Athlon 64 3400+ | 113 |
119 |
71.7 |
| Asus K8V Deluxe | Athlon 64 3200+ | 106 |
115 |
70.6 |
| Fujitsu Siemens D1607 | Athlon 64 3400+ | 107 |
114 |
65 |
| Fujitsu Siemens D1607 | Athlon 64 3200+ | 101 |
105 |
62 |
A PC with standard components and the new Intel Prescott chip dissipates nearly 50 Watts more under full load than the same system with the previous-generation Northwood chip. In Idle(no load) mode, the Prescott system still uses nearly 35 Watts more.
| Power dissipation (Watts, with Radeon 9800 Pro GPU) | ||||
| Motherboard |
CPU |
Idle (no load) |
Maximum (full load) |
Cool 'n' Quiet |
| Asus P4C800 | P4 3.2E GHz (Prescott) | 165 |
248 |
n/a |
| Asus P4C800 | P4 3.2 GHz (Northwood) | 125 |
179 |
n/a |
| Intel D875PBZ | P4 3.2E GHz (Prescott) | 145 |
242 |
n/a |
| Intel D875PBZ | P4 3.2 GHz (Northwood) | 113 |
182 |
n/a |
| Asus K8V Deluxe | Athlon 64 3400+ | 167 |
174 |
120 |
| Asus K8V Deluxe | Athlon 64 3200+ | 158 |
168 |
120 |
| Fujitsu Siemens D1607 | Athlon 64 3400+ | 157 |
166 |
114 |
| Fujitsu Siemens D1607 | Athlon 64 3200+ | 148 |
156 |
110 |
Talkback
When will IT departments realise that Intel processors are overpriced, over-hyped and over heated? Choosing AMD processors will be cheaper up front and cheaper in the long run (less power being used).
Unless the company wants to have a distributed, always on, heating system, there is now no reason to choose Intel over AMD.
Thanks for the great review. It was good to see a test that test each application/game in different configurations to find trends... Great job!
Thanks for the only slightly Intel-biased review (Dual DIMMs give the Intel mobos dual-channel RAM over AMD's single-channel single DIMM). While this doesn't make the Intel CPUs dominate the tests, it does make them look not quite as bad as they really are.
I'll remember to look for more reviews on ZDNet in the future - to make sure that I avoid them.
re. the previous comment:
Athlon 64 features only a single channel memory interface. Therefore it is irrelevant to have two DIMMs installed. The Athlon FX offers a Dual-Channel-Interface. With this processor it is important to have two DIMMs installed to get the best performance out of it, as with P4 systems.
re: Dual-Channel
Sorry Kai, I was thinking about the socket 940 FX.
Also, I shall retract my statement about ZDNet. After further review, some of the benchmarks you performed are different enough from the "standard" that (when balanced with other reviews) helps to paint a better overall performance picture. Thanks.
Well, gotta say, there might be some hope for ZDNet afterall. None of the usual Intel butt-kissing and making excuses for whenever it does worse than its competition.
And I'm impressed by the fact that you're using F1 simulations to do some testing of the hardware. Haven't seen that anywhere else. You must be an F1 fan.