The Inquirer-Home

In The Heat of the PC1066...

Granite Bay tested
Fri Jun 14 2002, 13:53
By Nebojsa Novakovic BACK FROM COMPUTEX, two PC1066 32 ns RIMMs -- Samsung 256 MB parts -- were waiting for me here in Singapore. You can guess how impatient I was to get them into the Intel 850EMV reference board for some "non-validated" fun.

And there we went: out goes 512 MB of PC800 45 ns RAM and in goes 512 MB of PC1066 parts. When I turn on the system, the expected "non-validated PC1066" message appears, I say "yes, I want to use it" and go on.

Guess what? The Windows XP refuses to continue, saying the system config has changed and Windows needs reactivation. Since re-activation, unlike activation, needs a dial-up modem to call a local M$ number (no broadband here) and I don't keep such prehistoric stuff in the system, I had to call M$ to get the new ID and log into the system. Geez, how much change is this to warrant the reactivation nuissance?

Just replacing two RIMMs with a same capacity, higher-speed grade?

The even funnier thing was when, before that, we swapped DDR400 (PC3200) with DDR266 (PC2100) DIMM on the Asus P4S533 (SiS 645DX) and Windows refused to boot at all on that system (it was a different installation than the RDRAM system).

Anyway, once that was over, we tested the new configuration with both 2.4 and 2.53 GHz (both 533 MHz FSB) CPUs, to not only test any speed-ups with the faster memory, but also to see how well the P4 scales with frequency, assuming the memory is the fastest possible.

Heh, what to expect? The performance delta between 2.4 GHz PC800 and PC1066 versions on the same 533 MHz bus is higher than between 2.4 and 2.53 GHz with the same memory. Except for the Sandra suite with its wavering results, the benchmark results give between 2% and 7% speedup with PC1066 over PC800 on the same P4-2.4/533 - except the PCMark2002, where the benefit jumps to 14% in the memory test. On the other hand, going from 2.4 to 2.53 GHz with PC1066 gives only between 1% and 3% speedup in the same tests.

The benefit is even less if using PC800, since higher-speed CPU will be stalled more with slower RDRAM.

Sandra did give us some inexplicable results due to varied outcomes in every run (there is no way a PC800 system would be faster than an otherwise identical PC1066 in anything).

So, if profiling a truly balanced high-performance P4 system, keep in mind: P4 is very memory speed dependent compared to some other X86 CPUs, and maximising the memory throughput may be more important than getting the fastest CPU speed parts.

Major problem: Samsung 32 ns 256 MB dual-sided PC1066 RIMMs are much hotter than the PC800 45 ns parts - I had to ensure that the casing has a side fan blowing cold air directly at the RIMMs to ensure truly stable long-term operation. So, if paying a premium for the PC1066 (and that premium is hefty: the only 10 pcs of available PC1066 128 MB Kingston RIMMs here retailed at around US$ 118 today per pc), get ready to add some more $$ for a good multi-fan chasis.

That chassis better not block the RIMMs with any 5.25" storage device bays or messy wires: the ventilation path must be clear.

Now further surprises: our own sources have also done some memory benchmarking on the upcoming Intel 7200 (Granite Bay) dual-channel PC2100 DDR chipset for the Pentium 4 and, guess what, its initial bandwidth results are not any better than the i850E with dual-channel PC1066. Of course, that PC2100 DDR memory is way cheaper. In the meantime, if you get some of the newer chipsets like SiS 648 or other with fast PC3200 (DDR400) support and some Samsung or Corsair PC3200 CL2 DIMMs to boot, you should have the second best performing P4 configuration possible, at quite a bit lower cost than PC1066 configurations.

Again, heat does become an issue for DDR as well at these speed: heat spreaders like those on Corsair XMS DIMMs do help, coupled with good memory area ventilation...

By the way, looks like most i850E mainboards will end up with the old south bridge (not the new ICH4) anyway. I'm curious to see how the upcoming SiS dual-channel DDR and RDRAM chipsets for the P4 will perform, especially once the new 2.66 GHz P4 grades are out in August. µ

See Nova's original review of Intel chipsets. Here.

CPU/Memory P4-2.4 DDRRAM P4-2.4 DDRRAM P4-2.53 RDRAM P4-2.4 RDRAM P4-2.4 RDRAM P4-2.4 RDRAM P4-2.4 DDRRAM P4-2.4 DDRRAM
FSB/Mem Type 533/PC800 400/PC3200 533/PC1066 533/PC1066 533/PC800 400/PC800 533/PC2100 533/PC2100
Chipset SIS 645DX SIS645DX i850E i850E i850E i850E i845E i845G
Motherboard Asus P4S533 Asus P4S533 Intel D850EMV Intel D850EMV Intel D850EMV Intel D850EMV MSI 845 Max2 Intel D845GBV
3DMark2001SE
Default 11162 10813 11971 11808 11239 10968 11188 11204
1600x1200 7019 6963 7482 7424 7280 7208 7334 7265
PC Mark 2002
CPU score 5891 5917 6144 5816 5792 5768 5869 5866
Mem score 5666 5332 6060 5995 5257 5041 5101 5235
SYSmark2002
Internet 328 324 339 329 309 301 300 302
Office 167 166 165 161 150 150 148 149
Sandra
CPU Dhrystone 4455 4646 4871 4623 4722 4603 4671 4611
SSE2Whetstone 2949 2975 3131 2965 2923 2928 2959 2945
Multimedia Int 9540 9628 9997 9472 9489 9496 9570 9523
Multimedia FP 11613 11735 12179 11553 11591 11584 11671 11615
Memory INT 2937 2842 3311 3272 2894 2802 2009 2022
Memory FP 2921 2833 3298 3271 2896 2803 2008 2023

Share this:

Comments

There are no comments submitted yet. Do you have an interesting opinion? Then be the first to post a comment.

aboutus
Advertisement
Subscribe to INQ newsletters
Advertisement
INQ Poll

Authorities in several countries raided Megaupload recently, shut down all of its services, seized hundreds of servers and arrested several of its executives on criminal charges.

Do you think the move was justified?