SAN FRANCISCO (03/27/2000) - Buy as much memory as you can afford. That's always been the conventional wisdom for Microsoft Corp. Windows users. When memory was dirt cheap and you were only talking about 16 megabytes at a time, that was great advice. But now that RAM is a little pricier and upgrading means adding 64MB at a crack, you're getting into serious money.
So, how much RAM do you really need? The PC World Test Center answered that question, but it wasn't easy: We needed four test machines, three memory configurations, three versions of Windows, our new PC WorldBench 2000 benchmark, and a boatload of patience.
Our conclusion: While more memory is better with all shipping versions of Windows, you don't always get much extra bang for your buck.
Not surprisingly, systems loaded with 256MB of RAM performed the best on our benchmark tests. But systems with just half that amount of memory did almost as well. However, we found that increasing system memory from 64MB to 128MB gave a much more noticable performance boost, making it the right combination of price and performance for most people.
As always, the right decision on what's cost effective for you is in the details. It depends on how you plan to use your PC, and how rich you feel.
How We Tested
We used four systems from mainstream vendors for this test. Each PC has a different processor: a 466-MHz Intel Celeron, a 500-MHz Celeron, a 500-MHz Pentium III, and a 600-MHz Advanced Micro Devices Athlon.
We tested the machines with our new PC WorldBench 2000, based on 11 real-world applications: Microsoft Word, Excel, PowerPoint, Access, and Visio; Lotus WordPro and 1-2-3; Netscape Communicator; Intuit Quicken; Adobe Photoshop; and Corel PhotoPaint.
We examined performance with 64MB, 128MB, and 256MB of standard PC100 Synchronous DRAM. At each memory level, we tested the systems running Microsoft Windows 98 Second Edition and Windows 2000 Professional. We also set up the systems with 128MB of memory under Windows NT 4.0, for comparison's sake.
Just a Little Boost
Adding memory to PCs running Win 2000 triggers the biggest boost in WorldBench 2000 performance. The averaged PC WorldBench score jumps about 7.5 percent with a move from 64MB to 128MB. However, the move from 128MB to 256MB nets an increase of only about 4 percent.
Also worth noting: Under Win 2000, some applications benefit from additional memory more than others. The averaged WorldBench scores for Microsoft Word 97 and Excel 97 don't show meaningful improvement with additional memory. However, Netscape Navigator 4.06's performance leaps nearly 9 percent with the move from 64MB to 128MB, and another 21 percent from 128MB to 256MB. Photoshop 5.0 performance improves nearly 13 percent moving from 64MB to 128MB, and 6 percent from 128MB to 256MB.
More memory also boosts Win 98 performance, but not as much. The move from 64MB to 128MB nets a nearly 6 percent improvement in the averaged WorldBench 2000 score. Jumping from 128MB to 256MB improves performance, but not by a statistically meaningful amount.
As with Win 2000, some WorldBench 2000 programs running under Windows 98 see larger performance gains than others. The meter doesn't move much for Word 97 and Excel 97. Navigator and Photoshop show larger overall improvement.
Navigator jumps more than 5 percent with the move from 64MB to 128MB, and another 8 percent in the move from 128MB to 256MB. Photoshop improves 7 percent, and then 3 percent.
Keep in mind that if you open any two or more of these applications at a time and jump between them (task-switching), or even have two or more actively working at a time (multitasking), you'll accentuate the need for more memory.
Choose Win 2000 for Speed
Of the three operating systems, machines running Win 2000 score the best across the memory segments, besting Win 98 at all levels and nudging past Windows NT at 128MB.
The averaged WorldBench 2000 score of Win 2000 machines running at 64MB is 128, compared to 112 for Win 98, making Win 2000 about 14 percent faster. At 128MB, the Win 2000 system's average score is 138; Windows NT hits 134, and Win 98 lands at 118. That makes the Win 2000 average about 17 percent faster than Win 98, and an insignificant 3 percent faster than Windows NT. At 256MB, the Win 2000 average is 143 and the Win 98 score is 121, making Win 2000 about 18 percent faster.
Our results here indicate Win 2000 is now performing slightly faster than our earlier testing indicated (see Windows 2000 introduction story link at right).
We attribute this change to a number of factors, most likely improvements in video drivers and other routine tweaks in hardware and software.
What's It Worth to You?
Memory costs are moderate these days. One example: A Dell Dimension XPS T desktop with a 600-MHz Pentium III and 64MB of PC100 SDRAM sells for $1349. Add $50 more ($1399) to get 128MB; another $200 ($1599) to reach 256MB. Gateway offers the same jumps for about $90 and $200; Quantex, for about $100 and $240.
Prices from aftermarket vendors also vary considerably, typically in the same ranges. You'll pay between $50 and $100 for 64MB, $150 to $200 for 128MB.
So here's the bottom line:
If you want the fastest machine around (processor and other hardware notwithstanding), go with 256MB of memory (or more) in a computer running Win 2000. You'll pay about $200 more for memory to get those bragging rights.
If money is an issue, but performance is too, go for the 128MB sweet spot, regardless of your operating system. For just a little extra money ($50 to $100), you'll get a nice performance boost over 64MB, especially if you use memory-hungry programs such as Photoshop.
And if you're on a tight budget and plan to stick mostly to word processing and spreadsheet chores, 64MB will get the job done perfectly well.
A Note About WorldBench 2000: These tests are among the first use of PC WorldBench 2000, an update to the PC WorldBench benchmark tests. WorldBench 2000 scores should not be compared with WorldBench 98 scores reported previously.The WorldBench 98 scores are not equivalent to WorldBench 2000 numbers. PC World will retest systems when appropriate, to determine their WorldBench 2000 scores.