Very good RDRAM board – Joe
SUMMARY: Stable at its highest FSB – a very good RDRAM motherboard.
The good guys at Iwill were nice enough to send over a sample of their Intel 850E motherboard, the Iwill P4R533-N (feature list). The board comes with integrated 5 channel audio and a Realtek LAN port (we’re seeing more LANs on boards). Both of these functions can be disabled by jumpers.
In addition to the usual feature list, we’re also seeing more boards with integrated Smart Card and Memory Stick ® ports. On the Iwill, there’s no install problem if you don’t install these ports; on some others, if you don’t disable these ports in BIOS, you’ll get an “unknown device” screen when installing Windows. If you don’t know what this refers to, you’ll go crazy figuring it out (at least I did).
For those of you considering an RDRAM board, there are some differences worth noting:
- RIMM modules must be installed in pairs;
- You can not mix different speeds or sizes;
- Sockets must be populated in a specific order;
- Unoccupied RIMM sockets must be populated with a CRIMM¹.
¹Continuity Rambus Inline Memory Module
Iwill packs the board with 2 CRIMMs. The other issue currently facing potential RDRAM users is the relative lack of PC1066 purchase options; Samsung and Kingston are two noteworthy manufacturers. This should (might?) change, but right now it’s a limited selection. Because the slots are populated in pairs, you might as well get 512 MB to start. The board supports PC800 and PC1066 RDRAM, up to 2GB.
CPU Voltage is adjustable up to 1.85v. FSBs can be set from 100 to 156 MHz; the setting at 120 runs PCI at 30 MHz, so running at 100, 120 and 133 will keep AGP/PCI speeds at or under spec. The board packs with a good manual, install CD, drive cables, two CRIMMs and a rear I/O panel.
One thing I did notice is that the 850E chipset gets hot – I measured its temp at the back of the board – it ranged from 40 – 45 C while I was using water cooling. I would strongly suggest that if the board does not have good airflow, look to an active chipset cooler or add a fan aimed at the chipset heatsink.
After setting up the board with a P4 1.6A and Leadtek GeForce 2 Pro, I installed W98se. Once installed, I proceeded with driver installation.
Iwill’s driver disk has an item called “Power Installer” – this is for automated driver installation. I tried this and it quickly installed Intel’s INF drivers – and that was it.
I subsequently found out why this was the only item installed – you can’t install the other stuff until the INF drivers are loaded. So after rebooting, I installed the IAA, Audio and LAN drivers – the first two went off without a hitch.
The LAN driver was a pain – first, clicking on it would open a text file which indicated installation steps. OK – no problem there – pretty standard stuff.
However, for some reason, it took 3 or 4 attempts to get the driver loaded. This happened twice (it went to a system back-up after it refused to boot at 11×156), so it was no first time fluke. I’ve seen this on other boards – why, when installing a LAN, the install disk can’t point to the right driver file is beyond me.
After the LAN was installed, it was on to testing. Recovery from no-boot situations generally were very good – pressing INSERT usually got me back to spec settings. If not, a 3 second CMOS Clear (power off) was all that was needed.
The first test series was to see how stable it was at different FSBs. I was able to run at 145 FSB at the highest RAM setting (RDRAM x4, Samsung PC1066-32), using a P4 1.6A (water cooled). To further test stability, I used an unlocked P4 1700 (an engineering sample) and found the Iwill was stable at its highest setting – 156 MHz. The RAM was the limiting factor, as it would not run at the highest (x4) setting, so I ran the 156 MHz stability tests at x3.
Stability was tested using the UXD RAM Stress Test Pro. This PCI card stresses RAM and I have found it to be an excellent stability test. After the stability tests, I ran a series of benchmarks with the 1.6A:
Benchmark | 16×100 (1600) | 16×133 (2128) | 16×145 (2320) | 11 x 156 (1716)¹ |
SiSandra CPU | 2977 952 1963 | 4000 1268 2619 | 4352 1385 2852 | 3200 1027 2119 |
SiSandra MM | 6283 7655 | 8390 10214 | 9130 11155 | 6791 8231 |
SiSandra MEM | 2484 2487 | 3222 3323 | 3624 3225 | 3331 3330 |
Quake | 185.4 181.5 | 244.8 240.8 | 265.9 264.5 | 199.0 194.6 |
3DMark 2001 | 4299 | 5064 | 5279 | 4522 |
Some impressive memory bandwidths! Quake eats it, as the scores suggest. Note that de-tuning RDRAM to x3 extracts a high toll; RDRAM that could run at 156 MHz x4 will turn in some impressive numbers.
I was curious to see how much FSB speeds contribute to performance, so I ran the P4 1700 at 100 (1700) and 133 (1729) FSB as a comparison:
Benchmark | 17 x 100 | 13 x 133 | % Diff |
SiSandra CPU | 3154 1009 2088 | 3215 1034 2129 | 1.9 2.5 2.0 |
SiSandra MM | 6669 8106 | 6824 8264 | 2.3 1.9 |
SiSandra MEM | 2498 2498 | 3259 3257 | 30.5 30.4 |
Quake | 180.6 176.4 | 203.2 198.6 | 12.5 12.6 |
3DMark 2001 | 4264 | 4553 | 6.8 |
About what I’ve seen in other FSB comparisons – an almost linear impact on memory bandwidth, with Quake benefitting the most and 3DMark 2001 trailing in the single digits. Memory intensive apps benefit the most from changing FSBs at similar CPU speeds.
This is the first RDRAM board we’ve tested and I was particularly interested in how it would compare to other CPU/memory platforms, so I compared it to the P4 Gigabyte GA-8SRX and the Shuttle AK31 with an AMD XP, both running Crucial DDR333. These test only changed CPUs, the motherboard and RAM – the video card was a Leadtek GeForce 2 Pro, Antec 350 watt PS. All CPUs were water cooled.
The P4 I used is an unlocked P4 1700 (an Intel engineering sample) which I ran at 13×133, as was the XP. RAM was Crucial DDR333, CAS 2.5, run at spec.
Benchmark | P4 RDRAM 1066 | P4 DDR 333 | XP DDR 333 |
SiSandra CPU | 3215 1034 2129 | 3189 1012 2085 | 4426 2218 |
SiSandra MM | 6824 8269 | 6668 8065 | 8782 10135 |
SiSandra MEM | 3259 3257 | 1940 1869 | 1977 1783 |
Quake | 203.2 198.6 | 172.3 168.1 | 176.7 174.4 |
3DMark 2001 | 4553 | 4022 | 4682 |
This is consistent with other comparisons I’ve seen – memory bandwidth intensive apps such as Quake shine, but others are not so impacted. Further tuning, such as running memory asynchronously, will further narrow the gap.
The Iwill P4R533-N is a competent performer, exhibiting the usual Iwill stability. There are significant tradeoffs to consider with RDRAM that, in my opinion, limit a user’s flexibility – a choice each must make considering potential benefits.
Further, the on-screen benefits of an RDRAM solution compared to either a P4 or AMD DDR motherboard appear negligible. I don’t find a compelling advantage for one platform against any other at comparable speeds.
Thanks again to Iwill for sending this our way.
Be the first to comment