Intel's E7205: Granite Bay Hits the Streets
by Evan Lieb on November 18, 2002 9:56 AM EST- Posted in
- Motherboards
ASUS P4G8X Deluxe: Stress Testing
The P4G8X had some pretty decent stress testing potential, mostly because it's based on a dual channel chipset (E7205), which is a more complex and difficult chipset to implement (due to signal integrity, among other issues) than the conventional single channel DDR chipsets that dominate today's markets. Still, we managed to test this board in several different areas and configurations, including:
- Chipset and motherboard stress testing was conducted by running the FSB at 160MHz.
- Memory stress testing was conducted by running RAM at 266MHz and 320MHz in dual DDR operation (two modules each) at the most aggressive timings possible.
Front Side Bus Stress Test Results:
Our FSB stress tests at 160MHz proved to be completely successful. Prime95 torture tests were run in the background for a grand total of 30 hours straight, during which time we performed some general tasks like data compression, various DX8 games (JKII, etc.), and light apps like Word and Excel. We also decided to further stress the P4G8X by randomly rerunning SPECviewperf 7.0, Sciencemark, and XMPEG, all of which couldn't faze the P4G8X.
Memory Stress Test Results:
Our first memory stress test tests how well the P4G8X is able to handle the stock, rated speed the Intel E7205 chipset is validated for (namely dual DDR266 operation). Here were the timings we were able to achieve at the following settings:
|
Stable DDR266 Timings |
|
|
Clock
Speed:
|
133MHz
|
|
Timing
Mode:
|
Turbo
|
|
CAS
Latency:
|
2
|
|
Bank
Interleave:
|
N/A
|
|
Precharge to Active:
|
2T
|
|
Active
to Precharge:
|
5T
|
|
Active
to CMD:
|
2T
|
|
Command Rate:
|
N/A
|
These are exactly the timings you want to see in a truly good dual channel motherboard. You won't be able to squeeze any more performance from your memory with these timings, as there aren't any other BIOS options that would increase performance.
Since the P4G8X doesn't have any other memory dividers other than 1:1, we decided to test how aggressive we could set DRAM timings to running at dual DDR320 (160MHz FSB). Here were our results:
|
Stable DDR320 Timings |
|
|
Clock
Speed:
|
160MHz
|
|
Timing
Mode:
|
Turbo
|
|
CAS
Latency:
|
2
|
|
Bank
Interleave:
|
N/A
|
|
Precharge to Active:
|
2T
|
|
Active
to Precharge:
|
5T
|
|
Active
to CMD:
|
2T
|
|
Command Rate:
|
N/A
|
We see that the P4G8X is able to handle the same aggressive timings at DDR320 as it did at DDR266. Very impressive, but we have to admit, it isn't anything we haven't seen before from dual DDR nForce boards. Still, it's nothing to complain about.
Our final memory test deals with one of the more stressful situations we can think of with the P4G8X. Lets see what happens with all four memory slots full running at 266MHz:
|
Stable DDR266 Timings |
|
|
Clock
Speed:
|
133MHz
|
|
Timing
Mode:
|
Turbo
|
|
CAS
Latency:
|
2
|
|
Bank
Interleave:
|
N/A
|
|
Precharge to Active:
|
2T
|
|
Active
to Precharge:
|
5T
|
|
Active
to CMD:
|
2T
|
|
Command Rate:
|
N/A
|
This is exactly what we like to see. With all four memory banks populated, CAS 2-2-2-5 in Turbo mode is very possible. We used four identical Corsair XMS modules (rated at CAS2 DDR400) for this test. However, mixing memory modules with all four DIMMs populated isn't impossible by any means. With two Corsair XMS modules and two Mushkin modules installed, we were able to just as reliably operate the P4G8X as we were with identical Corsair modules.
As usual, we ran several memory stress tests and general apps to make sure all these timings were stable. We started off by running Prime95 torture tests; a grand total of 24 hours of Prime95 was successfully run at the timings listed in the above charts. We also ran Sciencemark (memory tests only) and Super Pi. Neither stress test was able to bring the P4G8X to its knees.
Facebook
Twitter
RSS
2 Comments
View All Comments
hrumsey - Friday, January 7, 2005 - link
Regarding previous comment:And I told this thing to show e-mail address. hrumsey@charter.net if anyone has questions.
It also removed paragraph indents that would make the above post a bit more readable- apologies.
And a clarification: The ZCR card could be seen to be flashed only because a jumper change is needed to put them in flash mode. In normal mode, the Thunder K8S Pro S2882 BIOS was squashing the Adaptec 2010S / 2015S BIOS.
Damn, I hope Google indexes that comment well.
Speaking of which, for you-know-who:
Tyan Thunder K8S Pro Adaptec 2010S 2015S ZCR RAID BIOS problem incompatibility bug hang failure download flash PCI-X
Tyan 2882 K8S Pro Thunder ZCR Adaptec 2015S 2010S RAID bug hang failure problem incompatibility PCI-X flash BIOS download
Thunder Tyan 2882 K8S Pro ZCR Adaptec RAID 2010S 2015S BIOS incompatibility problem failure hang PCI-X BIOS bug flash download
wildly incompetent screen-reading technical support monkeys
beta-testing on customers
See previous comment
hrumsey - Friday, January 7, 2005 - link
Anandtech's evaluation covers how good Tyan's tech support is in the absence of any real problem for them to deal with. I would suggest that this is not an adequate criterion.Our experiences were different.
The issue of product quality is relevant here, since it makes the quality of technical support more important if the product is poor. My company tried Tyan boards several years ago, and gave up when along with 4 DOAs, 3 quick in-service failures gave a defective rate of almost 50%. I mistakenly thought almost 10 years would be enough for the company to straighten out.
We ordered 3 Thunder Pro S2882s for a client taking a website inhouse who wanted a 64-bit option- this was before Intel's 64-bit Xeons showed up.
All of the following happened under time pressure, which isn't unusual, and why better support than Tyan's is necessary:
One of the three boards was DOA; wouldn't flash any of three Adaptec 2010S ZCR cards; the other two would. Tyan's tech support essentially kept assuming we were doing something wrong and, and at one point asked if we had the current BIOS on the ZCR cards. They must not have any sort of decent database, since the problem had to be explained anew every call. After they admitted the board was bad, they failed to warn us of their shipping deadline for replacing the board (which they will do, and with an E. Coast vendor and them in CA was necessary).
All the boards failed to see the ZCR cards. First tech said that couldn't be happening, second knew about the problem and said the "E" BIOS fixed it. It didn't. We delivered servers with drives unmirrored.
Site setup was busy for a while. When I finally had a chance to work on ZCR problem, Tyan could find no record of the problem (none of the emails we exchanged except ones I sent had case #s in the header). I explained everything again, and once again had to assure them again that we'd gotten the obvious stuff right. First tech said he didn't know how it could be happening, and thought I was missing something. Got email next day from supervisor acknowledging there was a problem and saying (again) they had a new BIOS out that would fix the problem. Downloaded, sent tech onsite to install. Didn't work, same result- ZCR card option grayed out in BIOS, system hangs. When I had a chance to go down and work on it personally, once again, no record of case. I went through everything from scratch once more, assuring them that yes, we'd read the FAQs and yes, the system was plugged in, and yes, we had tried every possible combination of their two blasted relevant jumpers, and that in fact there were about eight other germane parameters we had tried which none of them had thought of- and all of this while wasting valuable onsite time. When I finally convinced them that 1) we were competent and 2) it wasn't working, I was told I'd get a call back "shortly" from the responsible engineer. Three hours later, in a darkened factory, at 5:14:55 just as I was leaving, I got a call back from the engineer who actually knew what was going on. He finally admitted we had everything right. He had no solution, but agreed with my suggestion for testing and said he'd check- he lacked authority(!)- to see if management would authorize the replacement board I'd been asking for. And they did, but there shouldn't have been any question.
Next trip down I replaced the board in one server, picking the server in whichhe Gigabit Ethernet ports had failed- and it still didn't #$%^& work. Tyan said it had been working the day before for them with a 2010S ZCR card, and until today, I didn't know whether they were lying or not. I cussed some and ordered $1200 worth of controllers to replace what Tyan couldn't get right 5 months after the product's release.
Today I checked and saw that they have a new BIOS for the board available that "Fixes PCI ZCR card hangs system during POST". It's the third BIOS for which they've made that claim, and you know, it really doesn't matter whether they're right this time or not. And if they're not, it doesn't matter whether they're just mistaken or actually lying- theend result is the same.
We saw five of their high-end server boards. One DOA, one in-service failure, all five with a major design flaw. Eight years is enough time to take care of company-wide failures. Any company that will release a $500 server board with a 40% failure rate, and without first ensuring that everything on it actually works, and who then can't tell for five months whether or not they've fixed the resulting problems, and whose tech support is staffed with folks who can't deal those problems- well, that's a company whose products you want to steer very clear of.