Original Link: https://www.anandtech.com/show/221
In a word: WOW! After last week's report I thought I'd been mail-bombed! My already high level of respect for Anand just ballooned exponentially. I've got a hunch he added that extra GB of RAM to his server just to handle his e-mail load.
The vast majority of you proved, once again, that the Anandtech family cannot be matched anywhere for intellect or knowledge. I'm very grateful to those of you who sent valuable information and insights. Much of that is included in this weeks report.
For those who respectfully disagreed with some of the information contained in the report, my gratitude also. Many of you brought to light information and/or points that I hadn't thought of. I don't think highly enough of my own opinion to believe that I can't be wrong.
And, lastly, for those who chose to express their opinion with, shall we say, a colorful use of one-syllable vernacular; hey, can't we just all get along?:-)
This week's report covers quite a number of chips compared to previous weeks. It seems that all my time was spent in the Celery kitchen! For the first time, I received a batch of the retail, SL32A CPU's and the results are very interesting. I also did a little experimenting with memory modules on the BH6 in an effort to confirm or deny the reported sensitivity of this board to RAM type/configuration.
One last thing before we move on. The information contained in these reports is simply a discussion of my experiences testing these CPU's. As is the case with all things in life, your results may vary and nothing here is unequivocal.
COMPEQ, Cloverleaf and other symbol trivia
If you remember last week's report, you know that I noted a difference in performance between 2 different symbols on the PCB of the CPU. The first is the COMPEQ symbol and the second, for lack of a better description, I called a Cloverleaf.
This issue generated a lot of mail. Many people provided information to help decipher the symbols, many argued that the PCB itself was not relevant, many related similar experiences, and still others reported no problem overclocking their COMPEQ chips.
It appears from this feedback that the chips with a PCB date prior to 9838 are equally as overclockable as the Cloverleaf CPU's. The majority who stated that theirs worked well, and provided the date, fell in to this category. The majority who experienced problems had a later dated chip. Typical of the latter group is this letter I got from Erics at Fastgraphics:
"Just wanted to let you know that my experience is about the same as yours: I didn't oc 77 of them ;-), but I have one with the second symbol which requires 2.2 volts, a huge PII heatsink with 3 fans and a dedicated air tunnel to run stable at 450.
I recently bought two more Celeron's (with the first symbol) and they appear to work just fine on a dual mainboard running at 450 MHz."
Now, I can't speculate as to whether or not the same would also apply to the "Cloverleaf" PCB as I haven't received enough feedback. It could just simply be that PCB as well as silicon quality was at a higher level in earlier production. This weeks testing, however, doesn't seem to indicate that to be the case.
I believe that the quality of the PCB varies week to week, just as does the quality of the silicon. I also believe that this quality variance can affect the overclockability of the CPU. While I am certainly no expert on this issue, I know that this theory applies to memory modules, so I see no reason why it wouldn't also apply to processors.
A Closer Look
Thanks to Brent Neufelder, we have some excellent shots of the symbols embedded in the PCB. These are located in the upper-right corner of the front side of the chip.
"Cloverleaf" |
"COMPEQ" |
Viewing these symbols is easy on the OEM version of the CPU, as it is not covered by a pre-mounted fan. For the Retail version, you must hold the chip at an angle, but they are still visible.
As it turns out COMPEQ is a manufacturer of a variety of PCB materials, including the substrate for Intel's line of Slot One CPU's. We still haven't been able to gather any information about the second symbol, bit I think it's safe to assume this is also a producer of PCB.
The reverse "RU" on the COMPEQ chips as well as the "94V-0" on both PCB's are Underwriters Laboratory marks. Information on the first can be found here, and on the second, here. I think we can logically assume that the UL marks bear no relationship to overclockability, so that leaves manufacturer and date.
The numbers "9829" and
"9832" are the dates the PCB was manufactured. As I stated earlier, I think that
the quality of this production can vary week-to-week and thus affect the quality of the
entire CPU itself. I've concluded that in order to divine a clue, in advance, we have to
look at both this date as well as the label date placed by Intel when the silicon is
actually mounted.
OEM CPU label
One thing I would like to note, and it's something that was pointed out by many of you who wrote; the sampling included in these reports are very small in comparison to the total of Intel's production of the Celeron. This is absolutely correct. That's why it's important to bear in mind that your results may vary. But, based on the overall overclockability of the Celery, I think some conclusions can be safely drawn when we encounter a dramatic variance in this success rate with an individual batch of CPU's.
In the report itself, I have created a table of results including the relevant information outlined above. More on that later, but right now let's take a look at....
The BH6 and memory
As most of you probably already know, there are reports everywhere of the finicky nature of the BH6 and it's diet as it relates to RAM. I can verify first hand that these reports are true because, during the course of testing the Celery, I've encountered these problems myself.
A couple of examples: One problem I ran in to was with 128MB modules that are only recognized as 64MB. The RAM worked perfectly stable as a 64MB and was recognized at full capacity on another brand of board. At the time, I didn't think to note the brand and type so, sadly, I don't have that information for you.
Another problem, and this one was reported by a customer, is that when 2 modules of 64MB are run at the same time, the board de-stabilizes. Running either one as a single makes the problem disappear. These modules are on their way to me, so when I have them, I'll report on product and chip number.
I've also found that one BH6 may not like a module, while another works fine with the same stick. This happened with a 64MB module of Micron "Elite" 8T. I placed it on a board for testing and it just layed their and beeped at me. Thinking I had a bad stick, I tried another one; same thing. I tried a third, same thing.
I tested each of these very same modules, all 8T's, on a different BH6 and, presto, they worked perfectly.
Many of my customers order RAM burned in with their combo's, so this week I decided to do some tests and take some notes about the modules and combinations that seem to work under all circumstances. Again, I want to emphasize that these results are no guarantee of anything; they are only what I experienced during testing.
All testing was done was done with a Retail Celeron 300A at 450Mhz, default voltage.
Other than as noted above, the Micron 8T's seem to be fine in 64MB size. I tried them as singles, pairs and all slots full with no stability problems.
The Micron 8E's also worked perfectly, both in 64MB as well as 128MB size. I tested these in singles and pairs, as I didn't have quite enough time to test in 3's. And, they do well in mixed double's of 64MB and 128MB.
We have another winner in the Samsung GH 128MB. From single stick to full population, these performed like a champion. I also tried a Micron 8E 128MB with a GH 128MB and they got along just fine.
Lastly, I had some NEC A80-9JF 64MB on hand, so I tried these. With single stick, it got along just fine with the BH6. However, when I tried a pair, I got the dreaded "beep, beep, beep". I only had time to try this on the one board so it's possible, as in the case of the Micron 8T, that this was an isolated problem specific only to this board.
In future reports, I'll try to include any relevant memory discoveries to help you select your RAM. I think it would be a good idea, if you can't find the above types, to but your motherboard and RAM from the same vendor and to ask specifically if they are compatible.
Now, on to this weeks report, but first, let's look at the testing process...
Cooking the Celery
A quick aside for those who have a stubborn chip. We've all heard of the "Celery Sandwich" from Computernerd. This cooler is based on the idea that with some processors it is necessary to cool the area on the PCB directly behind the silicon on the back. This is, however, a rather expensive solution. On this topic, I received this rather enlightening e-mail, with a low-cost alternative, from Peter Varga:
"When I first tried to overclock my Celeron, Windows crashed on boot. After reading your article, I just found out I have the COMPEQ version. Week 35. I didn't change the voltage.. and all I did was a add a small 486 fan to the BACK of the CPU (I have no case so it was a breeze). And guess what.. works beautifully.
"I had an uptime of about 2 days without crashes after which point I decided it was stable. Remember, no voltage change..just added a small fan to the back of the CPU.
"Windows won't even start without the fan."
So, before you give up on your CPU, try Peter's solution.
I test each CPU under initial identical conditions prior to "burning in" my combos. I do an initial test run using the same Abit BH6, same RAM (single stick 64MB Micron, CAS3) and the same video card, a Trident 975, 4MB AGP.
For those who are interested, this is a very fast 2D card that consistently scores better than 5 in Final Reality. Not very good in 3D, but for a business system, it’s a screaming bargain.
I use an old 270MB Quantum hard drive (actually have 3 set up identically, so I can burn multiple combos at once). I use an old hard drive because I assume that these setups will be going in to a variety of situations with a variety of drives and that if it’ll work with this clunker, it’ll probably work anywhere.
I run a quick test at 450 using Business Winstone 97. Why such an old version, you might ask? Because I’m not testing for performance, only stability. This works perfectly for that purpose. If it fails, I bump the core voltage up a notch until it passes.
If the CPU won’t do 450 or requires more than 2.2v core, I set it aside for later evaluation. After passing the first test on my evaluation board, I move the chip to the motherboard with which it will be mated.
I put each one through 6 full rounds of Winstone and 2 hours of Final Reality. By using these two tests, I’m covering both the 2D business users and the 3D gamers. As an aside, for those who haven’t used it, Final Reality is not only a great test, it is visually stunning.
One glitch at current voltage, and it’s bumped up a notch, until it’s completely stable. Because I believe that stability means having a little breathing room, I don’t sell a chip at 450 if it requires more than 2.2v core to stabilize.
Now, this week's results....
This week's report
It appears that there are still plenty of Celery sticks that "do the dance". This week's report covers a total of 60 processors.
For the first time, I got in a batch of Retail, SL32A chips, all of the Malay flavor; total of 15. 10 of these were COMPEQ PCB week 9823 and label code 98391399. Five were Clover, three PCB week 9837, and two were 9835. All five of these were label code 98410558. All fifteen of these CPU's cranked out 450Mhz at default voltage.
Now, there are reports all over the web that the retail versions are, on balance, more overclockable then the OEM's. While this is a small group, and from earlier production, it does seem to confirm this. Since all the processors are identical in CPUID and core stepping, this begs the question: WHY?
Some speculate that Intel "Cherry-picks" the chips for packaging for retail sale. I wonder. With the millions of chips being produced, do they really have time to do this, or is it just coincidence?
The fan that comes with the retails is an interesting piece of design. At first blush, it looks rather wimpy and it spins at a paltry average of 3800 RPM. But, you know what? These Intel engineers are pretty smart cookies; the little thing is very efficient. It keeps the chips barely warm at 450Mhz!
(Please do not e-mail me and ask if you got/are getting/can have one of these chips with your combo. They are already packed for shipping so I have no way of knowing the answer to that question.)
I was also able to grab 15 more of the Clover PCB week 9839/label 08451303 that did so well in last weeks report. They did just as well this week. Thirteen of the CPU's required only default voltage to run at 450, one at 2.1v and one at 2.2v. I think, since we now have 27 chips in this sample, that it's safe to assume this is a good week. 100% of these have overclocked with 89% at 2v.
The next group were thirty of PCB week 9841, label code 08461541. Twenty of these had the normal colored PCB substrate square under the silicon, and ten had the dark one. Of the twenty normal, 90% did 450Mhz, but 2 required 2.3v to make it stable. Of the ten dark, all did 450, but one required 2.3v.
To date, I've tested 137 Celery sticks. Of those, 126 have been stable at 450Mhz or better. This is a success rate of almost 92%. A certain percentage of these required 2.3v, but for most people this would be fine, as the temperature difference is minimal.
I created a chart for those who are in search of their Celery stick at Swap Meets and storefronts where they have an opportunity to inspect the chip prior to purchase. This table includes only the last two weeks results because I wasn't tracking the information this closely until now:
Label Code | # tested | PCB week | Symbol | Color | 2.0v | 2.1v | 2.2v | 2.3v | No Go |
08451303 | 27 | 9839 | Clover | Normal | 24/89% | 2/7% | 1/4% | 0 | 0 |
08450076 | 6 | 9838 | COMPEQ | Normal | 1/17% | 3/50% | 2/33% | 0 | 0 |
08450076 | 6 | 9838 | COMPEQ | Dark | 0 | 0 | 0 | 2/33% | 4/67% |
08461541 | 20 | 9841 | Clover | Normal | 6/30% | 5/25% | 5/25% | 2/10% | 2/10% |
08461541 | 10 | 9841 | Clover | Dark | 2/20% | 6/60% | 1/10% | 1/10% | 0 |
98391399 | 10 | 9823 | COMPEQ | Normal | 10/100% | 0 | 0 | 0 | 0 |
98410558 | 3 | 9837 | Clover | Normal | 3/100% | 0 | 0 | 0 | 0 |
98410558 | 2 | 9835 | Clover | Normal | 2/100% | 0 | 0 | 0 | 0 |
NOTE: Label codes beginning with
"98" are retail versions,
codes beginning with "08" are OEM versions.
One more time for clarity's sake: This information is no guarantee that the CPU you select will be overclockable even if it is one of the one listed at 100%.
Well, happy hunting!
Russ Stringham, Owner
CompuCheap