Rambus

Intel and Micron imagine a future beyond flash memory

Dave Farquhar Hardware , , , , , ,

In the shadow of Windows 10, Intel and Micron announced a new type of persistent memory that’s 1,000 times faster than the flash memory in today’s SSDs. It’s still not as fast as DRAM, but it’s fast enough that it’s going to make things possible that weren’t before.

Intel and Micron weren’t the first to develop something like this–HP has been working on something similar for years–but HP hoped the product would be out by now, and as far as I know, it didn’t happen. It looks like Intel and Micron’s similar technology is going to happen.

Read more

Hot tip: Memory probably isn’t going to get much cheaper

Dave Farquhar Hardware , , , ,
Hot tip: Memory probably isn’t going to get much cheaper

Friday I saw a story from a financial publication suggesting that DDR3 DRAM prices will be increasing soon due to increasing demand for PCs, thanks to Windows 10’s release and the back-to-school season.

That got me thinking, and while memory prices aren’t at an all-time low right now, they are pretty cheap. A Crucial Ballistix Sport 16GB kit runs about $105 right now. About two years ago, I paid $99 for the same kit. According to the pricing history available to me, the cheapest it’s ever been was $70, and the highest it’s been is $160.

Read more

Lionel bankruptcy

Dave Farquhar Toy trains , , , , , , , , , , , , , , , , , , , , , , , ,
Lionel bankruptcy

It was all over the news when it happened. Lionel, the train maker, filed Chapter 11 on Nov 16, 2004. But a lot of the news stories got some critical details wrong. It’s not the first time a Lionel bankruptcy confused people.

Lionel has been bankrupt before, but the company has changed ownership numerous times so it’s not the same legal entity that went bankrupt in the 1930s and 1960s. There have also been numerous rumors about bankruptcy after 2004. These are usually dealers trying to create artificial demand to clear inventory.

Read more

Reviving a laptop

Dave Farquhar Hardware , , , , , ,

The drive in my work laptop gave a S.M.A.R.T. error over the weekend. I never have had much luck with Hitachi laptop drives. Micron sent a replacement drive–an IBM, thankfully–and, doubly thankfully, the Hitachi hung on until today. So I whipped out Bart’s magic network boot disk–to which I’d added the 3c556 module necessary to get this Micron Transport LT on the network–and ran my copy of Ghost from a network drive. (It won’t fit on that disk, no way, no how. Not with all the other stuff crammed onto it.)
Depending on how far gone the drive is, Ghost can cope with failing hard drives, because you can use the -FRO switch to make it work around bad clusters to the best of its ability. So I initiated Ghost with ghost -z9 -fro (the -z9 tells it to use maximum compression, since the network is the bottleneck here) and made a copy of my disk to a network drive. An hour and a half later (ugh–do I ever miss Token Ring) I had a backup. So I swapped in the IBM drive and repeated the process in reverse. An hour and a half up, an hour and a half down. The data compression wasn’t the bottleneck.

And in the end, I had a healthy laptop again. The IBM drive is quieter and seems faster. I noticed it wasn’t the nice new 5400 RPM model (it’s a 4200 rpm drive) but it’s not a slouch. And it definitely doesn’t clunk as much as the Hitachi always did. I love Hitachi’s video equipment, but their hard drives have always given me trouble. IBM’s laptop drives have always been fine for me. And I know IBM took a lot of black eyes over the GXP desktop series, but think about the things that are known to cause problems with IDE drives:

Rounded cables
PCI bus overclocked beyond 33 MHz
Heat
Cables longer than 18 inches (the length of the wire–not the cable itself)
Certain VIA chipsets in conjunction with Sound Blaster Live! sound cards

IBM 75GXP and 60GXP drives were typically bought by people seeking performance. People seeking performance often do at least one of the above, intentionally or unintentionally. During the 75GXP’s heyday, the hottest chipsets on the block were made by VIA (Intel was still embroiled in the whole Rambus fiasco), and the sound card everyone had to have was the SB Live. I suspect the GXPs were more sensitive to these factors than some other drives and they really weren’t as bad as their reputation.

While rounded cables are good for airflow, they’re bad for signal integrity. Rounded SCSI cables are common, especially in servers, and have been for years, but SCSI takes precautions with its signals–most notably, termination–that IDE doesn’t. That’s part of the reason why IDE is cheaper. So yes, though ribbon cables do look really retro, replacing them with fancy rounded cables isn’t a good idea unless you like replacing hard drives. Get Serial ATA adapters and run your drives serial if you don’t think retro is cool. I’ve been conspiring for the last couple of years to get something semi-modern into my vintage IBM AT case, so I happen to like retro.

But I digress. I hope when the merger between Hitachi’s and IBM’s storage divisions happens, we get the best aspects of both rather than the worst.

Pretentious Pontifications: Finally, a respectable entry-level system

Dave Farquhar Hardware , , , , , , ,

David is still messing around with that ancient 500-MHz Compaq Proliant server, so I am filling in for him today. I threw all of my Pentium III-based systems out for the swine to trample months ago and I suggested David do the same. But, as usual, David refuses to listen to reason.
I see the peasants over at Ars Technica have finally started to show signs of coming to their senses. They have finally designed a personal computer that would be good enough to put in my bathroom. You can read about it here, if you must.

You can tell the people at Ars Technica are peasants, since people with special relationships with Intel (or people who know people with special relatonships with Intel have been running 3.6 GHz Pentium IV systems for weeks. Like I said, the entry-level PC described at Ars Technica is suitable for use in my bathroom. I feel sorry for those who have to putt-putt along on slower equipment in their main PCs. As I have said many times in the past (I am not a revisionist unlike some people), it is incredibly hard to get any serious work done at less than 3.5 GHz.

However, I must salute Ars Technica for getting it correct by using Rambus memory. Rambus memory is demonstrably superior in all regards to the DDR memory used by tyros. Any simple-minded twit can come to that conclusion simply by reading the benchmarks of trustworthy Web sites and looking at the price tag. One does not have to have insider sources like I do to know that.

Unfortunately, I must take issue with their use of IBM hard drives. IBM hard drives are demonstrably inferior to Seagate and Maxtor drives. Everybody knows you cannot power on an IBM hard drive for more than 8 hours a day. Why hundreds of thousands of people use IBM drives in their mission-critical servers is beyond me.

Two chipsets from the AMD front

Dave Farquhar General , , , , , , , , , ,

Yesterday AMD formally unveiled and shipped the AMD-760MP chipset. Right now there is one and only one motherboard using it, the ritzy Tyan Thunder K7, which runs about $550 minimum. (Wholesale cost on it is rumored to be $500.) Considering its 64-bit PCI slots, two built-in 3Com NICs, onboard ATI video, onboard Adaptec SCSI, and four available DIMMs, that’s not a half-bad price. It’s obviously not a hobbyist board. This dude’s intended to go in servers.
At any rate, reviews are all over the place and the quality varies. Far and away the best I found was at Ace’s Hardware, where he tested the things people actually likely to buy this board would do with it: workstation-type stuff.

Anand does his usual 10 pages’ worth of butt-kissing and he’s living under the delusion that people will buy this board to play Quake. However, he does test the board with plain old Thunderbird and Duron CPUs (they work, but AMD won’t support that configuration). Skip ahead to page 11 after reading the story at Ace’s. His tests suggest that for some purposes, a dual Duron-850 can be competitive with a dual P3-933. That information is more interesting than it is useful at this point in time, but we’ve all been curious about dual Duron performance, so if and when an inexpensive AMD SMP board becomes available, we have some idea what we’ll be able to do with it.

All the usual hardware sites put in their two cents’ worth; by the time I read Ace’s and Anand’s and Tom’s reviews I stopped learning anything new.

Some of it bordered on ridiculous. One site (I forget which) observed that the AMD 766 northbridge looks just like a K6-2 and said they must have made it look that way just to remind us where the Athlon came from. Whatever. The AMD 766 northbridge and the K6-2 use the same heat spreader. The intention is to keep the chip cool. It’s not there just for looks–the chip runs hot. But that’s the kind of quality information we get from most hardware sites these days, sadly.

More immediately useful and interesting, but not yet available, is the nVidia nForce chipset. You can read about it at Tom’s and elsewhere. This is technically nVidia’s second chipset, their first being the chipset in Microsoft’s X-Box. This chipset is a traditional two-chip solution, linked by AMD’s high-speed HyperTransport. It includes integrated sound better than anything Creative Labs or Cirrus Logic currently offer (now we know what nVidia was doing with those engineers they were hiring from Aureal) and integrated GeForce 2MX video connected via a high-speed port that would be equivalent to AGP 6X, if such a thing existed. And nVidia pairs up DDR controllers to give dual-channel, 128-bit memory with a bandwidth of 4.256 GB/sec. Suddenly DDR provides greater bandwidth than Rambus in addition to lower latency.

Just for good measure, the chipset includes Ethernet too.

What’s all this mean? High-speed motherboards with everything integrated (and with integrated peripherals definitely worth using) for around 200 bucks. By the end of the summer, last summer’s monster PC will be integrated onto two chips and priced for building PCs at the $600-$800 price point.

This summer’s computer revolution won’t be Windows XP.

And, in something not really related, here’s something you probably missed, unfortunately. Start rubbing your hands together if you enjoy the Mac-PC or Intel-AMD wars. This is a hard benchmark comparing AMD Athlon, Intel P3, and Motorola PowerPC architectures and their relative speed. The methodology: under Linux, cross-compile a Linux kernel for the SPARC architecture (compiling native isn’t a fair comparison; this way they’re all creating identical code and therefore doing the same work, or as close to it as you’re gonna get). You know those claims that a Mac is twice as fast as an equivalent-speed Pentium III running Photoshop? I always countered that with Microsoft Office benchmarks, where a Mac is about 1/4 the speed of a PC, at best, when doing a mail merge. Neither is a fair test. This benchmark resembles one.

Anyway… Yes, a G4 is faster than the equivalently clocked Pentium III. How much faster? Roughly 10 percent. And an Athlon turns out to be about 20 percent slower than the equivalent P3. Of course, the Athlon reaches clock speeds the P3 never will, and the Athlon is also much more than 20 percent cheaper than the equivalently-clocked P3, so who really cares?

This still isn’t a totally fair comparison of CPU architecture, since chipsets vary (and it’s entirely possible that the difference between the P3 and the Athlon in speed is due to chipset quality), but if indeed the G4 was twice as fast as the P3, it would surely outperform it by better than 10 percent in this test. But it’s a decent comparison of real-world performance, because it doesn’t matter how much better your CPU is if it’s burdened by a chipset that doesn’t show up to play on game day.

Most telling is the end, where he gives the cost per speed unit. AMD wins that chart handily.

Enough of my babble. Read all about it here.

More Like This: AMD Hardware

SiS rises from the ashes, and tries to bring AMD and DDR with it

Dave Farquhar Hardware , , ,

Well, I’m back from Bible study (I was teaching on one of those things that can change your life, so I put all kinds of pressure on myself, and I have no idea whether I delivered), but we won’t talk about that right now. No surprises on the Daynotes circuit today; the Weblogs circuit is mostly talking about Kaycee still. I think I’m done with that. I haven’t had time (or will) to go do the cable re-routing necessary to get my new Duron-700 working perfectly.
So, what to talk about…?

How about DDR chipsets?

VIA makes more DDR chipsets than anyone else, and they’ve surprised everyone during the past 18 months, producing chipsets that were much better than anyone expected while Intel produced chipset after chipset that was, for the most part, far worse than anyone’s come to expect of them. Current Intel chipsets work, but they’ve yet to deliver a truly worthy successor to the classic BX chipset. But so far, VIA’s DDR chipsets so far have been disappointing, which makes me wonder if inability to follow up is contagious.

AMD makes a pretty good DDR chipset–at least it gives better performance than PC133 SDRAM, unlike ALi’s DDR chipset and VIA’s DDR chipsets most of the time, and, to be fair, unlike Rambus chipsets–but finding a motherboard based on it can be difficult. AMD’s not very interested in producing the 760, and it shows.

So what’s the DDR chipset to get for AMD CPUs?

Right now, it’s the AMD 760. But very soon, it looks like it’ll be the SiS 735.

Yes, I know, it sounds like I’ve been smoking crack. SiS has a well-deserved reputation for making underachieving chipsets. Just ask Steve DeLassus what he thinks of his SiS 530 integrated video. He’ll throw an Okidata 180 printer at you (ouch) and then tell you it’s almost as bad as the service you get from GPS Computer Services, that’s what.

And the SiS 735 probably isn’t ready for release just yet, as the problems discussed in this review seem to indicate–though whether the problem is with the chipset, the prototype board, or the BIOS, who knows. But the benchmarks indicate the SiS 735 is about 5 percent faster than the AMD 760-based FIC AD11 while costing much less.

Yes, the AD11 isn’t the best-performing 760 board out there, but then again, prototypes aren’t known for stellar performance either. So this sounds promising. Based on these results, it would seem that an Asus or an Abit could produce a very nice-performing board with the SiS735. And as for SiS’s ability to produce a good chipset? Well, these are strange times. Two years ago, AMD bet the company on the Athlon. They had a new, expensive fab they couldn’t afford, dwindling market share and reputation, and a history of botching product releases. If they did everything right and Intel did everything wrong, they had a chance of surviving. Well, AMD executed while Intel fumbled and fumbled. And VIA executed. Intel got caught off guard, and while they’re still king of the hill, they’re embarrassed.

And there was a time, about five or six years ago, when SiS chipsets were actually very sought after. SiS was the first company to produce a chipset that truly brought out the best in Cyrix CPUs, and people who were concerned with raw applications performance sought them out, because the SiS/Cyrix combination outperformed anything Intel was making at the time.

Can SiS rise again? Maybe. It looks like we’re about to find out.

04/06/2001

Dave Farquhar General , , , , , ,

Mailbag:

File Name;Resume; CS; Ad Blocking; 602 Suite; Scary; Plextor

Three days down… The server was down while administrators removed dead sites, in hopes of increasing performance. Performance does seem better, but time will tell… Let’s get on to some serious business.

More memory alphabet soup. JHR wrote in with a good question that I realized I haven’t answered: Can you use your existing plain, cheap old SDRAM on a new DDR-capable motherboard?

The answer, unfortunately, is usually no. DDR comes on 184-pin modules. SDRAM usually comes on 168-pin modules. A few companies, like Fujitsu and Apacer, have talked about putting SDRAM on 184-pin modules. It’s been mostly talk. The price difference between DDR and SDRAM isn’t enough to justify it.

There are a few boards, like the Asus A7A266 (reviewed at http://www.dansdata.com/a7a266.htm ), with both types of sockets for both types of memory. But the A7A266 isn’t the best performer out there, so you pay the price of convenience by buying speed instead. It’s a mediocre DDR performer and a terrible SDRAM performer.

It’s a shame to throw away memory, but this isn’t the first time. As recently as 1997, 72-pin EDO memory cost less than SDRAM. The 72-pin SIMM replaced the 30-pin SIMM as the type of memory to have in 1994, though 30-pin-capable boards remained available for upgraders through 1996. Before 30-pin SIMMs, there were all sorts of weird memory technologies, like 30-pin SIPPs, and different types of individual chips, which generally were a huge pain.

Usually when memory was replaced, adapters came out. There were SIMMs with sockets to plug old chips into. There were adapters to plug a SIPP into a SIMM socket. There were riser cards to allow you to plug 30-pin SIMMs into 72-pin slots. The problem was, they tended to hurt speed and stability, and in many cases they were nearly as expensive as new memory.

History’s repeating itself. There are adapters to let you plug DIMMs into RIMM sockets, and 168-to-184 sockets, though they’re expensive and hurt speed and stability, especially in the case of those RIMM adapters. There’s no point in using them.

I really should have been shouting louder that PC133’s time in the sun is over. The problem is, nobody knows for sure what will replace it. There’s DDR and Rambus, both of which perform really well in certain benchmarks, neither of which seem to make much difference in the real world yet. DDR’s pricing is very close to PC133, assuming you’re buying Crucial. Rambus is still priced way too high. I suspect DDR will win, but there’s no way to know.

It’s a shame to throw out memory, but there usually isn’t much we can do about it. If it makes you feel any better, PCs using SDRAM should be useful for a number of years. I’ve still got two systems with 72-pin SIMMs in them doing useful work for me. One’s a Compaq 486 I bought back in 1994 that just finished a tour of duty as a DSL router; its next incarnation will be as a file/print server if I can find an ISA SCSI card to put in it. I’ll probably also have it automate some parts of my network, courtesy of cron. The other one is a Pentium-120, which has done time as a file server and also as a testbed.

Anything new enough to have SDRAM is new enough to make a very useful Linux box, and it can also make a good Windows box, particularly if you scale it back to just do a handful of things very well. If I ever get around to retiring my K6-2/350, my sister would love to have it because it’d make a great word processing/web browsing/e-mail box–better than the Cyrix 233 she’s using right now, though she doesn’t complain much about that computer. That computer was built out of a bunch of stuff Tom Gatermann and I pulled out of our spare parts bins. And if I did make that switch for her, I know who’d get that Cyrix 233, and that person won’t be complaining either.

The key to responsible upgrading, I think, is to buy stuff that you’ll be able to recycle whenever possible. A good SCSI card and hard drive, though expensive, will be good enough to be worth recycling when you make your next motherboard upgrade. The same goes for a good monitor, and unless you’re a 3D gaming freak, the same goes for a good video card as well. My STB Velocity 128 video card, even though it has an ancient nVidia Riva128 chipset in it, is still fast for the games I play and frankly, it’s overkill for business use. I’ve had that card for three and a half years. I expect I’ll still be using it in three years. Heck, my Diamond Stealth 3D card is still useful. It won’t do justice for my 19-inch display, but it’s fast enough for routine work and it’ll drive a 17-inch monitor at 1024×768 at refresh rates and color depths that won’t embarrass you. And that card’s five years old. It cost me $119 at a time when low-end cards cost $59, and it’s still better for most things than the $40 cards of today. The $25 cards of today will give you higher color depth and sometimes better refresh rates, but they’re not as fast. So that card saved me money. My STB Velocity 128 and my Diamond Viper 770 haven’t been recycled yet, but I’ll get at least three more years’ use out of both of those, even if I turn into a flight simulator fiend. The 770 would be decent for flight sims, and both of them are outstanding for what I do now.

Everyone I know recycles good keyboards and mice, when they think to buy them.

You’ll generally replace motherboards and CPUs on every upgrade cycle. Depending on how often you upgrade, you can expect to replace memory every other cycle.

A lot of people are recommending you buy a motherboard capable of either type of memory, then buy cheap PC133 and upgrade later. But the performance difference isn’t great enough to justify that. If you think you’re going to want DDR, I recommend you just bite the bullet and get DDR. Crucial’s now selling 128 MB PC2100 DDR modules for under $65, so 256 MB of PC2100 costs slightly more than a mid-range video card.

Mailbag:

File Name;Resume; CS; Ad Blocking; 602 Suite; Scary; Plextor

03/18/2001

Dave Farquhar General , ,

About DDR… I should have stated the difference between the two types yesterday. PC1600 DDR runs on a 100 MHz double-pumped bus. PC2100 DDR runs on a 133 MHz double-pumped bus. Obviously PC2100 is much more desirable, providing about 33% as much bandwidth. Crucial is selling PC1600–a fact I didn’t notice–at the price of PC133 SDRAM. That’s less than 50 cents a meg. They aren’t currently selling PC2100 directly, which is what you probably want. PC2100 is currently selling for about a dollar a meg from other sources.

The short term bang-for-the-buck option is to go with a KT133A-based board, a 133 MHz FSB Athlon, and PC133 SDRAM. You’ll get 85-90% of the performance for $100-$150 less. Long-term, however, a DDR solution will make more sense from a performance standpoint and an economy-of-upgrading standpoint. Take a look at what EDO memory costs today and you’ll see what I mean. It’s more expensive than Rambus memory–while Rambus sells for about $2 a meg, antiquated EDO memory sells for about $3 a meg. The price of FPM memory, an even older technology, is over $3 a meg.

So… If you’re swapping out a motherboard and can afford PC2100 DDR, it makes sense to go ahead and get a board that uses it.

What’s this PCxx stuff mean anyway? It’s fairly easy to understand SDRAM monikers–PC100 means the memory bus runs at 100 MHz, PC133 means the memory bus runs at 133 MHz. But manufacturers have gotten ridiculous with the naming schemes of new memory. Along comes Rambus with PC600, PC700, and PC800 memory. But the slowest Rambus memory isn’t 4.5x faster than PC133–far from it. And then comes DDR, not to be outdone, calling itself PC1600 and PC2100.

Here’s what it means. PC600 Rambus is running at a memory bus speed of 300 MHz. PC700 Rambus is using a 356 MHz bus speed. And PC800 Rambus is using a 400 MHz bus speed. CPUs still run at their old bus speeds of 100 or 133 MHz when using Rambus.

Now, PC1600 DDR runs on a 100 MHz bus, while PC2100 DDR runs on a 133 MHz bus. Their names refer to the amount of memory bandwidth available.

So, PCxx isn’t a direct comparison of speed at all. Comparing SDRAM, Rambus, and DDR by their names is like comparing apples, oranges and bananas.

And now for something totally different…

The height of hypocrisy. The RIAA is saying  that paying royalties to songwriters for their work is too difficult–a similar argument to the one Napster used in its defense. The RIAA can’t have it both ways. (Never mind everyone else has to pay to use the songs, and rightfully so.) Hopefully the government will agree. Otherwise the only thing the past year has proven is that the RIAA can bully around anyone who’s smaller than they are.

The story goes like this. Now that the RIAA has turned Napster (who had little ground to stand on) and MP3.com (who had all the ground in the world to stand on) into shells of their former selves, they’re poised to launch their own online service(s). But the RIAA, who represents the record labels, has tried to cut the NMPA, who represents the songwriters, out of the deal.

I’ve heard people advocate pirating music, then tracking down an address for an artist and paying the artist directly. That’s more honorable than paying the RIAA. An honorable and legal approach is to just buy music from artists who also own their record label–when you constantly bend the rules in your favor, it’s hard to keep friends, as the powers that be at the RIAA seem to have not learned on the grade-school playground.

01/28/2001

Dave Farquhar General , , , , ,

I liked how yesterday’s experiment went. So here’s the good stuff I found yesterday.

Laptop intro (Tom’s Hardware Guide)

Aside from spelling errors (notebooks have “gismos,” and PCMCIA network cards connect to CAT5 cable through the use of a “dangle”), this is a pretty good introduction to notebook PCs, covering recent developments like miniPCI and MDC as well and explaining oft-confusing battery technology.

The roundup of video chipsets common in notebooks is nice, and includes the important but easily overlooked power consumption of each solution.

I was disappointed that there was no mention of a previous THG notebook article, http://www4.tomshardware.com/cpu/00q4/001107/index.html , which talked about little-known upgrade paths–by replacing the MMC in a notebook, it’s possible to cross generations. Yes, you can upgrade an old Pentium-based notebook to a P2 or Celeron, assuming you can find an aftermarket MMC.

When you have information like that, there’s nothing wrong with mentioning it whenever another article with similar information gets posted.

These two articles are essential reading if you’re in the market for a laptop, or if your job includes spec’ing and ordering laptops.

EPoX EP-8KTA3 review (AnandTech)

Good discussion of the board’s weaknesses, especially in regards to routing cables and heat dissipation. Heat might be less of an issue if they didn’t assume everyone overclocks, but heat is your PC’s enemy, whether you’re running out of spec or within it. Also good coverage of this board’s special features, including a two-digit diagnostic LCD display on the board. If something goes wrong and it can’t boot, this board will tell you what happened.

Benchmarking is limited to Content Creation, Sysmark, and Quake III Arena under Windows 98, so this is hardly an authoritative evaluation of performance. If you’re into flight sims, racing games, strategy games, or RPG games (let’s face it, first-person shooters aren’t everyone’s thing, and for good reason), Anand’s benchmarks are worthless to you.

This is a decent review, but hardly authoritative. If you’re thinking about buying a KT133A-based Athlon board and you’re considering the EP-8KTA3, you’ll definitely want to look for reviews on another site. You’ll know from reading the KT-133A roundup at THG  that the EP-8KTA3 is a better all-around performer than the Abit K7TA, but you won’t get that from this review.

Mosel Vitelic “PC143” SDRAM (Hardware Daily)

Dangerous, dangerous, dangerous. To wit: “This Mosel Vitelic ram is actually the same as Mushkin Rev2.0 ram. But this one doesn’t have the Mushkin stickers on it and it doesn’t comes with the bubble delivery bag.” Wrong, wrong, wrong. Same chips doesn’t mean same module. Same PCB and same chips doesn’t necessarily mean same module. Here’s the scoop: a 7ns chip may not necessarily run at 7 ns. If a chip runs at 6.9 ns, it’s marked as 7. If it runs at 6.6 ns, it’s marked at 7. If it runs at 7.1, it’s marked as 7.5. What Mushkin’s doing is testing and putting the very fastest 7s on their rev. 3 modules. The second-best go on the rev. 2s. This takes additional testing, which adds to the cost. Buy your Mosel Vitelic memory elsewhere, and you’ll have some 6.6 ns chips and some 7.0s–your results won’t be predictable. One module may run a lot faster than the next. But we’re way ahead of ourselves here.

“According to sisoft Sandra 2001, the chips on this ram is made by Apacer rated at 133mhz.” Wrong again. The reviewer’s hardware knowledge seems as limited as his knowledge of proper English grammar. The chips are made by Mosel Vitelic ( www.moselvitelic.com ), a Taiwanese memory manufacturer who’s been around since 1991 (it was a merger of two companies, each founded in 1983) whose memory is gaining a reputation among overclockers because of its use by Mushkin. Apacer ( www.apacer.com ), on the other hand, makes memory modules.

He then ran some tests in SiSoft Sandra that make this memory look very impressive, but they didn’t do anything to stress-test the RAM to ensure that indeed it was stable at 160 MHz. They also encourage running it at 160 MHz CAS3, which is dubious advice–you get better burst speeds but higher latency that way. That’s precisely the problem with Rambus. How about some benchmarks that more closely resemble real-world performance?

Mosel Vitelic is getting such a reputation that you’ll soon see cheap, generic PCBs with Mosel Vitelic chips on them being sold dirt cheap and bought by misinformed people who read reviews like this and think they’re getting Mushkin-calibre memory for half price.

Mosel Vitelic does make and market their own modules, but that’s not what this is. Manufacturers like Mosel Vitelic and Apacer will be pretty safe, but what you’re paying for when you buy Mushkin is their hand-picking of chips, so you’ll get better, or at least more consistent, results with a Mushkin module.

If you want a near clone of Mushkin memory, you’ll have to look for a module manufactured by Mosel Vitelic themselves (good luck), or by a brand-name maker like Apacer containing 7 ns Mosel Vitelic chips. But you won’t necessarily get the same results.

The review concludes with this: “I highly recommend this ram for people who are looking for good overclocking performance. This teaches us a lesson that good ram isn’t always expensive!”

Unfortunately, the reviewer recommended the wrong thing. The true lesson of this review is that you don’t always get burned when you buy cheap memory, but a few runs of SiSoft Sandra isn’t a good way to test system stability, so this reviewer really doesn’t know what he’s got. He only thinks he does.