Looks like Corsair is releasing their own "Gold" PC3700 modules!

[TysonM]

Looks very similiar to OCZ's GOLD series; high latencies but for a higher clock.

http://www.xbitlabs.com/news/memory/display/20030620030336.html

:rocker:

What's next to find out is its PRICE..... :eek:!!!!!!!

:lol:

[saaya]

errrm you should read the article again :rolleyes:

The new ultra-fast memory from Corsair works at 466MHz clock, but only features CAS3 instead of more aggressive timings.

[Kunaak]

holy crap, what a bunch of crap timings.

"The timings of the new memory modules are relaxed to 3-4-4-8"

theres no way on earth I would consider this ram.

[saaya]

yeah, i think all good 2700 sticks can run 3700@3-4-4-8 ... this is just a bad joke if you ask me.

[Kunaak]

they better sell this ram for like $100 dollars if they expect anyone to take it serious.

[sandman]

I'm willing to bet their 3500 would do better.......

[saaya]

100$? i got hyper x 3500 that does 230+ with cl2 for 80€...

i wouldnt even pay 100$ if it was a 1gb stick! who cares about cl3 memory at all? everybody who buys 3200 or better memory wants performence, so i dont think anybody will buy this ram.

[Resurrected]

I'd rather get GEIL Golden Dragon any day...

[burnout]

sounds like some kind of bad joke to me...

maybe they'll release LL modules later, but what will they kost? 150$?

[Drozdov]

http://www.googlegear.com/jsp/ProductDetail.jsp?ProductCode=85015

$153

[JCviggen]

They rate Corsair XMS3500 at much higher timings than it actually runs as well... I think in real life the sticks could do pretty damn well, they could potentially beat everything else on the market... or it could just be XMS3500 but with CH5 chips and crap timings which will never go much better. Guess someone needs to try

[burnout]

http://www.forumdeluxx.de/gallery/watermark.php?file=500/3259corsair3700-med.jpg
no useable values, but in percents it looks quite good...

[Major]

All the sites that list the XMS 3700 for sale say its rated at cas 2.5 ?? which is it ? I doubt it would be rated at cas 3 ! hehe

My PC3500 goes to 244 @ 3.15v 2-3-2-5.

Corsair and Kingston are pretty conservative on their ratings, where Ocz, Geil, Mushkin, etc.. "push" there ratings.

[WxChaser]

All I can tell you guys, is I saw Ram Guy make a post the other day on the House of Help forums, that addressed the new memory. He was hesistant to discuss a lot of details, but it's designed for the new Canterwood and Springdale chipsets which are not so dependent on the tighter timings (what he said). I've never not had Corsair memory that didn't perform to at least at spec or much higher.

[saaya]

i havent heard anything about what the spd is programmed at, all they say is that it approves the jedec standard for pc 3700 memory, which happens to be a ridiculous 3-4-4-8.

like jc says, it could mean nothing and the meory runs great with low timing, but if it runs with good timings why dont they say so but only state that it approves the jedec 3700 spec? i think it bad memory, if it would run with better timings they would have said that it runs 2.5-3-3-7 like most manufacturers say...

[TysonM]

If youre using an AMD system, RUN AWAY FROM THIS MEMORY. That much is obvious.

The thing is, reviewers are showing that timings do not have as much effect on Canterwoods. So its not as much of a big deal..

Lets see how the reviews on it's performance and overclockability pan out....

[Dissolved]

Originally posted by TysonM
If youre using an AMD system, RUN AWAY FROM THIS MEMORY. That much is obvious.

The thing is, reviewers are showing that timings do not have as much effect on Canterwoods. So its not as much of a big deal..

Lets see how the reviews on it's performance and overclockability pan out.... .


looks like companys quit careing about amd computers, there all trying to go test intel setups i guess for braggin rights?

Seems sad how no one wants to make Good memory for amd systems anymore.

[xgman]

I can only hope that this is not using the new Winbond BH-4 chips.

[Zroc]

It quite true. Raw memory speed is much more important than timings for Canterwood boards. Still and yet, quite possible (very probable) the mem will do better timings than it's conservative ratings.

What JC said is quite true...we won't know about this memory until somebody gets it, strips it, and tells us what the actually memory is.

My guess it's the same stuff OCZ is using.

[TysonM]

Originally posted by Dissolved
.


looks like companys quit careing about amd computers, there all trying to go test intel setups i guess for braggin rights?

Seems sad how no one wants to make Good memory for amd systems anymore.

Actually theres plenty of memory out there that can get good timings up to the 230FSB range..

The problem isnt that they are ignoring AMD, it's that there isnt any better memory out there that can do low latency out there at 250+ FSB yet.

Canterwoods with PAT are showing that they can benefit more from higher FSB/relaxed timings, so thats something Corsair CAN DO while the demand is there. People have been complaining why Corsair hasnt been releasing anything new lately.. well, here they are.

[Creative]

I know everyone says that the memory timings dont really matter on an Intel Canterwood/Springdale setup but come on guys.....I dont really like saying to my mates;

"Yer I run my ram at 300fsb/240mhz but I have to use the timings of 3-4-4-8" :rolleyes:

I dunno, maybe its all in my head, but anything less than 2-6-3-3 is just not an option for me......

can someone slap me and tell me im a fool pls? :p
hehehehe

[dmitriyaz]

:rolleyes:
same exact sticks as the lower-clock-lower-timings ones,
but more expensive.
its been done so many times before :slap:

[WxChaser]

From XMS3700 Press release 6/20:

Corsair Announces XMS3700 DDR Modules

Fremont, CA, June 20, 2003: Corsair Memory, Inc., an industry leader in ultra-performance DDR DRAM modules for gaming and overclocking applications, has announced new XMS3700 additions to their popular XMS family of memory modules. These ultra-fast new modules raise the bar for extreme performance, giving PC enthusiasts further headroom for overclocking their memory bus. Corsair's XMS3700 products are available immediately from resellers worldwide.

Corsair's offering includes 256 MByte and 512 MByte modules available individually or as matched "TWINX" pairs. The new additions consist of eight products, the following four clad in either black or silver (PT) heat spreaders:

CMX256A-3700: 256 MByte, 467 MHz DIMM module
CMX512-3700: 512 MByte, 467 MHz DIMM module
TWINX512-3700: 512 MByte, 467 MHz DIMM module pair, kit consists of two matched CMX256A-3700 DIMMs
TWINX1024-3700: 1 GByte, 467 MHz DIMM module pair, kit consists of two matched CMX512-3700 DIMMs
Each module has been tested at 467 MHz on an Asus P4C800 dual channel DDR motherboard based on the Intel 875P "Canterwood" chipset. TWINX kits are tested as a matched pair and physically packaged together immediately after passing test, guaranteeing that the customer receives a product that has been verified in the dual channel environment. JEDEC standard latency settings of 3-4-4-8 allow these modules to achieve extraordinarily high clock speeds for optimum performance in many system configurations.

After extensive testing of a very broad selection of RAM chips, Corsair engineers designed the XMS3700 around the only RAMs that could pass their stringent qualifications. Corsair's benchmarking showed that faster clock speeds were more critical to improving performance than lower latencies. Therefore the XMS3700 delivers the highest attainable benchmarks with today's fastest available technology.

According to Don Lieberman, Corsair's VP of Engineering, "Our new XMS3700 memory is designed for true enthusiasts who demand the highest levels of performance. By easing up the latency specs on these modules we were able to support faster clock speeds that result in big performance gains...while maintaining the rock solid stability that only TWINX delivers."

Further information and specifications on this product and other XMS products can be found at Corsair's web site (www.corsairmemory.com/xms/). These modules can be purchased through Corsair’s authorized sales channels listed at http://www.corsairmemory.com/xms/xms_sales.html.
About Corsair Memory
Corsair Memory, a member of JEDEC, has been a leader in the design and manufacture of high speed modules since 1994. We have earned our reputation as being the first to market with leading-edge products supporting new computing platforms and technologies. Corsair supplies memory for applications ranging from mission-critical servers to ultra-high performance gaming systems. The performance and reliability of Corsair memory products makes them ideal for the high bandwidth internet infrastructure.

More information from their PDF on XMS3700:

http://www.corsairmicro.com/main/products/specs/twinx-3700.pdf

[Method Man]

3-4-4-8 timings...WTH are they crazy? that will give you §§§§ preformance:stick:

[macci]

I know everyone says that the memory timings dont really matter on an Intel Canterwood/Springdale setup
I wonder what makes em say so. As far as I can tell the timings do make a considerable difference. Here are the results I've got with my Hyperx PC3000 + P4P800 combo. Make your own conclusions:
http://www.solidhardware.com/macci/timings.gif

3DMark2001SE Lobby hi detail test is an excellent bandwidht/latency tester and as you can see 2-2-2-5 timings at 285FSB are equal to 2-3-3-7 timings at 300FSB. So in other words in this case if you decide to drop timings from 2-2-2-5 to 2-3-3-7 you need to OC the FSB by 15MHz and CPU by 180MHz in order to reach the same performance level.
Of course these results will apply only when HyperX modules are used (I'd say that XMS 3200/3500/any BH-5 behaves in similar way too).

We have already seen that OCZ PC3700 performs quite nicely at slow timings and I'm almost certain that Corsair will follow this trend.

[emp]

damn macci, you must be having a lot of spare time, you dont have anything similar for AMD? :)

These PC3700 looks to be teh crap but I really like my 3200LLs :)
Better luck next time Corsair

[BrainStorm]

Personally, I think it's just some of their lower latency stuff run at higher latencies. Otherwise they'd be bringing out new low latency stuff, too. :rolleyes:

[TysonM]

Thats what I love about these forums.. if theres a piece of wisdom being repeated often someone eventually comes out with information that reshapes it more clearly.

I wonder if these modules are coming out at 2.5v? You know how the others are releasing theirs at higher voltage (and with better timings). If so would it overclock a lot better? Ah, cant wait for the reviews to tell the truth soon! [Whether it smacks down the 3-4-4-8 as an Edsel or if a surprise pops out via overclockability]

Tyson

[Creative]

macci, thank you heaps for that little lesson :D

With some extra juice, I believe my ram could be running at 2-2-2-5 too so see how it goes.....That small increase that your pic shows, would be very beneficial for me :)

Thanks again man :)

[xgman]

"I am sorry but its not public as of yet as we are testing a few IC's that may be used but they are -5's
Ram Guy"

[WxChaser]

xgman - I am sorry but its not public as of yet as we are testing a few IC's that may be used but they are -5's"
That's exactly what I was referencing earlier, thanks!

I couldn't find it again this morning when I started this post.

[GVCryan]

I bet they are CH-5's.

Corsair runs bh-6 in the 3200 whcih sould have bh-5 to match the speed.

The run bh-5 in what should run 4.5.

I think they have ch-5 at crappy timings to get 3700. It is very likely that any ch-5 can run 466 3-4-4-8 at 2.5 in an intel board.

If anyone can back this up, I would be grateful.

[Major]

Originally posted by Major Slaughter
My PC3500 goes to 244 @ 3.15v 2-3-2-5.


I'm assuming my chips are winbond on the XMS 3500, that is on a P4C800 plus thats where my cpu tops out so I don't know how much higher they can go. I'll find out with a vapo or promie real soon though ! hehe

Edit: I guess to that I'd have to try it at 2.8v or so for a 1:1 comparison with the XMS 3700 though.

[Zroc]

Originally posted by macci
I wonder what makes em say so. As far as I can tell the timings do make a considerable difference. Here are the results I've got with my Hyperx PC3000 + P4P800 combo. Make your own conclusions:
http://www.solidhardware.com/macci/timings.gif

3DMark2001SE Lobby hi detail test is an excellent bandwidht/latency tester and as you can see 2-2-2-5 timings at 285FSB are equal to 2-3-3-7 timings at 300FSB. So in other words in this case if you decide to drop timings from 2-2-2-5 to 2-3-3-7 you need to OC the FSB by 15MHz and CPU by 180MHz in order to reach the same performance level.
Of course these results will apply only when HyperX modules are used (I'd say that XMS 3200/3500/any BH-5 behaves in similar way too).

We have already seen that OCZ PC3700 performs quite nicely at slow timings and I'm almost certain that Corsair will follow this trend.
macci, they DO matter a little more for 865PE's at a ratio, because there is no PAT or 'psuedo PAT' in effect.
In particular, what most site's are refering to is CAS Latency barely has an effect anymore with PAT (either real PAT or psuedo-PAT at 1:1).

For example, note the last two columns I did awhile ago:
http://www.monsterboxpc.com/roundup2/800.gif

While Sandra shows a bit of difference, everything else is practically equal (some faster, some slower...none big enough to mean anything).

That's the main thing people refer to. In everything I've tested on any of these boards, RAS-to-CAS is the biggest difference maker in real benchmarks and performance. On the flip side, it tends to be the one that allows the biggest gain if you relax it from 2 to 3.

Regardless...in general, timings don't matter as much as they used to...particularly with PAT on (875P at any ratio, or the 'super-865PE's at 1:1 only).

[macci]

In particular, what most site's are refering to is CAS Latency barely has an effect anymore with PAT
Cas latency didn't make any difference on i845 chipset either so thats not a very big news. But that does't mean timings dont make a difference. I'd say that timings matter just as much as they used to matter on i845/E/PE or nForce2 boards.


psuedo-PAT
Why do you say Psuedo-PAT when i865PE is using excatly the same optimizations the 875 is using? :)

At 533FSB 865/865PE/875 are all using the same optimizations:
http://www.x86-secret.com/pics/tweak/pat/parano.gif

Here (http://babelfish.altavista.com/babelfish/urltrurl?lp=fr_en&url=http%3A%2F%2Fwww.x86-secret.com%2Farticles%2Ftweak%2Fpat%2Fpatsecrets-3.htm) is a good article about 865/875 and PAT. Worth reading if you haven't already seen it.

[Tom Holck]

Originally posted by JCviggen
They rate Corsair XMS3500 at much higher timings than it actually runs as well... I think in real life the sticks could do pretty damn well, they could potentially beat everything else on the market... or it could just be XMS3500 but with CH5 chips and crap timings which will never go much better. Guess someone needs to try

I will have them here in a day or to

We vill see:D

[Petrus]

Thanks Macci for that link, I was looking a way to access the device 6 of the NB and finally found it ;)

[Zroc]

Originally posted by macci
Cas latency didn't make any difference on i845 chipset either so thats not a very big news. But that does't mean timings dont make a difference. I'd say that timings matter just as much as they used to matter on i845/E/PE or nForce2 boards.


Why do you say Psuedo-PAT when i865PE is using excatly the same optimizations the 875 is using? :)

At 533FSB 865/865PE/875 are all using the same optimizations:
http://www.x86-secret.com/pics/tweak/pat/parano.gif

Here (http://babelfish.altavista.com/babelfish/urltrurl?lp=fr_en&url=http%3A%2F%2Fwww.x86-secret.com%2Farticles%2Ftweak%2Fpat%2Fpatsecrets-3.htm) is a good article about 865/875 and PAT. Worth reading if you haven't already seen it.
Hmm...well, from the actually design schematics of the chipset that I've seen myself, I'd say that that 'there is no PAT' theory is complete bull§§§§.
I'll give a call to my chipset engineer buddy over at Intel, see if I can get the straight scoop. Gotta love it ;)

There's no spoon, either ;)

[althes]

So does anyone have them yet.;)

[Liquid3D]

Hopefully Soon. I have to give them a lot of credit for releasing a "High Latency" memory, in a trendy LL market. Their more concerned with the idiosyncrasies of the Canterwood/Springdale chipset, which has not meshed too well with any particular memory, regardless of the benchmarks. At least as far as the Springdale's concerned. I don't mean to negate the excellent benchmarks posted here, with 300FSB, but imagine where the Corsair will take you? Doesn't the BIOS have a 400FSB limit?

BTW Excellent article;
"In short, the conclusion is simple: To activate the STALEMATE on a chipset i865PE in mode FSB800/DDR400, it is enough to configure Northbridge in mode FSB533/266 (mode where the STALEMATE is activated by defect) via a modification hardware and then to ask the generator of frequency a frequency adapted to the processor (200 MHz for 800QDR). It is not more complicated."

EDIT: Re-working the OCZ problem.

[PiLsY]

Dont hold your breath guys - these certainly arent OCZ Gold killers. Theyre Samsung 5ns chips.....

The same chips that is on the crucial PC3200 for a third the price (they even have exactly the same timings as crucial at 3-8-4-4) and on most of the cheapest generic 3200 sticks out there for a quarter the price. Theyve just hammered the latencies to get high FSB.

OCZ are using true 4ns chips on their 3700 Gold, meaning theyre using chips from one of three manufacturers - Hynix, Infineon or Micron/M-Tech. My guess would be Hynix. Infineon is historically an average performer, and Micron have always had compatibility problems. Winbond dont have 4ns chips yet, neither do Samsung. Both are sampling for a 4th quarter release from what I can gather though, so we should see some more modules in a couple of months capable of hitting DDR500 with ease.

For now though I wouldnt go near the corsair stuff :stick: . The 3200 Samsung has clocked quite nicely on the more recent boards (early problems on release seem to have been more to do with SPD problems with older single channel motherboards than actual poor clocking), but always at a high price - ie massive latencies. The OCZ Gold has shown that despite needing slacker timings that theyve worked very hard on optimising the module for those slacker timings and pulled back a lot of the potentially lost performance. The Corsair stuff performs inline with results youd expect to see from crap timings (3-3-7 or higher), whereas the OCZ performs far better than you would expect. From results I have seen the OCZ seems to perform basically 1 setting faster than is actually set at (ie 3-3-7 performs like 2-2-6).

When you take into account that the Corsair is retailing for a higher price than the OCZ Gold there really isnt much of a choice :).

That said im certain the corsair module will reach DDR500 and give decent performance, but why bother when you can have better for cheaper? And with nice shiny copper/gold heatspreaders and not nasty grey heatspreaders made from recycled Tuna cans too ;).

PiLsY.

[Liquid3D]

PiLsy I thought OCZ was manufacturing their own Chips, that they actually have access to a Fab? This is claimed in their literature, by Ryanpgroovy, and from a review I found the pIc below?

[charlie]

S-I-L-K-S-C-R-E-E-N

[Liquid3D]

Charlie do you really think their "lying" to us? I don't don't understand in todays market why? It's commonly known companies, purchase, optimize, and assemble the IC's of the major Fab's, so why? The only reason I could think of to go to such trouble, is if their buying really slow chips and optimizing the SPD, or worse selling all different type of crappy chips? But why then are there so many satisfied OCZ users, and how is their memory seems to require less VDIMM to overclock?

[PiLsY]

From what I can surmise OCZ manufacture lower end chips for their slower modules (if they manufacture any at all) and buy in "blanks" for the high end modules.

Blanks are totally unbranded blank chips bought in from a manufacturer of choice, be it winbond, samsung or whoever. These would then be tested and speed binned to the appropriate rated stick (or as many companies do, tested to 3200 speed if they need 3200 that day, even though the ICs may be capable of say 3500). Obviously they would chose a reputable manufacturer and stick to them depending on what stick theyre going on.

Be under no illusions, the sticks themselves are designed, tested, branded and manufacturered by OCZ, but to make ICs capable of the speeds their 3700 Gold is hitting would require manufacturing prowess in excess of the likes of Winbond and Samsung. My own opinion is that they expend all their resources on designing the absolute best possible PCBs, SPD settings and voltage regulation they can around the best ICs currently available. This way you can always go with whatever manufacturers ICs give the highest performance. You arent solely limited to your own brand IC, which may be surpassed by brand X or company Y further down the road, losing you massive quantities of revenue. Afterall we are talking about modules designed for overclockers, who can be a fickle crowd at best ;). We spend our money where the performance is, blind loyalty in the face of inferior performance just isnt in our veins :p.

You also hit another problem which is as you are supplying memory to an enthusiast crowd, how do you continue to produce the lower end modules (such as 2700, or even 3200 now) whilst maintaining production of the high end 3500 and 3700 modules? Its not as simple as just pouring in a diffent mix of silicon into a machine and getting a different IC out the other end ;). Each different IC would require a different production line. When you consider that Samsung's latest factory has 8 production lines and cost them $340 million you begin to see my point ;).


Please note - the above is CONJECTURE ONLY. This is not based on fact or even rumour, its merely my own opinion drawn from a little common sense and some intuition into how the industry works

Taken from the OCZ site product info on the 3700 Gold :

"256MB modules based on OCZ brand EL DDR IC"

You'll notice the key word "brand", ie labelled :). Nowhere on their site do they claim to manufacture their own ICs for the 3700 Gold. This key phrase is repeated throughout the high end modules. I cannot find anywhere on their site where they claim to manufacture ICs, only modules.


The sheer amount of resources required to design a memory module are incredible. You not only have to design the PCB within your own requirements and JEDEC requirements, but also have to set the SPD (which is not just cas, trp, trcd and tras defaults, but a combination of many, many factors such as drive strengths, strobe delays, refresh delays etc), design the voltage regulation circuitry, optimise trace lengths, hell the list is practically endless. If you had to design and manufacture the IC as well it would be nigh on impossible to release and keep up with the competition, let alone stay ahead as OCZ have done, without the backing of billions of dollars and thousands of employees. Only the true giants like micron, samsung, hynix and winbond can actually design, build and produce their own memory, and half the time its on PCBs designed and built by outsourcing and not in house at all.

My own personal opinion is that the silkscreening of the OCZ logo onto the ICs is more to prevent counterfit modules than anything else. OCZ had massive problems with counterfit modules not so long ago which cost them thousands, let alone HUGE amounts fof negative PR when these "dodgy" sticks got into retail. Of course blank modules would be cheaper than branded, but the cost of silkscreening would offset that saving.

I have nothing but the highest respect for OCZ, and have recommended nothing but OCZ 3500, 3700, and Twinmos 3200 modules both online and to friends for the last 6 months. However, to assume that they produced their own ICs for the high end dimms would be taking a very nieve view on things.

What people have also failed to notice is that Corsair have until 2 months ago claimed on their forum and on their site that they "Corsair has been an leader in the design and manufacture of high-speed ICs since 1994", and when you lift the heatspreaders on any module right back to the PC150 sdram they launched their XMS range with in '99 theyre either winbond, micron or samsung chips. Still labelled winbond and samsung. Its now been revised to "Corsair has been an leader in the design and manufacture of high-speed modules since 1994".


Anyway, enough of a ramble for now :).

PiLsY.

[Liquid3D]

I do understand the industry somewhat, perhaps not to the degree you do, but I have read 300mm and even 200mm Wafer Fab's can cost in excess of $2.6-billion. I alos understand were heading into a "thinning out" of Fabs, apparantly their more Wafer's produced (or will be) then is currently needed by the market;
In total, the world's three largest foundry providers-Taiwan Semiconductor Manufacturing Co., Chartered Semiconductor Manufacturing Pte. and United Microelectronics Corp.-are expected to produce 369,000 equivalent 200-mm wafers a week in 2003, said Semico Research Corp. Meanwhile, newer foundries have brought on another 150,000 wafers per week, mostly in "trailing edge" 0.35- to 0.18-micron technologies, Semico said. That astounding 40 percent increase in capacity since 2000 has depressed average selling prices (ASPs) and, arguably, lengthened the current semiconductor recession...

On top of that, seven new fabs-six of them in China-are expected to come online in the next 12 months at 0.18-micron or thicker line widths (see table below), with plans to churn out another 200,000 wafers a week. "There is more capacity than the industry needs today," said Chartered president and chief executive officer Chia Song Hwee...."The capacity glut will continue in 2003, especially for the trailing technology, due to the newly added capacity from new foundries. New foundries will also affect wafer ASPs for 0.35-, 0.25- and some 0.18-micron technologies," said Peter Chang, vice chairman of UMC. http://www.eetimes.com/story/OEG20030317S0012

[Zroc]

Originally posted by Zroc
Hmm...well, from the actually design schematics of the chipset that I've seen myself, I'd say that that 'there is no PAT' theory is complete bull§§§§.
I'll give a call to my chipset engineer buddy over at Intel, see if I can get the straight scoop. Gotta love it ;)

There's no spoon, either ;)
Well holy shiznat, macci...you're right.
They argue semantics way to much in that article, but the theory is right, just mislabled. They're actually both right, oddly enough.

The 875P is ALWAYS on the PAT path. Period. No matter what proc.

On the 865PE, it is on the PAT path with a 533 proc, and on the standard path when an 800 FSB proc is enabled.

What the 'pseudo-PAT' 865PE boards are doing to put it back on the PAT path is tricking it to thinking a 533 proc is installed.
Which actually explains why these boards 'PAT' fails at 5:4 or 3:2...because it DOES. The 533 setting doesn't actually have 5:4 or 3:2 capabilities...only 4:5 and 3:2. So what is happening when one sets these 'PAT' boards to a 5:4 or 3:2 ratio, the CPU setting is being forced back to 800...and back off the PAT path again.

Trippy! Basically, they explain the 'effects' correctly (there is no PAT with the 865PE and an 800 Proc), but not the cause.
So the results are the same, it's just the 'why' that backwards all across the net.

I'll explain more later...

[TysonM]

The question I had was resolved-- ignore this post. --edit

[Zroc]

Oops...fixed it, Tyson.
Yeah, I'll make a post on it, as soon as I remove all the 'marketing speak' from the info ;)

[Major]

Ok my .02 !

I don't care if there made in prison labor camps in china, asymbled in mexico, and silkscreened in someones garage. For that matter I could care less if they are made by one company and remaked to sell by another.

Bottom line = performance/price

[BrainStorm]

Originally posted by Major Slaughter
Ok my .02 !

I don't care if there made in prison labor camps in china, asymbled in mexico, and silkscreened in someones garage. For that matter I could care less if they are made by one company and remaked to sell by another.

Bottom line = performance/price

:rocker:

[Liquid3D]

Originally posted by Major Slaughter
...I don't care if there made in prison labor camps in china, asymbled in mexico, and silkscreened in someones garage. For that matter I could care less if they are made by one company and remaked to sell by another.... That has to be the most common sense statement ever made on these, or any Hardware Forums! People wondered why I defended OCZ for over a year, now look at the numbers of satisfied OCZ users! I was flamed out of Icronic forums because I opened with a statement defending OCZ. Remember the fodder started because somone "claimed" to have photos. He took pictures of a strip-mall, basically making the claim" OCZ uses chimpanzees to glue someone elses silcon to PCB's" And I'll never forget the logical retort at the OCZ site. (my paraphrase follows) We do not want to offend our customers, but we find it preposteous anyone would would perpetuate such ignorance. It's commonly known there are but a few IC manufacturers with Fabs. From these most memory makers purchase their chips...

And now reading MajorSlaughter's remarks I realize how I've been pulled into the semantics of it all. Who cares if OCZ makes their Chips in Zero gravity on the Space Station, as long as they perform, AND THEY HAVE been performing. I'm only glad I chose to stand behind them.

If people want gossip to chew on; read the "Statement of Allegations" made by the OSC (Ontario Securities Commision) which has proven ATI's top exec's inside traded $7.6-million (494,000 shares) of stock in Q3/2000 prior to their earning report which was to show a loss. They've been brought in front of the OSC this Feburary for this; http://www.osc.gov.on.ca/en/Enforcement/Allegations/2003/soa_20030116_ati.htm

Here's the Fabless facts, true the trend is reversing, but MOST semiconductor companies do not have their own Fab;
Would-be semiconductor entrepreneurs are playing by a different set of rules these days. The time when, as the old joke went, all it took to start a company were four Indian engineers and an American guy to do the marketing are over, and startup chip makers are beginning to look like an endangered species.

A total of five chip companies that didn't have their own manufacturing plants were founded last year, according to the Fabless Semiconductor Association. That's a paltry figure compared with the 53 that launched in 2001 and the 92 founded in 2000....The rising costs of designing and producing a chip, at a time when funding is tight and the customer base more discriminating, have turned the screws.

The fabless startup model is under stress...It costs you a lot of money to get to market and get proven silicon working. There's no room for error...With the advent of dedicated foundries like Taiwan Semiconductor Manufacturing Co. in the late 1980s, young companies could focus on design and buy manufacturing capacity as needed, making them partially immune to the wild swings of the industry. Altera, Broadcom, Nvidia, Xilinx and others owe much of their success to this model....

But the days when design and production costs could be ignored because money was easy to come by are no more. Many young fabless companies, for example, can no longer afford the $3 million it takes to buy their own design tools, so they are contracting with ASIC vendors. But this just defers the cost by shifting it to the price of the wafer. Materials, operations and miscellaneous costs can add up to about $4 million. The big one is labor, which can run to $8 million. Factor in the uncertainties of keeping up with Moore's Law and the picture gets more bleak...We'll invest in a fabless company but it's got to address a $200 million to $300 million socket opportunity...
http://www.eetimes.com/story/OEG20030519S0056

[PiLsY]

Arent the above replies exactly what I said but far less diplomatically stated?

You seem to have done an about face, Liquid3D, from saying OCZ do make their own ICs to they definitely don't? For someone who is so adept at quoting other websites you dont seem to be able to read the content that well (regarding OCZ claiming to make their own ICs comment) :p: .

Ah well, as I said before, we pay our cash where the performance is. Id draw the line at chinese prison labour camps though ;).

PiLsY.

[EDIT] Apparently I should listen to my own advice somewhat - I totally missed the "have access to a fab" comment in your first post in our little sub section here Liquid3D, and assumed you thought they HAD a fab. My bad, and my apologies :am: .

Have a banana :banana:. (Just promise you wont try this with it.. :bananal: )

[ryanpgroovy]

I could have taken the time to read the entire thread above but I did not so if I miss something tell me

If someone cares what is done at OCZ here is the short explanation

OCZ does not own a wafer fab , ( in case your wonder neither do many folks including , Nvidia , ATI and VIA ). We will likely never own a wafer fab

OCZ Brand ICs - not the EL DDR ICs , but the lower speed grades ( 2700 , ect ) use blanks , as pilsy mentioned . They are then speed binned marked ( not with a silkscreen but with a laser )
and integrated into our modules . There many reasons for this

The die source on these ICs is Winbond ( the same source for kingston branded IC's) and these ICs are used the premier series of OCZ modules

The package ( IE the part of the chip you see ,the black square on the module ) is a industry standard TSOP



.:banana: Now EL DDR is a bit of a diferent story .:banana:

The wafers used for EL DDR products , are still not manufactured at OCZ , but are packaged using the OCZ designed EL DDR TSOP( a way to reduce heat and increase IC speed) package and as all OCZ ICS do , go through our speed binning process. The speed bin is acomplished using an industry standard advantest 5992 ( http://www.advantest.com/aac/Products/otherFiles/t5592.pdf )

This is the same machine that is used at most tier one DRAM companies ( micron , samsung , ect)

So to be clear, OCZ EL DDR ICs are produced on a contract basis according to the OCZ requirements , we have strict wafer requirements , they are then wafer probed and then shipped to a 3 party for packaging, finaly they are speed binned and the modules are assembled .

This is the simple reason why the OCZ EL DDR parts run so fast :D
much faster I am sure than the samsungs and hynix that people can run above 466 but only at CL3.

If the people want a OCZ Part based on the samsung E rev ICs ( same as corsair is using ) let me know I will tell R&D to build OCZ PC-3700 premier series using Samsung blanks and we can sell it very cheap as a bargain part



We have wafer contracts currently with Winbond , PCS, UMC, TMTC and were currently doing a pilot run with PROMOS .



We also outsource quit a bit of engineering , in case you wondered



Just wanted to put everyone straight as to our current situation
as times move on we will of course adapt .:banana::D

[PiLsY]

And there you go :). Interesting stuff Ryan - I guessed the lower grade memory now (ie not the EL stuff) would be made using the likes of winbond blanks (as you are now guaranteed damned good performance in the 2700 to 3200 range when using these generic ICs), but wasnt sure about the EL stuff. I honestly couldnt see how you had managed to take existing ICs and produce the performance the 3700 Gold has managed to - now its all become clear ;).

As I mentioned in my long and boring essay, theres a lot more to manufacturing ram as people would think. Your post above has prooved this - there are stages there that I didnt know about :p:. A company would have to be truly monolithic to be able to manage all that in-house. Its amazing how little attention people pay to stuff like this - they seem to just assume that the name on the module made the ram from ground up.

The EL series has matured very quickly, when you consider how short a time EL DDR has been available - very impressive. I cant help but think theres more to come from this product line as well - when you consider that the slowest (ie earliest) EL module is a 2700 dual channel part (ie a new design given the dual designation) there has to be more improvements on the way. PC4000 by the end of the year maybe? We can only hope :). 300FSB 1:1 is looking more and more realistic. Id say DDR600 (not as a stock release, but overclocked and rammed with voltage) may even be possible by the end of the year (thats a challenge Ryan :D).

Things must be looking rosy over at OCZ HQ :).

PiLsY.

[ryanpgroovy]

If the people want a OCZ Part based on the samsung E rev ICs ( same as corsair is using ) let me know I will tell R&D to build OCZ PC-3700 premier series using Samsung blanks and we can sell it very cheap as a bargain part



We are here to please the public after all

[PiLsY]

ROFL you sarcastic git :p:.

Very true though - of all the generic modules you find around the samsung quipped are always the cheapest. If this is Corsairs response to the 3700 Gold theyre in a very sorry state. Still, im sure all the blinkered clowns who are part of the "CORSAIRS ARE TEH BESTEST!1" clan will love it.

Still, theres one tactic left for Corsair that youve overlooked Ryan...

[Liquid3D]

Originally posted by PiLsY
Arent the above replies exactly what I said but far less diplomatically stated?
You seem to have done an about face, Liquid3D, from saying OCZ do make their own ICs to they definitely don't? For someone who is so adept at quoting other websites you dont seem to be able to read the content that well Your absolutely right, no need to aplogize. I've been advocating for OCZ for quite awhile, and of recent had been purporting they were manufacturing (as far as owning a Fab) their own memory, based on my seeing those photo's (OCZ IC's) and a few other points I won't get into. Suffice it to say, it was your opinion, which began changing my mind. Having been completely self-taught I used to rely heavily on the word of others in the forums. This was a mistake as even reviewers are in error sometimes. So for the last few months I've completely relied on trade Journals, abstacts, technical papers to educate myself. I didn't mean to doubt your theory, and in fact EVERY point you made which you so humbly claimed as conjecture was right on the money. For this I am thankful your here.

Prior to your mentioning it just yesterday I'd never even heard the term "blanks". However it makes perfect sense, and I do realize what the quote from above says. That most companies start with design tools, and certain facilities, rareky do those include wafer production. It also says that trend is changing, and perhaps that's good as many companies will have full control of the process from "silicon to sale". And start-up monies will require legtimacy, and some real R&D potential to earn those start-up funds!

With Ryan confirming your statements, I am now humbled. "Humility through humiliation, it's not always comfortable, however what's most important is that I continue my intellectual (technical) growth, and not let ego get in the way of the truth. In my reply I mentioned even after reading at OCZ's site a year ago, about their out-sourcing, I began to get into the semantics, and over-complicate things to the point where my believing they had a Fab (perhaps through A-DATA) was feasable. I was wrong.

In so far as not paying attention to the quotes I use, i do realize why I used the quote in the reply above, however where I did err, was on the page before. That quote specifically mentions .35 micron to .18micron line widths (DRAM IC's), which is superflous to our discussion since were discussing either .18-micron, or .13-micron die sizes.

Beyond that, I'm happy Ryan has enough respect for us to share the info he did. And you had very astutely described the process in your claimed "conjecture." I respect your humility as well. You could have just said, "this is the way it is".
I'm happy I learned about "blanks" (and that they don't have the semantic equivilent of, well your banana scenario for example) obviously the blanks OCZ sources have some "life" in them (so to speak). By the by one reason I use quotes, is that people can either verify, or further read where I get my knowledge from. As I said being self-taught, I've a long way to go. I'm not only learning technical facts, I learning a lot about people too.

[charlie]

Yeah, Ryan
Good job being factual, that's cool. I'm about thiiiiisssss (holds up forefinger and thumb) close to trying a pair of Gold PC3700. However I'm concerned about the ability to run middle of the road timing settings...I personally like cas2, 6,3,3 at around ddr530!!
C

[gobbo]

Hmm, why do they recomend it for P4 systems only??

I've still got some PC3200LL stuff from them in my NF2 and its happy at 220fsb @ 2-3-3-7

[TysonM]

What's making me put off the OCZ Gold are the 512mb [1 GB total in dual channel] modules.. until theyre out I'm still searching for the right modules.. :( :bsod: :frag:

[ryanpgroovy]

lquid , no need to be humbled , you were both almost right , I just wanted to try and help clear things up

[Liquid3D]

Originally posted by TysonM
What's making me put off the OCZ Gold are the 512mb 1 GB total in dual channel modules.. until theyre out I'm still searching for the right modules.. I'm running 4x256MB OCZ 3700 GolD and their overclocking like crazy now. If it's easier to overclock 256MB modules (which I've read has always been the case) then 4x256MB might more sense anyways. And if your not requiring more then 1GB total RAM, you may do better top find a reseller that might give you a "two pair" deal? Last night I came close to 6000MB/s (285FSB 3.4GB 5:4) which I know isn't insane, but there are two factors holding me back which are not applicable to others. Number one: I'm using the Abit IS7-E and the chipset just doesn't see the bandwidth the i875 does, and I think the numbers are much greater then 3%-5% difference. Number two: VDIMM, at 2.8V max, I can't tickle the I/O buffers nor effect the (VDDQ) core voltage on these babies? By the by, Ryan. Are these the newer modules which have the "internal volltage regulators" which step down Core IC voltage to 1.8V? OR are these IC's able to benefit from the higher VDIMM?

[ryanpgroovy]

to follow up with those who asked for this

here it is as I promised

http://www.ocztechnology.com/products/memory/OCZ+DDR+PC-3700+Premier

[OPPAINTER]

Cas3,, nice stuff Ryan, how much does it go for??

OPP

[ryanpgroovy]

Its in our bargain line of products , so pretty cheap

[Xenogias]

I'd love to see a comparison between your three lines of pc3700!

[TysonM]

These are the same Samsung rev E modules as in the Corsair?

Should be interesting to see how competitively you can price them!

The thing about the Canterwood is that it's life can be extended in approximately a year by buying a Prescott (bigger cache!) which should be more affordable then.

I want 1 gig of memory and I dont want to sink a huge amount into memory that I may not be using next year [it seems likely that I'd be forced to upgrade memory to broaden the OC for that CPU]....

:banana4:

[eva2000]

Originally posted by ryanpgroovy
Its in our bargain line of products , so pretty cheap hopefully, they make their way down to OZ soon :)

[Zroc]

Ok, wait...so the Corsair PC3700 and OCZ Premier 3700 use Samsung 'E' chips? What exactly are those?

I have a pair of Samsung Original 256MB sticks here that have K4H560838E-TCCC on the chips...are these the same thing?

I'd played with the TCC4 sticks before, and they SUCKED...haven't tried the TCCC sticks yet.

If I'm reading this correctly:
http://www.samsung.com/Products/Semiconductor/DRAM/DDRSDRAM/DDRSDRAMcomponent/256Mbit/K4H560838E/K4H560838E.htm

TCC4 is rated for 3-4-4 timings and 200MHz, but TCCC is rated for 3-3-3 timings and 300MHz ??

[TysonM]

I remember reading a memory review in which something like a (revision C?) Samsung got to DDR500 or slightly higher but at really relaxed timings like youre seeing here.

Pilsy I believe posted in these forums saying that the Corsair 3700 chips were the Samsung Revision E chips..

[Liquid3D]

Originally posted by Zroc
...I have a pair of Samsung Original 256MB sticks here that have K4H560838E-TCCC on the chips...are these the same thing?
... I think those are the ones' K4H560838E-TCCC you may have already tested them by the time I write this, but they may do very well in an 875 board. With high voltage of course. If you read Samsung's spec, the VDDQ is 2.6V as well, as VDIMM so I'd say they would need like 3.1V to get up and dance, but get and up and dance they will. albeit at high latency's 3-3-3-8. My guess is their good for 270FSB t 1:1?

Let's see if I'm completely off the mark, or close?

[ryanpgroovy]

the E-CC are good for 466 at CL3 only though 480 starts to err in memtest

[Zroc]

Originally posted by ryanpgroovy
the E-CC are good for 466 at CL3 only though 480 starts to err in memtest
Cool, thanks. So it started to error for you at 240 1:1?
What motherboard and DDR voltage?

[ryanpgroovy]

IC7 2.85

[Liquid3D]

Nothing flies like OCZ at low voltage! As I've said for a year now. :)

[eva2000]

Originally posted by ryanpgroovy
IC7 2.85 Doesn't IC7 only max out at 2.8v vdimm ? i haven't installed my IC7 yet..

[texuspete00]

It does. I'd bet he's reporting actual. Abit's seems to give a little more here.