Figured it out, seems like the cpu speed plays a role here. 2.25 ghz maxes out at 6600-6700. 2.34 gets 6900-7000. I wonder if going even higher will set a new bandwidth record.
Hmm, I can't seem to duplicate the results here and I don't know what I'm doing wrong. I have 2 sticks of pc3200 ocz vx and a dfi ultra (non-sli). I have the ram running at ddr500 at 2-2-6-2 T1 @3.3v and sandra reports a bandwidth of 6000-6100. The weird part is that sandra reports a timing of 2T but if I drop the timing from 1T to 2T in the bios, the results drop to 5700 so I'm pretty sure 1T is enabled. Any ideas?
iirc, the dfi was tuned for tccd and bh-5/vx.
other stuff that overclocks somewhat does terribly on the dfi, that includes ballistix's micron 5b-g chips.
For everyone hung up on the on-board memory controller...
The motherboard will make a difference for the on-board memory controller, not because of the chipset, but due to the characteristics of the routing. The type of material used, the thickness and width of the copper traces, and how well the traces are matched in length, will all make a huge difference in signal integrity. Even differences in what layers of the board you route to make a difference, and what power plains they're coupled against are. The list goes on...
DFI put a lot more effort into their routing, and it makes a significant difference.
With most comments before this one praising this RAM as if it were the Holy Messiah Himself, my comment will surely be seen as the Devil but the truth is, this article is biased. Mr. Fink doesnt know what he's on about. He's gone and tested the VX Gold and pitted the test results against the results obtained from other RAM on a completely different board. If you would take the time to indulge me, you will see that he's simply copied and pasted the performance results of RAM taken from the MSI Neo2 (NF3) and then compared these results to what he obtained with the VX Gold on the DFI LANParty nF4 SLI-DR. I cannot express in any written language how inconsistent this is. If I didn't know better, I'd say OCZ handed him a nice little duffle bag sporting a $ sign after writing this article.
The fact is, OF COURSE the OCZ VX Gold will look like God when you compare RAM the way Mr.Fink has done. The test bed he's used for comparison has not been consistent. So you say "Who cares, its an onboard memory controller. It shouldn't matter which board you use". Do some research and you'll see that the previous statement is rediculous. Mr. Fink tested the DFI LANParty nF4 SLI-DR when it was first available and had the following results with the OCZ PC3200 Platinum Rev.2: "Quake 3 ran at 642FPS and SiSoft Sandra 2004 standard memory bandwidth was 8,300 MB/s. The Sandra unbuffered memory bandwidth was at 4000 MB/s" ( http://www.anandtech.com/mb/showdoc.aspx?i=2337&am... ). A direct comparison of these results to when the SAME RAM was tested on the MSI Neo2 ( http://www.anandtech.com/memory/showdoc.aspx?i=231... ) will show that the board itself makes all the difference, in DFI's favor. Considering the Corsair twinx1024-4400c25 outperformed the OCZ on the MSI, I still cant understand why we havent seen this RAM tested on the DFI yet. In any case, you cannot compare the test results of the VX Gold to those obtained by other RAM on the MSI board and thus the results and conclusion of this article is invalid.
I've emailed Mr.Fink regarding the inconsistencies of his articles and have outlined what we want as consumers. Because of the nature of the DFI board, I believe anandtech should test the top 5 RAM modules using this board (including the Corsair twinx1024-4400c25) and present us with results from an unbiased, completely controlled testbed so we, as consumers, can make a more informed decision when deciding which RAM to purchase.
Guys, have a look at the front page of the ORB for 3D2001, many of the top benchers use ram thats running 2-2-2 at 250+.
You have to remember here the VX beat out OCZ's 3200Plat rev2 which is also TCCD based so Wesleys review while showing one product in a good light shows another in a not so good.
Nvidia have said that the NF4 and NF3 chipsets clock ram and perform equal with the same memory, i have tested both boards here and found the same, if you want proof do a few days of testing yourself, remember the memory controller is on the CPU, not the chipset ;-) so actual bandwidth will probably be identicle.
This debate has been runmbling for years, whether high fsb's high latency beat low latency at lower fsb's, we debated for ages with the 875 chipsets but the cross over for performance was a lot lower down the fsb range, and boards are now clocking ram better than they ever did. Many prefered 250fsb 5:4 with 2-2-2 over 250 1:1 with 2.5-4-3 as the benches were always higher in games running async, you needed 265fsb for the high latency 1:1 to equal the 5:4 score and then pull away.All we see now is a much more flexable memory controller which just like the P4 C Northwoods is low latency, it will perform much better with low latency ram...and as the fsb of this low latency ram increases we will need to push higher latency ram to much higher speeds to match it.
I know many overclockers who are now running old Corsair/OCZ/Kingston BH5 bought off Ebay at some incredible speeds and overclocks professing its the best ram for A64...the only difference is now OCZ are pretty much the only manufacturer to offer an alternative to buying second hand.
If you have both types of ram, run a few benches off and see for yourselves which is faster.
TCCD has its place, 2-2-2 at low voltage and incredible clocks at reasonable voltages, but for all out benching high voltage and tight latency to many is the king.
"Wesley, you're claiming that I will get a 10% increase in FPS by *just* switching memories"
Yeah, but that's only at 1024 x 768 (and Quake 3?). Anyone buying this sort of memory would most likely be gaming at a much higher resolution. Is there much gain at all at 1600 x 1200? A video encoding benchmark would have been nice...
Wesley, you're claiming that I will get a 10% increase in FPS by *just* switching memories, and you dont want to go ahead and do a more thorough review? I find the results hard to believe, and it doesnt help that OCZ runs so many ads on here.
I also find it hard to believe that you couldnt get a DDR booster. As far as I can tell, OCZ was the one to send you the memory, so they should also have the boosters for the test. If OCZ doesnt have DDR boosters, who does?
I dont know a better way of saying this, but I'd prefer if the advertisements stayed in the boxes at the top of the screen, and out of the content of the pages.
#49 & #50 - I have updated the test configuration to include all components used in the series of memory tests. Comments have also been added to the configuration page clearly stating the configuration used and the rationale for our test methods.
Ideally all memory would have been retested on the MSI or the VX would have been tested in the MSI using a DDR Booster. However, a Booster was not available, and the time for retesting on the DFI did not seem justified based on the close nF4/nf3 performance we have seen in past benchmarking.
Wesley, you yourself demonstrated in the SLI roundup that nf4 boards have an ideal tras at 6-8 which is clearly lower than the ideal tras of older athlon 64 boards. There is plenty of reason to think using the DFI could skew the results. Why not just put it to rest once and for all and compare results of the 2 boards keeping all other things equal? Well except the whole PCIe and AGP thing. But everything else.
Further, why not just test *one* of those other sticks of RAM on this board and with these drivers? Under the section "Performance Test Configuration" we have listed: 8 kinds of RAM, Forceware 71.80, DFI LANParty nF4. This is NOT representative of the actual numbers to come later in the review, where again, all 8 kinds of RAM are presented on the same graph, implying that the differences we se here are solely because of the RAM, and all other factors have been held constant.
One thing that hasn't even been mentioned is that you ran this OCZ VX at 6 tRAS while the other numbers come from 10 tRAS. Who knows how that might have changed things on an nF4 board. I'm not saying that that changed things, but it *could* have, and we would never know from this review because this review doesn't present the results in a way that is valid.
Please, for the sake of AT's benchmarking integrity, take JUST the top-performing RAM (the Corsair?) from the earlier test. Put it through the wringer with the DFI and the newer drivers. Then re-do the charts in this review, comparing JUST the OCZ VX and the Corsair 4400C25. Then you can say "Look at this other review to get an idea of the comparison of OCZ VX vs. Geil, G.Skill, Crucial, etc. We won't compare directly, but OCZ VX crushed Corsair, which beat all the others."
We all know that your conclusion is right. We know that this OCZ VX RAM really is performing the way you claim -- way better than anything else available. The problem is with the essentially dishonest way that you're presenting those results, even if the results themselves are true. I hope that you can understand this sentiment that makes us hold on to this most important principle of benchmarking or any kind of scientific testing: hold everything else constant!
Wesley, it's disturbing that you refuse to re-test other memories on the same test set-up that you used for this memory, or make some other effort to do a fair test set-up.
Anandtech really shouldnt be making any assumptions. I dont want to bash... but... how can anyone take these tests seriously? How do we know that OCZ didnt *ask* that this memory to be tested on the DFI board for boosted results? I dont want to say such things, but you're leaving yourself open to these questions.
#38, #40, #41 -
In the review we never stated that the other memory chips were tested on the DFI. We pulled the results from our earlier benchmark, since we have already established that the nF3 and nF4 perform virtually the same, and the AGP and PCIe perform virtually the same. In fact, if you check closely, the DFI was a very average performer at stock speeds, so the DFI is not the reason for higher speeeds.
As much as it goes against the grain of many peoples thinking, 2-2-2 on one memory has never performed the same as 2-2-2 on another chip. There are performance differences that can only be explained by difference in the memory chips.
We would have tested on the MSI Neo2 had it supported the voltages needed by VX, but it can't supply them. We do not have a DDR Booster at present so we could not test on another motherboard, so we tested on the only production motherboard to supply voltages needed by the VX.
The 71.80 drivers ARE a bit faster than the 61.77 used for some earlier tests, but the difference is still small and does not change the performance pattern seen in this review. I have posted those benchmarks earlier in these comments.
I know most BH5 or BH6 will do 250fsb 2-2-2- timing as well with like 3.3Vs. Wesley, how does the BH5 at 250 2-2-2 compare with VX at 2-2-2 250? is bh5 slower or just as fast?? (since u mention "If we had results from older BH5 chips you would likely have seen BH5 perform between Samsung TCCD and OCZ VX", would that be at ddr400? or ddr500? Thankq for the great review, btw when are we goin to see the review for the 24'' and 20'' widescreen dell lcd??
this smells like winbond, especially the settings where it settled best around tras 5 or 6. that is unique to BH5/6. also the voltage is unique to BH5/6 only.
#35 - That's not what #34 was asking. It's no surprise that sticks of RAM using the same chips will perform similarly.
What IS a surprise is that the PQI 3200 Turbo gave 512.9 FPS in Quake 3 Arena in a DFI nF4 motherboard with nVidia 71.80 drivers, and it also happened to get 512.9 FPS in Quake 3 Arena in an MSI nF3 motherboard with nVidia 61.77 drivers (as shown in your January 4th review of the Corsair PC4400). This exact sameness in benchmark numbers is the same down the line, with every type of RAM and every benchmark -- both gaming and synthetic. I checked every single number.
There's no need to re-test all of the 7 types of RAM in the DFI board with the newer drivers. Just a couple, say the Crucial and the Geil, so we can know that the conclusions are valid.
"With nForce3 motherboards, we achieved the fastest performance on AMD Athlon 64 chipsets (nForce3, VIA K8T800 PRO) at Cycle Time or tRAS of 10."
Are you saying there are nForce3 motherboards out there with VIA chipsets? ;-) I know it is probably a typo, nevermind.
However, i thought the tras 10- thing was a odd behaviour of the nforce2/3 chipsets only, not the VIA K8T800 PRO too.
Otherwise, good review, as always. These chips would be something to look for here in Norway, considering the fact that vx3200- ram is no more expensive than TCCD- chipped ram.
#35-
wes, you're right there, but it doesn't help explain the profound performance differencce seen in the gaming tests and unbuffered bandwidth... it all adds up to the dfi providing better memory/graphics performance in my mind.
Some of the figures show over 10% performance increase (the actual game benchmarks, not just memory benches) - is this performance intirely attributable to the RAM of odes the motherboard\gfx come into play at all?
Previously high-end RAM with tight timings has shown an improvement of a couple of fps - say max 2% overall, yet this is showing figures of 10% improvement - which like I say is like going from a 3500+ to a 4000+ - and I'm pretty certain that the DFI board, this ram and a 3500+ would be cheaper than a 4000+, generic ram and a cheaper mobo.
Wesley - can I suggest a follow up, with this RAM on other boards and a direct comparison of RAM on the same setup, because I think the benchmarks you have shown certainly warrant this being done. You cant blame me for being sceptical when the figures are so unbelievably awesome!
#34 - The other RAM did NOT perform identically, it is just that there was much smaller variation in the results than we saw with the leap in VX performance.
Perhaps it will be easier to swallow if I point out that the OCZ 3200 Platinum R2, PQI 3200 Turbo, G. Skill TCCD, Geil PC3200 Ultra X, and Corsair TwinX1024-4400C25 are all based on the same Samsung TCCD memory chips. The Crucial Ballistix is based on Micron chips, and the OCZ 3700 Gold R3 is based on Hynix DT-D5 chips. Until VX, all recent memory has had to compete with Samsung TCCD, which quickly became the performance standard for current memory. If we had results from older BH5 chips you would likely have seen BH5 perform between Samsung TCCD and OCZ VX.
I find some of these results rather difficult to believe, not so much the results of the OCZ DRAM but I'm confused as to how the other DRAM in this test managed to perform identically in every test on the MSI K8N Neo2 that was used in your last DRAM review and the DFI motherboard utilized in this review.
It would seem rather unlikely that every piece of DRAM would perform exactly the same in every test on two different motherboards.
I have the 3200vx and they perform very similarly(i think they are the same ram but just different auto settings lol) I will post screenie and stuff on forums when i can run some concrete bench marks.
Wesley: I'm confused by your answers to #15/#17. Did you actually test all 8 types of RAM on the DFI LANParty nF4 SLI-DR with nVidia 71.80 drivers, or did you take the benchmark results for the other 7 sticks of RAM from the previous review of the Corsair 4400C25, which used an MSI nF3 motherboard and the 61.77 drivers? (All the numbers seem to be the same, which I find hard to believe with a new motherboard, chipset, and video drivers. I would expect at least a couple of variations of 0.1 fps or something.)
These are Winbond chips. The BH/CH-5/6 chips were their own branded chips, while these are unlabelled OEM chips that are sold for relabelling. The supply of the labelled Winbond chips are gone, but there are still plenty of these unlabelled chips.
This really gives you an idea of the cartel nature of the memory industry. Small memory makers like Winbond, despite having a superior product, cannot really stay alive in this market of giants.
thanks for the info wes. if it's not too much trouble, how about a test with the VX on the neo2 with a ddr booster? you say that it's just a better performing ram chip, i'd like to believe you, but it really is difficult to swallow.
and.. the fact that this ram is so much faster really is piquing my interest, and shows that even if you don't have extreme voltage, this ram may be nice to have for the higher performance at the same speed/timings.
if it doesn't take too much time, i, and i'm sure a few other people, would be very apprecitative of some benches of vx on the neo2. thanks wes.
#26 - You need to take a close look at DDR400 2-2-2 VX vs. DDR400 2-2-2 with any other ram we have tested. The biggest news with VX is how fast the chips are at the same speed and timings compared to other chips, which we talked about in the article. There IS a difference in performance from DDR400 2-2-2 to DDR533 2-2-2 at the same clock speed but it is smaller than many imagine as I pointed out in the review. We have set up our AMD tests to really measure the impact of memory - removing as many other variables as possible.
Sorry but I don't see any justification for over-voltage RAM as the system performance increase is marginal at best. The cost doesn't justify the minimal gains. In addition more voltage = more heat, any way you slice it. This is more marketing hype and no measurable system performance increase as we've seen before from OCZ. Between OCZ and "DFI's gullible PC enthusiast product line", I'm sure they are laughing all the way to the bank. PT Barnum was correct...
Obviously the RTCW-ET results with VX at 2-3-2 are 116.7 with 61.77 vs. 119.3 with 71.80. This compares to our previous fastest 2-2-2 DDR400 at 110.8. We really do need the ability to correct in the comments section :-)
#15 - I did test with 61.77 to make sure the results were not too skewed by the newer 71.80 driver. Q3 and RTCW are slower with 61.77 but this is offset by the memroy bandwidth scores being higher. I just quickly ran 61.77 at 2-3-2-6 and 2,.6v since that was your greatest interest. Q3-554.1 (vs. 567.4), RTCW-ET-116.7 (vs.199.3), Sandra UNbuffered-2949 (vs.2927). The 61.77 results at 2-3-2 are still significantly higher than any 2-2-2 results at 400.
#17 - In our launch review we clearly showed the MSI Neo2 to be the same performance as nF4. We have also shown in the past that AGP and PCIe using the same card are the same performance. Since the MSI does not support these extreme voltages we had to make some changes to test VX - which we detailed in the review.
I realize many of you do not want to believe that a memory running 2-3-2 can outperform another memory running 2-2-2 at the same speed, but this is not unique to VX. It is just that VX is an extreme example of a chip being faster at the same speed. Go back to benchmarks that included BH5 and you will see it is faster than TCCD at the same speed and timings.
#12 - The memory ran quite cool at 3.0V, and was even cool to about 3.2V. Above that, however, it started to get quite warm and I did mount a fan over the dimms at speeds above DDR500 to get higher stable overclocks. The VX ran fine at higher speeds, stable and no crashes, but the extra cooling gave a few FSB more in overclocking.
#13 - The Value Ram roundup is in the works and will hopefully publish next week while I am out of the US.
umm, i have a problem with the video card benchmarks..
go back to the 61.77 drivers you used for the rest of the benchmarks, vx at 2-3-2 shouldn't be that much faster than other ram at 2-2-2 at the same mhz.
#10 - I don't know how the wording got turned around, but the sentence has been corrected. It now reads:
"As we raise the memory speed from 200 to 267 (DDR400 to DDR533), keeping the CPU speed constant, memory Read increases over 25% while memory Write over the same range shows just a 14% increase. That means that while all operations benefit from memory speed increases, operations more dependent on memory Read will benefit much more from memory speed boosts than those that are memory Write dependent."
Nice review, and great memory. But for those of us who aren't willing to piss away that much money for memory, are you still planning on the value memory round-up that was promised last summer?
how were the temps runnin that stuff at 3.6v? was it super hot? were case temps drastically higher? did you need active fan cooling over the ram? open test bed? was it all good in the hood?
"That means that while all operations benefit from memory speed increases, operations more dependent on memory write will benefit much more from memory speed boosts than those that are memory write dependent."
Should be
"That means that while all operations benefit from memory speed increases, operations more dependent on memory write will benefit much more from memory speed boosts than those that are memory read dependent."
I'm just wondering, though..... can there possibly be a (real-life, practical) application that writes to memory more than it reads from memory? I mean, what's the point of writing to memory, if the stored values are never accessed? Seems like a pretty inefficient program to me :)
Good article, I agree this is one of the few non-boring RAM reviews I've ever seen :)
normally, I hate reading memory reviews. Looking at the charts, I don't ever get excited about advances in the stuff, simply because you never see all that much of an improvement on the current king of performance.
However, I must admit I was enjoying the article much more than I have ever have before. I guess when you see 10's of fps better, and there is such a marked improvement in performance over the competition, you can't help but want some of that action. ;)
now, if we can just see that kind of performance for the sub-$100 market ;)
The figures are correct, everyone has become fixated with DDR600+ using TCCD and higher latency, this proves tight latency is the king on A64 by a mile as long as the clock is high enough.
The price is not that expensive, there are 3 versions of this ram. vlaue VX, 3200VX and 4000VX, all of it clocks well with voltage although for the guaranteed highest clocks i would go with the 4000. A note to #3, the value VX is i hear cheaper than twinmos memory, with the launch of 4000VX the 3200 price has dropped also, please remember OCZ are the ONLY company to warrant high voltage here and for peace of mind the extra few $$ you "may" have to spend would be well worth it in my opinion.
when you mention the price, if the figures given are correct this ram seems to give across the board over 10% of performance increase; thats probably more than say 3500+ to 4000+ if im not mistaken. Im sure the cost of a 3500+ and a pair of this is cheaper than a 4000+ and a pair of other ram. I spose the voltage is a limitation, but if the performances increases are that great I would foresee a lot more enthusiast boards offering Vdimm upto say 3.3V or more.
I still can't really believe the figures though; can Wesley confirm if the other ram was tested on the same DFI board (or is the DFI also contributing to the performance increase?).
the kinds of chips that these ocz's using -- they have to be sourced from somewhere and I doubt they have exclusive acess.... or is it that only they are crazy enough to produce memory running at 3.6v and still give a lifetime warranty??
the competition should catch on soon and hopefully the price wont be so horrendously expensive :)
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
67 Comments
Back to Article
Lymphatik - Wednesday, March 30, 2005 - link
Hi I would like to know if u think that thoose dimms could be enough cooled by a 120 mm fan in extraction and a 120 mm fan from a power suplyBecause i would like to put thoose dimm in antec Sonata and as it's a small case i don't know if i will have enough space to put a fan
Lymphatik - Wednesday, March 30, 2005 - link
OCedHrt - Thursday, March 24, 2005 - link
Figured it out, seems like the cpu speed plays a role here. 2.25 ghz maxes out at 6600-6700. 2.34 gets 6900-7000. I wonder if going even higher will set a new bandwidth record.OCedHrt - Tuesday, March 22, 2005 - link
Hmm, I can't seem to duplicate the results here and I don't know what I'm doing wrong. I have 2 sticks of pc3200 ocz vx and a dfi ultra (non-sli). I have the ram running at ddr500 at 2-2-6-2 T1 @3.3v and sandra reports a bandwidth of 6000-6100. The weird part is that sandra reports a timing of 2T but if I drop the timing from 1T to 2T in the bios, the results drop to 5700 so I'm pretty sure 1T is enabled. Any ideas?zumbi9in - Saturday, March 19, 2005 - link
I would like you to help me with a MotherBoard/RAM question.I’m about to buy a DFI Lanparty NFORCE 4 Ultra MB with 2 GB of RAM.
This machine will be for audio sampling and recording, and some gaming too J
I’m worried if it can run 2 GB of RAM at DDR400, I don’t know if it will drop the memory speed with this config.
I would like to buy 4x512 KINGSTON PC3200 ULTRA LL 2-2-2-5, but as they are double sided, maybe they will run at DDR333
The 2nd option would be 2x1024 of OCZ memory PC 3200 3-3-3-7 OCZ4001024PF, is its performance good with this latency?
I have searched lots of forums and didn’t find a final word about this issue.
What would be the best performance choice? Do you think it can work?
Thank you very much,
Alexandre Zumbi.
Rand - Wednesday, March 9, 2005 - link
ozzimark - Wednesday, March 9, 2005 - link
#60-iirc, the dfi was tuned for tccd and bh-5/vx.
other stuff that overclocks somewhat does terribly on the dfi, that includes ballistix's micron 5b-g chips.
L3p3rM355i4h - Wednesday, March 9, 2005 - link
hmm...good performance, but also a lot of questions. How well does ballistix do on the mobo?Quanticles - Wednesday, March 9, 2005 - link
For everyone hung up on the on-board memory controller...The motherboard will make a difference for the on-board memory controller, not because of the chipset, but due to the characteristics of the routing. The type of material used, the thickness and width of the copper traces, and how well the traces are matched in length, will all make a huge difference in signal integrity. Even differences in what layers of the board you route to make a difference, and what power plains they're coupled against are. The list goes on...
DFI put a lot more effort into their routing, and it makes a significant difference.
slashbinslashbash - Tuesday, March 8, 2005 - link
#56 - Wow. Thanks for pointing this out. This review is bogus. I hope that you cc'd Mr. Shimpi on that e-mail.ozzimark - Tuesday, March 8, 2005 - link
#56-it's easier to just go and get a booster and retest on the neo2 that was previously the benchmark system.
renzokuken - Tuesday, March 8, 2005 - link
With most comments before this one praising this RAM as if it were the Holy Messiah Himself, my comment will surely be seen as the Devil but the truth is, this article is biased. Mr. Fink doesnt know what he's on about. He's gone and tested the VX Gold and pitted the test results against the results obtained from other RAM on a completely different board. If you would take the time to indulge me, you will see that he's simply copied and pasted the performance results of RAM taken from the MSI Neo2 (NF3) and then compared these results to what he obtained with the VX Gold on the DFI LANParty nF4 SLI-DR. I cannot express in any written language how inconsistent this is. If I didn't know better, I'd say OCZ handed him a nice little duffle bag sporting a $ sign after writing this article.The fact is, OF COURSE the OCZ VX Gold will look like God when you compare RAM the way Mr.Fink has done. The test bed he's used for comparison has not been consistent. So you say "Who cares, its an onboard memory controller. It shouldn't matter which board you use". Do some research and you'll see that the previous statement is rediculous. Mr. Fink tested the DFI LANParty nF4 SLI-DR when it was first available and had the following results with the OCZ PC3200 Platinum Rev.2: "Quake 3 ran at 642FPS and SiSoft Sandra 2004 standard memory bandwidth was 8,300 MB/s. The Sandra unbuffered memory bandwidth was at 4000 MB/s" ( http://www.anandtech.com/mb/showdoc.aspx?i=2337&am... ). A direct comparison of these results to when the SAME RAM was tested on the MSI Neo2 ( http://www.anandtech.com/memory/showdoc.aspx?i=231... ) will show that the board itself makes all the difference, in DFI's favor. Considering the Corsair twinx1024-4400c25 outperformed the OCZ on the MSI, I still cant understand why we havent seen this RAM tested on the DFI yet. In any case, you cannot compare the test results of the VX Gold to those obtained by other RAM on the MSI board and thus the results and conclusion of this article is invalid.
I've emailed Mr.Fink regarding the inconsistencies of his articles and have outlined what we want as consumers. Because of the nature of the DFI board, I believe anandtech should test the top 5 RAM modules using this board (including the Corsair twinx1024-4400c25) and present us with results from an unbiased, completely controlled testbed so we, as consumers, can make a more informed decision when deciding which RAM to purchase.
slashbinslashbash - Tuesday, March 8, 2005 - link
#51: Thank you.bigtoe36 - Tuesday, March 8, 2005 - link
Guys, have a look at the front page of the ORB for 3D2001, many of the top benchers use ram thats running 2-2-2 at 250+.You have to remember here the VX beat out OCZ's 3200Plat rev2 which is also TCCD based so Wesleys review while showing one product in a good light shows another in a not so good.
Nvidia have said that the NF4 and NF3 chipsets clock ram and perform equal with the same memory, i have tested both boards here and found the same, if you want proof do a few days of testing yourself, remember the memory controller is on the CPU, not the chipset ;-) so actual bandwidth will probably be identicle.
This debate has been runmbling for years, whether high fsb's high latency beat low latency at lower fsb's, we debated for ages with the 875 chipsets but the cross over for performance was a lot lower down the fsb range, and boards are now clocking ram better than they ever did. Many prefered 250fsb 5:4 with 2-2-2 over 250 1:1 with 2.5-4-3 as the benches were always higher in games running async, you needed 265fsb for the high latency 1:1 to equal the 5:4 score and then pull away.All we see now is a much more flexable memory controller which just like the P4 C Northwoods is low latency, it will perform much better with low latency ram...and as the fsb of this low latency ram increases we will need to push higher latency ram to much higher speeds to match it.
I know many overclockers who are now running old Corsair/OCZ/Kingston BH5 bought off Ebay at some incredible speeds and overclocks professing its the best ram for A64...the only difference is now OCZ are pretty much the only manufacturer to offer an alternative to buying second hand.
If you have both types of ram, run a few benches off and see for yourselves which is faster.
TCCD has its place, 2-2-2 at low voltage and incredible clocks at reasonable voltages, but for all out benching high voltage and tight latency to many is the king.
kmmatney - Tuesday, March 8, 2005 - link
"Wesley, you're claiming that I will get a 10% increase in FPS by *just* switching memories"Yeah, but that's only at 1024 x 768 (and Quake 3?). Anyone buying this sort of memory would most likely be gaming at a much higher resolution. Is there much gain at all at 1600 x 1200? A video encoding benchmark would have been nice...
Quanticles - Monday, March 7, 2005 - link
Wesley, you're claiming that I will get a 10% increase in FPS by *just* switching memories, and you dont want to go ahead and do a more thorough review? I find the results hard to believe, and it doesnt help that OCZ runs so many ads on here.I also find it hard to believe that you couldnt get a DDR booster. As far as I can tell, OCZ was the one to send you the memory, so they should also have the boosters for the test. If OCZ doesnt have DDR boosters, who does?
I dont know a better way of saying this, but I'd prefer if the advertisements stayed in the boxes at the top of the screen, and out of the content of the pages.
Wesley Fink - Monday, March 7, 2005 - link
#49 & #50 - I have updated the test configuration to include all components used in the series of memory tests. Comments have also been added to the configuration page clearly stating the configuration used and the rationale for our test methods.Ideally all memory would have been retested on the MSI or the VX would have been tested in the MSI using a DDR Booster. However, a Booster was not available, and the time for retesting on the DFI did not seem justified based on the close nF4/nf3 performance we have seen in past benchmarking.
sangyup81 - Monday, March 7, 2005 - link
Wesley, you yourself demonstrated in the SLI roundup that nf4 boards have an ideal tras at 6-8 which is clearly lower than the ideal tras of older athlon 64 boards. There is plenty of reason to think using the DFI could skew the results. Why not just put it to rest once and for all and compare results of the 2 boards keeping all other things equal? Well except the whole PCIe and AGP thing. But everything else.slashbinslashbash - Monday, March 7, 2005 - link
#46 -- BTW, "In the review we never stated that the other memory chips were tested on the DFI."That's EXACTLY what "Performance Test Configuration" says. Or are we to assume that every other AT benchmark test is similarly flawed?
slashbinslashbash - Monday, March 7, 2005 - link
Further, why not just test *one* of those other sticks of RAM on this board and with these drivers? Under the section "Performance Test Configuration" we have listed: 8 kinds of RAM, Forceware 71.80, DFI LANParty nF4. This is NOT representative of the actual numbers to come later in the review, where again, all 8 kinds of RAM are presented on the same graph, implying that the differences we se here are solely because of the RAM, and all other factors have been held constant.One thing that hasn't even been mentioned is that you ran this OCZ VX at 6 tRAS while the other numbers come from 10 tRAS. Who knows how that might have changed things on an nF4 board. I'm not saying that that changed things, but it *could* have, and we would never know from this review because this review doesn't present the results in a way that is valid.
Please, for the sake of AT's benchmarking integrity, take JUST the top-performing RAM (the Corsair?) from the earlier test. Put it through the wringer with the DFI and the newer drivers. Then re-do the charts in this review, comparing JUST the OCZ VX and the Corsair 4400C25. Then you can say "Look at this other review to get an idea of the comparison of OCZ VX vs. Geil, G.Skill, Crucial, etc. We won't compare directly, but OCZ VX crushed Corsair, which beat all the others."
We all know that your conclusion is right. We know that this OCZ VX RAM really is performing the way you claim -- way better than anything else available. The problem is with the essentially dishonest way that you're presenting those results, even if the results themselves are true. I hope that you can understand this sentiment that makes us hold on to this most important principle of benchmarking or any kind of scientific testing: hold everything else constant!
Quanticles - Sunday, March 6, 2005 - link
Wesley, it's disturbing that you refuse to re-test other memories on the same test set-up that you used for this memory, or make some other effort to do a fair test set-up.Anandtech really shouldnt be making any assumptions. I dont want to bash... but... how can anyone take these tests seriously? How do we know that OCZ didnt *ask* that this memory to be tested on the DFI board for boosted results? I dont want to say such things, but you're leaving yourself open to these questions.
Wesley Fink - Sunday, March 6, 2005 - link
#38, #40, #41 -In the review we never stated that the other memory chips were tested on the DFI. We pulled the results from our earlier benchmark, since we have already established that the nF3 and nF4 perform virtually the same, and the AGP and PCIe perform virtually the same. In fact, if you check closely, the DFI was a very average performer at stock speeds, so the DFI is not the reason for higher speeeds.
As much as it goes against the grain of many peoples thinking, 2-2-2 on one memory has never performed the same as 2-2-2 on another chip. There are performance differences that can only be explained by difference in the memory chips.
We would have tested on the MSI Neo2 had it supported the voltages needed by VX, but it can't supply them. We do not have a DDR Booster at present so we could not test on another motherboard, so we tested on the only production motherboard to supply voltages needed by the VX.
The 71.80 drivers ARE a bit faster than the 61.77 used for some earlier tests, but the difference is still small and does not change the performance pattern seen in this review. I have posted those benchmarks earlier in these comments.
bigtoe36 - Sunday, March 6, 2005 - link
#44BH5 is about the same speed as VX is
JoKeRr - Sunday, March 6, 2005 - link
I know most BH5 or BH6 will do 250fsb 2-2-2- timing as well with like 3.3Vs. Wesley, how does the BH5 at 250 2-2-2 compare with VX at 2-2-2 250? is bh5 slower or just as fast?? (since u mention "If we had results from older BH5 chips you would likely have seen BH5 perform between Samsung TCCD and OCZ VX", would that be at ddr400? or ddr500? Thankq for the great review, btw when are we goin to see the review for the 24'' and 20'' widescreen dell lcd??JoKeRr - Sunday, March 6, 2005 - link
Rand - Saturday, March 5, 2005 - link
Wesley, I understand the DIMM's are using the same chip and hence unsurprisingly they perform similarly.What I am finding hard to believe is that all of the DIMM's perform identically on two different motherboards, using different graphics drivers.
In every single test the bandwidth never deviates by even 1MB/s or so much as 0.1 FPS.
I would imagine the odds of two different platforms never devaiting in anything by even the sammest margin is bordering on non-existent.
cryptonomicon - Saturday, March 5, 2005 - link
this smells like winbond, especially the settings where it settled best around tras 5 or 6. that is unique to BH5/6. also the voltage is unique to BH5/6 only.a winbond chip if ive ever seen one.
slashbinslashbash - Saturday, March 5, 2005 - link
#35 - That's not what #34 was asking. It's no surprise that sticks of RAM using the same chips will perform similarly.What IS a surprise is that the PQI 3200 Turbo gave 512.9 FPS in Quake 3 Arena in a DFI nF4 motherboard with nVidia 71.80 drivers, and it also happened to get 512.9 FPS in Quake 3 Arena in an MSI nF3 motherboard with nVidia 61.77 drivers (as shown in your January 4th review of the Corsair PC4400). This exact sameness in benchmark numbers is the same down the line, with every type of RAM and every benchmark -- both gaming and synthetic. I checked every single number.
There's no need to re-test all of the 7 types of RAM in the DFI board with the newer drivers. Just a couple, say the Crucial and the Geil, so we can know that the conclusions are valid.
frodin - Saturday, March 5, 2005 - link
"With nForce3 motherboards, we achieved the fastest performance on AMD Athlon 64 chipsets (nForce3, VIA K8T800 PRO) at Cycle Time or tRAS of 10."Are you saying there are nForce3 motherboards out there with VIA chipsets? ;-) I know it is probably a typo, nevermind.
However, i thought the tras 10- thing was a odd behaviour of the nforce2/3 chipsets only, not the VIA K8T800 PRO too.
Otherwise, good review, as always. These chips would be something to look for here in Norway, considering the fact that vx3200- ram is no more expensive than TCCD- chipped ram.
ozzimark - Saturday, March 5, 2005 - link
#35-wes, you're right there, but it doesn't help explain the profound performance differencce seen in the gaming tests and unbuffered bandwidth... it all adds up to the dfi providing better memory/graphics performance in my mind.
theOracle - Saturday, March 5, 2005 - link
sorry, of odes = or does!intirely = entirely
Any ideas when this RAM is available\where from?
theOracle - Saturday, March 5, 2005 - link
Some of the figures show over 10% performance increase (the actual game benchmarks, not just memory benches) - is this performance intirely attributable to the RAM of odes the motherboard\gfx come into play at all?Previously high-end RAM with tight timings has shown an improvement of a couple of fps - say max 2% overall, yet this is showing figures of 10% improvement - which like I say is like going from a 3500+ to a 4000+ - and I'm pretty certain that the DFI board, this ram and a 3500+ would be cheaper than a 4000+, generic ram and a cheaper mobo.
Wesley - can I suggest a follow up, with this RAM on other boards and a direct comparison of RAM on the same setup, because I think the benchmarks you have shown certainly warrant this being done. You cant blame me for being sceptical when the figures are so unbelievably awesome!
Wesley Fink - Saturday, March 5, 2005 - link
#34 - The other RAM did NOT perform identically, it is just that there was much smaller variation in the results than we saw with the leap in VX performance.Perhaps it will be easier to swallow if I point out that the OCZ 3200 Platinum R2, PQI 3200 Turbo, G. Skill TCCD, Geil PC3200 Ultra X, and Corsair TwinX1024-4400C25 are all based on the same Samsung TCCD memory chips. The Crucial Ballistix is based on Micron chips, and the OCZ 3700 Gold R3 is based on Hynix DT-D5 chips. Until VX, all recent memory has had to compete with Samsung TCCD, which quickly became the performance standard for current memory. If we had results from older BH5 chips you would likely have seen BH5 perform between Samsung TCCD and OCZ VX.
Rand - Saturday, March 5, 2005 - link
I find some of these results rather difficult to believe, not so much the results of the OCZ DRAM but I'm confused as to how the other DRAM in this test managed to perform identically in every test on the MSI K8N Neo2 that was used in your last DRAM review and the DFI motherboard utilized in this review.It would seem rather unlikely that every piece of DRAM would perform exactly the same in every test on two different motherboards.
Forsa - Saturday, March 5, 2005 - link
I have the 3200vx and they perform very similarly(i think they are the same ram but just different auto settings lol) I will post screenie and stuff on forums when i can run some concrete bench marks.Live - Friday, March 4, 2005 - link
These are winbond chips that can also be found in twinmos. Dont know if this is mentiond in the article as I did not read all :)Check out this link for info on speeds and which ones to get:
http://www.akiba-pc.com/article.php?45.0
slashbinslashbash - Friday, March 4, 2005 - link
Wesley: I'm confused by your answers to #15/#17. Did you actually test all 8 types of RAM on the DFI LANParty nF4 SLI-DR with nVidia 71.80 drivers, or did you take the benchmark results for the other 7 sticks of RAM from the previous review of the Corsair 4400C25, which used an MSI nF3 motherboard and the 61.77 drivers? (All the numbers seem to be the same, which I find hard to believe with a new motherboard, chipset, and video drivers. I would expect at least a couple of variations of 0.1 fps or something.)Jeff7181 - Friday, March 4, 2005 - link
Awesome... I always wondered about Tras and the Athlon-64... this answers my question! :)Slaimus - Friday, March 4, 2005 - link
These are Winbond chips. The BH/CH-5/6 chips were their own branded chips, while these are unlabelled OEM chips that are sold for relabelling. The supply of the labelled Winbond chips are gone, but there are still plenty of these unlabelled chips.This really gives you an idea of the cartel nature of the memory industry. Small memory makers like Winbond, despite having a superior product, cannot really stay alive in this market of giants.
ozzimark - Friday, March 4, 2005 - link
thanks for the info wes. if it's not too much trouble, how about a test with the VX on the neo2 with a ddr booster? you say that it's just a better performing ram chip, i'd like to believe you, but it really is difficult to swallow.and.. the fact that this ram is so much faster really is piquing my interest, and shows that even if you don't have extreme voltage, this ram may be nice to have for the higher performance at the same speed/timings.
if it doesn't take too much time, i, and i'm sure a few other people, would be very apprecitative of some benches of vx on the neo2. thanks wes.
Wesley Fink - Friday, March 4, 2005 - link
#26 - You need to take a close look at DDR400 2-2-2 VX vs. DDR400 2-2-2 with any other ram we have tested. The biggest news with VX is how fast the chips are at the same speed and timings compared to other chips, which we talked about in the article. There IS a difference in performance from DDR400 2-2-2 to DDR533 2-2-2 at the same clock speed but it is smaller than many imagine as I pointed out in the review. We have set up our AMD tests to really measure the impact of memory - removing as many other variables as possible.Beenthere - Friday, March 4, 2005 - link
Sorry but I don't see any justification for over-voltage RAM as the system performance increase is marginal at best. The cost doesn't justify the minimal gains. In addition more voltage = more heat, any way you slice it. This is more marketing hype and no measurable system performance increase as we've seen before from OCZ. Between OCZ and "DFI's gullible PC enthusiast product line", I'm sure they are laughing all the way to the bank. PT Barnum was correct...ChiefNutz - Friday, March 4, 2005 - link
The Article said that fourm rumors said the chips were built by winbond. I Though winbond left the memory market? anyone know any different??Wesley Fink - Friday, March 4, 2005 - link
Obviously the RTCW-ET results with VX at 2-3-2 are 116.7 with 61.77 vs. 119.3 with 71.80. This compares to our previous fastest 2-2-2 DDR400 at 110.8. We really do need the ability to correct in the comments section :-)Wesley Fink - Friday, March 4, 2005 - link
#15 - I did test with 61.77 to make sure the results were not too skewed by the newer 71.80 driver. Q3 and RTCW are slower with 61.77 but this is offset by the memroy bandwidth scores being higher. I just quickly ran 61.77 at 2-3-2-6 and 2,.6v since that was your greatest interest. Q3-554.1 (vs. 567.4), RTCW-ET-116.7 (vs.199.3), Sandra UNbuffered-2949 (vs.2927). The 61.77 results at 2-3-2 are still significantly higher than any 2-2-2 results at 400.#17 - In our launch review we clearly showed the MSI Neo2 to be the same performance as nF4. We have also shown in the past that AGP and PCIe using the same card are the same performance. Since the MSI does not support these extreme voltages we had to make some changes to test VX - which we detailed in the review.
I realize many of you do not want to believe that a memory running 2-3-2 can outperform another memory running 2-2-2 at the same speed, but this is not unique to VX. It is just that VX is an extreme example of a chip being faster at the same speed. Go back to benchmarks that included BH5 and you will see it is faster than TCCD at the same speed and timings.
LX - Friday, March 4, 2005 - link
Another typo: "OCZ LANParty nForce4"Hey, Anand, let Wesley have some sleep between reviews!
chr6 - Friday, March 4, 2005 - link
typo on last page, its enhanced bandwidth not extended, if i remember correctly.chr6 - Friday, March 4, 2005 - link
EddNog - Friday, March 4, 2005 - link
"Pwned."No, really...
"Pwned."
Rocket321 - Friday, March 4, 2005 - link
Now that you wont need all that extra ocz platinum r2, feel free to send some my way ;)ozzimark - Friday, March 4, 2005 - link
oh wait, i forgot this is also on a different motherboard.. are the nf4 pci-e boards really that much faster?Wesley Fink - Friday, March 4, 2005 - link
#12 - The memory ran quite cool at 3.0V, and was even cool to about 3.2V. Above that, however, it started to get quite warm and I did mount a fan over the dimms at speeds above DDR500 to get higher stable overclocks. The VX ran fine at higher speeds, stable and no crashes, but the extra cooling gave a few FSB more in overclocking.#13 - The Value Ram roundup is in the works and will hopefully publish next week while I am out of the US.
ozzimark - Friday, March 4, 2005 - link
umm, i have a problem with the video card benchmarks..go back to the 61.77 drivers you used for the rest of the benchmarks, vx at 2-3-2 shouldn't be that much faster than other ram at 2-2-2 at the same mhz.
Wesley Fink - Friday, March 4, 2005 - link
#10 - I don't know how the wording got turned around, but the sentence has been corrected. It now reads:"As we raise the memory speed from 200 to 267 (DDR400 to DDR533), keeping the CPU speed constant, memory Read increases over 25% while memory Write over the same range shows just a 14% increase. That means that while all operations benefit from memory speed increases, operations more dependent on memory Read will benefit much more from memory speed boosts than those that are memory Write dependent."
eetnoyer - Friday, March 4, 2005 - link
Nice review, and great memory. But for those of us who aren't willing to piss away that much money for memory, are you still planning on the value memory round-up that was promised last summer?elrolio - Friday, March 4, 2005 - link
my question is:how were the temps runnin that stuff at 3.6v? was it super hot? were case temps drastically higher? did you need active fan cooling over the ram? open test bed? was it all good in the hood?
thanks, just wondering...
AnnihilatorX - Friday, March 4, 2005 - link
This ram rocks solid oO255 USD according to haelduksf
That's a bargain.
Again I am not in US and sometimes I just get depressed when I cannot find a single computer equipment as cheap as US in UK and HK.
slashbinslashbash - Friday, March 4, 2005 - link
Typo on Page 4 (last sentence):"That means that while all operations benefit from memory speed increases, operations more dependent on memory write will benefit much more from memory speed boosts than those that are memory write dependent."
Should be
"That means that while all operations benefit from memory speed increases, operations more dependent on memory write will benefit much more from memory speed boosts than those that are memory read dependent."
I'm just wondering, though..... can there possibly be a (real-life, practical) application that writes to memory more than it reads from memory? I mean, what's the point of writing to memory, if the stored values are never accessed? Seems like a pretty inefficient program to me :)
Good article, I agree this is one of the few non-boring RAM reviews I've ever seen :)
Tiamat - Friday, March 4, 2005 - link
Wow, the huge performance delta is incredible! Just WOWQuiksel - Friday, March 4, 2005 - link
normally, I hate reading memory reviews. Looking at the charts, I don't ever get excited about advances in the stuff, simply because you never see all that much of an improvement on the current king of performance.However, I must admit I was enjoying the article much more than I have ever have before. I guess when you see 10's of fps better, and there is such a marked improvement in performance over the competition, you can't help but want some of that action. ;)
now, if we can just see that kind of performance for the sub-$100 market ;)
cHodAXUK - Friday, March 4, 2005 - link
Stunning performance and great review. Nice work Mr Fink :)bigtoe36 - Friday, March 4, 2005 - link
The figures are correct, everyone has become fixated with DDR600+ using TCCD and higher latency, this proves tight latency is the king on A64 by a mile as long as the clock is high enough.The price is not that expensive, there are 3 versions of this ram. vlaue VX, 3200VX and 4000VX, all of it clocks well with voltage although for the guaranteed highest clocks i would go with the 4000. A note to #3, the value VX is i hear cheaper than twinmos memory, with the launch of 4000VX the 3200 price has dropped also, please remember OCZ are the ONLY company to warrant high voltage here and for peace of mind the extra few $$ you "may" have to spend would be well worth it in my opinion.
haelduksf - Friday, March 4, 2005 - link
Hey- the price isn't bad- I picked a gig of this up for $315 after cancelling my order for the Corsair 4400C25 @ $345 (All CAN$).And the performance is right on- google this stuff, especially at Xtremesystems.org, and you will see nothing but 260-270mhz @ 2-2-2-1T
And that's for the PC3200 ;)
theOracle - Friday, March 4, 2005 - link
when you mention the price, if the figures given are correct this ram seems to give across the board over 10% of performance increase; thats probably more than say 3500+ to 4000+ if im not mistaken. Im sure the cost of a 3500+ and a pair of this is cheaper than a 4000+ and a pair of other ram. I spose the voltage is a limitation, but if the performances increases are that great I would foresee a lot more enthusiast boards offering Vdimm upto say 3.3V or more.I still can't really believe the figures though; can Wesley confirm if the other ram was tested on the same DFI board (or is the DFI also contributing to the performance increase?).
xsilver - Friday, March 4, 2005 - link
the kinds of chips that these ocz's using -- they have to be sourced from somewhere and I doubt they have exclusive acess.... or is it that only they are crazy enough to produce memory running at 3.6v and still give a lifetime warranty??the competition should catch on soon and hopefully the price wont be so horrendously expensive :)
Tokat - Friday, March 4, 2005 - link
theOracle - Friday, March 4, 2005 - link
wow.the figures are so dramatically ahead something just cant be right - what motherboards were the other ram timings on?
then again, if they are right, that is some awesome ram!