TCCD vs. UTT results...have a look!

[infinitevalence]

Brainpower vs. JDEC

http://www.legitreviews.com/article.php?aid=53

 

[Super Nade]

I'm running 2.8V. I'm in touch with a PDP Engineer regarding my problems, they are running a few tests to determine if I've been stuck with a bad set of modules.

/Apologies for the threadjack/

 

[NicePants42]

I just wanted to let you guys know that I purchased 4 512mb sticks of the DDR400TwinMos Speed Premium DDR400 from Newegg, and all 4 sticks appear to be Winbond UTT chips, according to the huge thread over at XS.

Unfortunately, my A7N8X doesn't supply adequate voltage to really test them, so I don't have any benches yet, but everything looks good on ALL sticks:

-P/N ends in AA4T
-Each chip has the two large circular indentations near the center of it
-Each chip's TMD # ends in 44D
-Each chip's 5 digit central # ends in 4
-BrainPower PCB is used

For those of you who haven't made it through all 80 pages of the XS thread (I have read it all), these are all the indicators that the chips are UTT. Also, the German TwinMos rep. seemed to think that the MORE RECENT the sticks are, the better your chances of getting UTT.

I purchased these on 2/15/05 about 3 minutes after Newegg got them back in stock. I obviously can't make any guarantees, but it appears to me that those of you who might want to gamble on UTT have a very decent shot at getting them for $160/gig shipped at the Egg. Just realize that they have been selling out within minutes for the past month or so - I got mine by refreshing like a psychopath every few minutes on the top of the hour on the 15th. I was able to purchase them at either 7 or 8pm (I forgot which) EST.

 

[Reefa_Madness]

I just wanted to let you guys know that I purchased 4 512mb sticks of the DDR400TwinMos Speed Premium DDR400 from Newegg, and all 4 sticks appear to be Winbond UTT chips, according to the huge thread over at XS.

Unfortunately, my A7N8X doesn't supply adequate voltage to really test them, so I don't have any benches yet, but everything looks good on ALL sticks:

-P/N ends in AA4T
-Each chip has the two large circular indentations near the center of it
-Each chip's TMD # ends in 44D
-Each chip's 5 digit central # ends in 4
-BrainPower PCB is used

For those of you who haven't made it through all 80 pages of the XS thread (I have read it all), these are all the indicators that the chips are UTT. Also, the German TwinMos rep. seemed to think that the MORE RECENT the sticks are, the better your chances of getting UTT.

Those are all signs of Winbond UTT,except for the 44D at the end of the TMD code line. That denotes the 4.4ns rating or DDR454. The non-Winbond Speed Premiums have the same TMD line coding. The most important CODES are the "4" vs the "D" in the Part Number (AA4T vs AADT) and in the mfg date code (the five digit central number as you describe it above). The first 2 digits are the year and the next two are the week, with the last digit being either a "4" for Winbond or a "D" for Powerchip ICs (these don't respond to voltage).

As stated above, the dimples, plus the small hole in the lower left hand side of the chip, as well as the two metal staples/bands on both the left and right outer edge of the IC are the typical Winbond physical attributes to look for.

Based on what guys are reporting, the PC3200 dual-pack of non Speed Premiums are also shipping with Winbond UTT.

 

[bobmanfoo]

nice review collin, tighter timings always felt better to me, even when i had my pdps i always ran it at 2-3-3-5, gunna get me another pair of VX when i save up enough for the 90nm FX, decided to keep me neo2 and booster to use with my 3200+ when it gets replaced

 

[RC64]

Hey, think you could run some game benches? like doom3 and hl2 maybe?

 

[NinjaZX6R]

Hey, think you could run some game benches? like doom3 and hl2 maybe?

Sure! I will as soon as I get my new board in the mail. Should come today, then I'll run some more tests!

-Collin-

 

[Overclocker550]

try running your patriot at 2.5-3-3 and lets see how much ahead it comes out. 4's just suck......

 

[NinjaZX6R]

try running your patriot at 2.5-3-3 and lets see how much ahead it comes out. 4's just suck......

Don't you think if it were stable I would?

Anyway, looks like I am delayed for a few days. Waiting for my new NF4 DFI board to come in the mail. These tests will get interesting now that I have a board with 4v going to the ram and easily capable of insane HTT speeds. Will DEFINATELY UPDATE when I get it.

-Collin-

 

[rseven]

I look forward to hearing about happier and more successful exploits. Good luck and please don't tell your girlfriend I was talking about her!

 

[bobmanfoo]

Don't you think if it were stable I would?

Anyway, looks like I am delayed for a few days. Waiting for my new NF4 DFI board to come in the mail. These tests will get interesting now that I have a board with 4v going to the ram and easily capable of insane HTT speeds. Will DEFINATELY UPDATE when I get it.

-Collin-

nice! i love my board, although 3.4v-3.5v is the max you would need set voltage in bios, with ddr booster i had to set to 3.6v to get my current clocks, with the dfi board i only need to set it at 3.4v to get the same clocks, i have read that anything above 3.4v in caused instability but i've ran up to 3.5v no problem, good luck!

 

[TimoneX]

Nice job on the testing bro!

The testing I've done indicates similar performance between low latency and high bandwidth memory. The difference between DDR400@2-2-2-x vs 2.5-3-3-x is about 50Mhz(DDR100) in dual channel w/ 1T timings. There's something else that handicaps TCCD though and that's diminishing returns. Over 260Mhz or so I saw less and less to be gained from increasing memory frequencies. That's not to say there's nothing to be gained, just that the difference between DDR550 & DDR650 isn't as great as the difference between DDR400 & DDR500. I've sold my plat r2s now so I cannot unfortunately run more direct comparisons, but I sold them primarily due to the results of the testing I did run. I believe that if you can give BH5/UTT memory the voltage and get them over DDR500 or so they'll be a match for TCCD up to and beyond DDR600.

 

[Clevor]

In regards to this shootout: 2.5-4-4-7 is pretty slow, and so is 2.5-4-3-7. Heck, on the LEs I tried, I did not even bother to see how they run 2.5-4-3-7. . . I don't go there; I don't care even 300+ at 1T. That is uh, Hynix territory, breached several months ago. Still, 2.5-4-4-7 at 275 is stuff the average overclocker can identify with.

Now if you compared 275, 2.5-3-3-7, that would come out on top over 225, 2-2-2-5, although at 245x11, 2.5-2-2-5 I see it catching or beating TCCD at 270x10, 2.5-3-3-7 (both ram only needing 2.8 volts or so). Every mhz you gain on the BH-5 or Ballistix from 245 up gives you the advantage over TCCD at 2.5-3-3-7, clock-for-clock on the CPU, e.g, 251x11 vs 276x10.

I would do a shootout of 251x11 (2761 mhz), 2-2-2-5 versus 276x10, 2.5-3-3-7 except I have tried three sets of LEs so far, and none will bench 3D at 276x10. I said BENCH 3D, not run Memtest clean - two different things with a high multi. And I've tried 431s, 437s, and 440s, same deal - no go. I also tried two Neo2 Plat boards and two 3500+s that can do 2760, no good. The kickouts and error boxes I get indicate memory issues. 270x10 seems fine though.

Meanwhile, no problem getting all 3D benchmarks in at 251x11 (2761 mhz) with my BH-5 or Ballistix on the 9NDA3+, so the comparison is a moot point, I would say . Moreover, my prelim results indicate the lower latency stuff is indeed faster in the 'Marks.

 

[hitechjb1]

Based on some analysis and 3Dmark01 runs of TCCD from 200 MHz 2-2-2-5 1T to 317 MHz 2.5-4-4-8 1T, I preminlarily established some number for frequency increase needed to break-even with low latency.

Just post it here for another perspective, ....

Originally posted by hitechjb1
...

So preliminary result, between 2-2-2-5 1T (such as BH-5/UTT at 220-250+ MHz) and 2.5-3-3-7 1T (such as TCCD PC4400 at 280-300+ MHz), the latter would require 25% higher frequency to break even with the former low latency memory setup for memory performance.

In conjunction with the 30-42% for memory read of 1 to 8 burst, and the 33% typical based on analytical estimation by counting read access cycles (see link below), it is fair to establish that memory with 2.5-3-3-7 1T would need 25-30% higher bus frequency to break even with memory with 2-2-2-5 1T timing for memory performance in memory intensive applications.

So if BH-5/UTT is able to run at 250 MHz 2-2-2-5 1T, 3.3+ V. TCCD 4400 such as G. Skill LE has to run at around 300 - 310 MHz 2.5-3-3-7 1T, 2.8 V to break even, and in many cases it is doable using some Nforce4 motherboards.

Besides the performance comparison, these are some pros and cons for BH-5/UTT vs TCCD.
- The TCCD modules which require less voltage would lessen concern about chip reliably due to the high 3.3+ V, especially medium to long term impact (if any) of such voltage level on the CPU's memory controller interface (Vmemref).
- The TCCD modules offer a wider range of memory frequency and timing for tweaking, from 200 - 300+ MHz, cas 2/2.5/3 (if motherboard allows).
- On the other hand, the frequency of around 250 MHz for 2-2-2-5 1T memory modules is more easily achievable in many setups for top performance vs the 300+ MHz for 2.5-3-3-7 1T memory.

Memory frequency and latency tradeoff
This gives the analytical reasoning behind the memory frequency and latency tradeoff, and establishing the amount of frequency increase needed.

How much frequency increase is needed to break-even with low latency

 

[Clevor]

"On the other hand, the frequency of around 250 MHz for 2-2-2-5 1T memory modules is more easily achievable in many setups for top performance vs the 300+ MHz for 2.5-3-3-7 1T memory."

From what I've seen of whatever 'Marks I can complete at 276x10, 2.5-3-3-7 versus 251x11, 2-2-2-5, I don't think the spread is quite that big, but it's big enough . From the standpoint of 3DMark anyway.

Yeah, try benching all four 'Marks with 2x512 at 300, 2.5-3-3-7, 1T- no screenies from mem manufacturers please.

It's as I always say, you can't just look at sheer bandwidth. What is the highest 3DMark2000 score I ever got on S939? Maybe 26,973 running 2760 mhz, 30 sec SuperPI 1M, yadda yadda. The highest score I got with a S754 is 27,043 running a Clawhammer at 2580 mhz! Yet I get 7600 buffered bandwidth with S939 and 3500 with S754. So let's not forget the 1MB L2 cache and the short bus path of an on-board memory controller, which make all that bandwidth superfluous.

Guys knock the Intel for poor 3D performance compared to A64. Well in many ways, you are comparing apples to oranges. It would be interesting to compare the effect of L2 cache on 3DMark2000 scores between a Prescott and Northwood, except that you can't run that bench due to some issue with Hyperthreading. I dare say it won't be as dramatic as with A64s.

 

[NinjaZX6R]

Based on some analysis and 3Dmark01 runs of TCCD from 200 MHz 2-2-2-5 1T to 317 MHz 2.5-4-4-8 1T, I preminlarily established some number for frequency increase needed to break-even with low latency.

Just post it here for another perspective, ....




Memory frequency and latency tradeoff
This gives the analytical reasoning behind the memory frequency and latency tradeoff, and establishing the amount of frequency increase needed.

How much frequency increase is needed to break-even with low latency

Man, that is some NICE work there. Thanks for the great info!

-Collin-