Lord of the RAM, Return of the MushkinG

Memory by KeithSuppe @ 2005-05-16

Mushkin has been a supplier of quality PC and Server based memory since the days of SDRAM. Today [M] has been invited along on Mushkin?s latest excursion as they once again prove themselves worthy as the DDR Ring Bearer, and the Lord of the RAM.

  • prev
  • next

1GB Redline HP3200 (2x512MB)

512MB (x2 tested) Redline HP3200


A few words on UTT memory chips

While Mushkin's new Redline series is designed for the ultimate in performance. Based on Winbond UTT memory chips, these modules offer some of the tightest latencies at the highest speeds ever found on DDR. There has been, however; some concern on the quality of these "UnTesTed" memory chips as in this recent article. The article states the following;

"....do you know what UTT stands for? It stands for UnTesTed memory IC's. This means that after the IC's were packaged they were NOT speed binned and offer no ratings from the factory. This is done so DRAM manufacturers can cut 15 to 20 percent in costs per DRAM by skipping DRAM chip reliability burn-in and speed/functionality testing on high-end chip testers. Not only does this save money, but it also allows them to dump large volumes of IC's on the market faster because there is no testing done...For those looking to build systems that meet JEDEC specifications, it is bad news because the majority of these IC's are Non-Conforming (NC). For system builders or corporate builders these IC's are known to be bottom of the barrel and are something they wouldn't touch with a ten foot pole. This is why these UTT IC's come blank from the factory with nothing written on them as they are unrated."

Further on the author shares his thoughts on manufacturing costs stating memory makers use "...BrainPower PCB's (which are cheaper to build than the JEDEC reference design)..." There seems to be a semantic contradiction in the term Brainpower and it's association with cost saving measures. Since I don't know the actual cost of the PCBs compared to JEDEC reference designs I dare not refute the claim, however; I surmise a PCB capable of regulating such high current and dictating Low Latncecies at speed would be a more costly PCB to design or procure? Insofar as JEDEC specifications, if manufacturers strictly adhered to JEDEC there would be no memory available beyond DDR400 speeds. While verbatim implementation of JEDEC speed ratings wouldn't deter hardcore PC-Enthusiast's, the implications are scary. Once we begin to criticize manufacturers whom design specifically for the overclocking community (often at a financial loss) incentive to manufacture such products may disappear quicker then half-priced BH-5 at a water-cooled Malibu-Stacy LAN-Party.

The term "re-labeler" brings with it negative connotations since being misused as a convenient dispargement due to ambiguity among the layperson. To my knowledge there is only one company on the planet whose product bear their original name from silicon to package and that's Samsung. Just about every other manufacturer outsources some componnent, whether it's IC's, PCB's, or some other part and at at some time may have used blank IC's. Once they've placed their name on the package, technically they've relabled. For example Kingston arguably the largest memory supplier on the market outsources IC's as I've yet to hear or see any IC's, or PCB's bearing the Kinsgton name. Pertaining to the reliabiity of UT's IC's the article goes on to state; "...who knows how these modules will hold up down the road. With a lifetime warranty on memory running 3.5Volts possibly 24/7, we might be seeing a lot of burnt out modules one or two years from now." Fortunately there are two factors working in our favor. First; these modules are warranted for life. Second; at no time did raising voltage to the UTT based modules tested in this round-up require, or benefit from anything more then 3.3V. Quite often with our busy schedules we're too impatient to test our memory at the smallest voltage increments, and we tend to leap over potential sweet spots to the higher end of the spectrum; 3.5V - 3.7V. While I did place active cooling on the modules testesd, I also chose to insert a thermistor between memory heat-spreader and PCB to monitor temps. I did this with the proprietary fan on and off, finding temps to be just a few degrees warmer then any other DDR at the highest voltage tested (3.3V - 3.4V).

Mushkin has incorporated Redline HP (UTT based) kits throughout their High Performance line. Redline kits such as these are tested at more relaxed timings to keep costs down. At just $206 (SRP) for the 1GB pair featured here, they've paid homage to the art of overclocking in it's truest form, which was founded by budget minded Enthusiast’s wanting to squeeze every MHz possible for their money. While Overclocking has evolved into several categories such as the Xtreme approach where one employs a cost-no-object method to extract the ultimate in performance, Mushkin has not forgotten where Overclocking originated.

The results below were achieved after spending many hours testing and re-testing on what is arguably the most extreme overclocking AMD motherboard available. I was able to run these modules to 260HTT at their default timings and just requiring the voltage jumper for 3.3V ran the memory at a lower latency of CL2-2-2-6 to 245HTT. Mushkin was surprised I achieved the results I did and while they were happy for me they also want to indicate such results are not guaranteed. To reiterate, overclocking is always the luck of the draw and manufacturers only test each part for a certain performance ceiling, that I've found a kit with what may be an exceptionally high performance ceiling should be noted. It didn't hurt to have 4.0V available, the number of memory settings on the DFI, and proprietary cooling on the memory. I ran Mushkin's 1GB HP3200 at: 11x200FSB CL2-3-2-6 (2.7V), 10x245FSB 2-2-2-6 (3.3V), 10x260FSB CL2-3-2-6 (3.3V).

Madshrimps (c) Madshrimps (c) Madshrimps (c)


Madshrimps (c)


Madshrimps (c)


Madshrimps (c)


Madshrimps (c)


Madshrimps (c)


Madshrimps (c)


Madshrimps (c)
  • prev
  • next