For
the casual user, Mushkin
may not be a name they'll be familiar with, but for enthusiasts,
Mushkin is actually quite well known. They have a history of making
quality products, and speaking of their ram products in particular,
they tend to be quite speedy and overclockable.

There's
no shortage of PC4000 ram manufacturers, and Mushkin actually has
a couple of different products for you to choose from, depending
on the budget. Today, we'll be looking at their High Performance
Black PC4000 memory, which is their top-of-the-line product.
Specifications
512MB (2x512MB as tested today)
4.0ns Hynix IC's
500MHz (PC4000) Unbuffered DDR Modules
CAS latency: 3 clock cycles
RAS precharge: 4 cycles
RAS active to precharge: 8 cycles
Command rate: 1T (1 cycle)
Aluminum heat spreader for thermal diffusion (Flat Black)
Latency: 3-8-4-4-1T
Jet Black 6-Layer PCB
2.75V
The
sticks we'll be looking at aren't normally sold as a kit for Dual
Channel users. We'll be testing them in Dual Channel mode though,
but just in case you have some reservations, Mushkin did test them
on DC motherboards.
"Tested
compatible with VIA, AMD, Intel, NVidia and Sis chipsets. Overclockability
of your system is dependent on all other components including but
not limited to Motherboard, CPU, graphics and PCI cards and other
peripheral devices. Our PC4000 eliminates the memory as a potential
bottleneck when overclocking. Tested on motherboards with the KT400,
P4X400 and NForce2 and Intel 865, 875 chipsets most of which are
officially supporting the PC3200 (DDR 400) spec."
Mushkin High Performance Black PC4000 Memory
I don't
have any pictures of the packaging, as I destroyed
it while getting the ram out. I am uncertain if what I got was the
retail packaging, but both ram modules arrived in an anti-static
bag, encased in a large, plastic air bag to protect it during shipping.
There's
no documentation, but the Mushkin's target audience will certainly
not need any.
Mushkin's
products carry a limited lifetime warranty, meaning that if it's
a manufacturer defect, they will replace your ram (shipping not
included). If you run your memory out of spec, modify it, or simply
put it through a ton of abuse, your warranty can be kissed good-bye.
This is the case with most manufacturers, but at least Mushkin states
it clearly so there's no misunderstanding.

Like
many other manufacturers, Mushkin uses Neng Tyi aluminum heatspreaders.
These are anodized black, which gives them a sophisticated look
with the silver tabs. I've never felt heatspreaders were mandatory,
as our pasts tests have shown that they don't always help in overclocking,
but they do do the job as the Mushkin modules became quite warm
during testing.
Mushkin's
ram modules are built on a 6-layer PCB process. Lower quality PCBs
are made using a four-layer process. The power and ground layers
are sandwiched between the two outer signal layers. Though this
process is cheaper than six-layers, the modules are vulnerable to
signal noise which can cause instability and corruption. In a six-layer
design, two additional grounding layers are placed on the outside,
shielding the modules more efficiently, which is especially important
as we move from PC3200 and up.

The
memory label has very little information about the ram. Only the
brand, the memory size, and speed are indicated here. There's no
mention of memory timings, though on their website (and confirmed
via the BIOS of our test system) the timings are indicated as 3-8-4-4-1T.
Overclocking
and Stability Testing
Since
we're using a 200FSB (800MHz) ABIT IC7-MAX3 motherboard, the first
thing I wanted to try was what are the tightest timings I could
run the Mushkin PC4000 at 200FSB. After a bit of tweaking, it seems
that 2-3-3-5 is about as low as I can go at 1:1.
 |
|
200FSB
(400MHz) @ Tweaked
|
vCore
needed a boost to 2.65v though, but the ram was able to past Memtest
at these settings. Next up was the rated speed of PC4000 (DDR500
or 250FSB) at 1:1. At default timings, 250FSB was no problem at
all, though we needed another bump in voltage up to 2.75v.
Next
thing we did was to adjust the timings and to see what the Mushkin
would allow us to tweak to. At 250FSB, the best we managed was…
 |
|
250FSB
(500MHz) @ Tweaked
|
The
ram was completely stable, and passed MemTest without any problems
at 2.5-4-3-7. Memory voltage was at 2.75v, and even with an additional
bump to 2.9v, we were not able to tweak it any further. I have read
reports that there are some voltage fluctuations on the IC7-MAX3
at 2.9v of -/+ 0.1v, which I can confirm by watching the BIOS readings,
though at 3.0v, the problem seemed to have gone away. I was still
not able to tweak the ram further though, and the system would repeatably
hang at POST.
We
relaxed the timings back to the Mushkin default of 3-4-4-8, and
went forward with the FSB once again, adjusting the voltages as
required until we settled on our final overclock.

280FSB
Max OC
At
1:1 CPU and memory, we settled on a final OC of 280FSB. I'd be lying
if I said the ram was running 100% reliably, as I did experience
lockups when running stress tests. All our benchmarks ran
fine however, and although I am going to present those numbers,
just keep in mind that we were at what I would say 75% stable. This
was at 3.0v, and a Delta 80mm 68cfm fan blowing directly on to the
ram modules. I removed one module, and retested the overclocking
in single channel mode, but our end result did not change.
Lowering
the speed down to 274FSB at 2.8v was enough to keep things running
smooth.
Test
Setup
ABIT
IC7-MAX3: Pentium 4 2.4C, 2 x 512MB Mushkin PC4000, ATI
AIW Radeon 9800 Pro, 120GB Seagate, Windows XP SP1, ATI Catalyst
3.6.
Benchmarks
will be presented at 200FSB, 250FSB and maximum OC. Keep in mind
that the numbers at maximum OC will be skewed since we're running
1:1 and the CPU's speeds will be different when showing each memory's
OC speed. There isn't anything I can do with this since it's impossible
to set the memory speeds independently with the CPU FSB unless we
go with a lower memory divider.
Test
Software for both platforms will be:
SiSoft
Sandra 2003 Memory
PC Mark 2002
Memory
PiFast
TMPGEnc
AVI-to-MPG Encoding
Unreal Tournament 2003
Game
Accelerator will be configured at Street Racer for 200FSB, and Auto
for the rest. Competing sticks will be OCZ's
1GB PC4000 EL Gold, 1GB
Kingston HyperX PC4000, and the Corsair
TWINX1024-4000 Pro Series. At 200FSB, all the competing ram
kits will be running at 2-3-3-5, and the HyperX at 2-3-3-6. At 250FSB,
the ProSeries will be running at 2.5-3-3-6, whereas the OCZ and
Kingston kits will be at 2.5-4-3-6, since neither could do 2.5-3-3-6.
The Mushkin will run at 2.5-4-3-7, which was the best I could manage
at those speeds. All the ram modules will be running at 3-4-4-8
at their maximum overclocks.
SiSoftware
Sandra 2003 Memory - Pentium 4 @ 12x200
Although
a synthetic benchmark, it's a popular one, freely available if you
wish to make comparison benchmarks. We will be testing the memory
speeds only. All memory timings are at 2-3-3-5,
with the exception of the HyperX, which will be at 2-3-3-6.

SiSoftware
Sandra 2003 Memory - Pentium 4 @ 12x250

At 200 and 250FSB, Mushkin only trails the Corsair
TWINX by a slight margin, and edges out the other two, despite running
at a slight timing disadvantage at 250FSB. Not bad at all.
SiSoftware
Sandra 2003 Memory - Pentium 4 @ Maximum OC (Mushkin @ 280FSB)

In case you missed it in our last reviews, the TWINX
kit is clicking away at 285FSB, the HyperX at 272FSB, and the OCZ
EL Gold at 280FSB. Trailing the TWINX by 5MHz, Mushkin still performs
within percentage points of Corsair's finest.
PC
Mark 2002 -
Pentium 4 @ 12x200

PC
Mark 2002 -
Pentium 4 @ 12x250

Very close across the board, and the Mushkin PC4000
finishes in a solid second place.
PC
Mark 2002 -
Pentium 4 @ Maximum OC (Mushkin @ 280FSB)

The CPU clock speed differences between the ram modules
account for the differences in performance. Given that the IC7-MAX3
doesn't allow memory speed adjustments independent of the CPU speeds,
there isn't much we can do here. Anyhow, this is a purely synthetic
test, so let's see if things look up as we move to real world tests.
NEXT
|