In The Package

Kingston HyperX DDR3-2000 4GB Memory Kit Review

Add Your Comments

When it comes to memory, Kingston is no spring chicken: they’ve been in the game since 1987. Their global sales in 2009 reached over $4.1billion (Dr. Evil would be proud) and their HyperX series of RAM is backed by a lifetime warranty. This makes it popular with overclockers.

For this review, we’re looking at a new mid-level dual-channel 4GB DDR3 kit, their DDR3-2000 with rated timings of 9-10-9-27.

Packaging, Specifications and First Impressions

The Package

Kingston’s model numbers aren’t the shortest – this kit is model # KHX2000C9AD3T1K2/4GX. It comes in the typical formed plastic pack; with a decent packing job from Newegg or whomever you buy them from, it will protect the sticks just fine.

In The Package

In The Package

Specifications

This particular kit is rated for timings of 9-10-9 (tCL-tRCD-tRP) at 1.65 V. You may notice it shares a very similar model number with an older revision of it that is rated 9-9-9. This is a new revision of that kit, hence the “A” in the model number. That revision means that they are using a different build that carries with it loosened timings of 9-10-9 due to chip availability.

The 27 tRAS timing used in the review comes from all of the other kits on their page rated at 9-9-9. In a similar fashion, other kits rated at 8-8-8 have a tRAS of 24 and 7-7-7  of 20. The specifications haven’t been updated on Kingston’s site yet, but this is what Kingston tells me they have now changed to.

In the screenshots below, some of you may notice the tRFC was run at 110. This is because the kit didn’t get along with my board at 200BCLK with tRFC of 88, where I had set the other two kits that are referenced in this review. According to Kingston, 110 is Intel’s recommended spec for tRFC and sure enough, it was stable throughout testing. At their rated speed and timings but at 167BCLK (and 12x memory multiplier), the memory was just fine and stable as they could be running the tighter tRFC.

First Impressions

HyperX  modules are Kingston’s performance series and they all look similar. The heat spreaders are a striking metallic blue and rather large. If their size is any indication of their ability to remove heat, they’ll keep your sticks nice and cool.

RAM Unpacked

RAM Unpacked

Label Side Up

Label Side Up

They’re almost 1/4″ taller than the G.Skill Pi sticks reviewed previously, which were already tall to begin with. If you use a wide air cooler with 38mm fans clearance may be an issue depending on your board’s RAM placement. This is common with a lot of high performance RAM these days (the G.Skill kit included) so it’s not a new concern for overclockers.

Kingston v. G.Skill Height

Kingston v. G.Skill Height

Kingston was also kind enough to include one of their HyperX fans. This model is stronger than the plastic fan included with the G.Skill kit. It’s also quite large, so be sure to pay attention to clearance issues if you choose to get one. The fans are silent yet move a lot of air.

HyperX Fan

HyperX Fan

Fan Parts

Fan Parts

Fan Assembled

Fan Assembled

Fan Stood Up

Fan Stood Up

The RAM and fan look quite striking when installed:

Installed

Installed

Installed in the Dark

Installed in the Dark

Fans Top Down

Fans Top Down

They definitely look quite pretty all lit up. The fan itself will cost you $22.99 at Newegg. Its metal construction and quiet fans are worth a look if you think your sticks will need a little extra cooling or if you case’s air flow isn’t the best.

Looks can only take a pair of memory sticks so far though, so let’s see how they perform!

Testing and Performance at Rated Speed

Before performance testing, it’s always good to look at stability. Ultra-X has generously supplied their RAM stress tester, the R.S.T. Pro 3 PCI Express. There are only a couple review sites that have one of these and we are proud to be one of them. It’s not your standard Memtest++ bootable CD. This is a standalone, bootable piece of hardware that plugs into a PCIe x1 slot and, man, does it ever beat on some RAM. By the time it was done the sticks were very hot, even with the case side panel off and the supplied fan blowing air on them, just like the G.Skills before them.

Several manufacturers use this to test their memory before it goes out the door to reduce RMAs, Kingston included. If it passes this test, it is stable and then some. Manufacturers generally run the full memory test for three passes. We’re overclockers though and tend to abuse sticks a little more than other people, so for our purposes, doubling the manufacturer’s requirement to six passes sounds good.

These sticks were set to their rated speed using the 10x multiplier at 200BCLK with timings 9-10-9-27. vDIMM was 1.65V as specified and VTT was set at 1.4V. They made it through all six passes with zero errors. Nice job Kingston!

Ultra-X R.S.T. Pro 3 Pass

Ultra-X R.S.T. Pro 3 Pass

To establish a baseline for performance all tests were run on a 4GB set of G.Skill Trident DDR3-2000 with timings 9-9-9-24. Also included in the results is the G.Skill Pi DDR3-2400 kit reviewed earlier. Every benchmark was run three times and the best score is recorded below.

Obvious anomalies, if they existed, were discarded and the benchmark was run again. The only tests that this happened on were the Everest read/write/copy tests, where they’re known to glitch about 10,000MB/s over every other run. This only happens rarely, but does happen.

All tests were run with the CPU at 4.0 GHz (200 BCLK x 20) and RAM set to its rated speed and timings. The rest of the system in question consists of:

  • CPU: Intel i7 870, kindly supplied by Intel.
  • Motherboard: EVGA P55 FTW
  • GPU: HIS HD4890 Turbo
  • PSU: Corsair TX650
  • OS for stock testing: Windows 7 Professional x64.
  • OS for overclocking: Vista Enterprise x86.

3D Performance

I decided to drop a few 3DMarks this time, both to save time and because the difference was so minimal. The 3D benches of choice are not only the most recent, but are the two that showed the largest difference in the prior review – 3DMark06 and 3DMark Vantage

3DMark Graph

3DMark Graph

As you can see, even with the two benches where there was a larger difference, that difference is still very minimal. Both of the other kits tested lost out to the tighter timings of the baseline kit running at DDR3-2000 / 9-9-9-24. Just barely though: with the HyperX set losing 1.41% and 1.46% on 3DMark06 and 3DMark Vantage, respectively.

Rendering & Compression Performance

For some “real world” testing I used Cinebench R10 64-bit to test graphics rendering & WinRAR for compression performance. To flesh out this section a little more, I’ve also included results from Cinebench R11.5 64-bit.

Cinebench R10

Memory Score Percent Change
G.Skill Trident DDR3-2000 23419 n/a
G.Skill Pi DDR3-2400 23354 -0.28%
Kingston HyperX DDR3-2000 23422 + 0.29%

Cinebench R11.5

Memory Score Percent Change
G.Skill Trident DDR3-2000 6.86 n/a
G.Skill Pi DDR3-2400 n/a n/a
Kingston HyperX DDR3-2000 6.89 + 0.44%

This one was surprising. The HyperX set managed to edge out both previous kits in R10 and the baseline kit in R11.5. The image rendered isn’t that large in size, so this one’s less of a memory test than a CPU test, but still, a win is a win!

WinRAR

Memory KB/s Percent Change
G.Skill Trident DDR3-2000 4345 n/a
G.Skill Pi DDR3-2400 4504 +3.66%
Kingston HyperX DDR3-2000 4281 -4.95%

Compressing files definitely shows a substantive difference, losing 4.95% to the baseline kit and 8.61% to the (much more expensive) DDR3-2400 kit.

Synthetic Number Crunching Performance

Here we’ll have a look at how each kit performed when running benchmarker favorites SuperPi 1M, SuperPi 32M, WPrime32M & WPrime 1024M.

SuperPi 1M

Memory Time (Seconds) Percent Change
G.Skill Trident DDR3-2000 10.477 n/a
G.Skill Pi DDR3-2400 10.422 -0.52%
Kingston HyperX DDR3-2000 10.437 +0.14%

SuperPi 32M

Memory Time (min:sec) Percent Change
G.Skill Trident DDR3-2000 9:21.109 n/a
G.Skill Pi DDR3-2400 9:15.844 -0.94%
Kingston HyperX DDR3-2000 9:21.797 +0.12%

Nothing to call home about here really. They did lose a little from the baseline and a little more from the high-end kit, but all in all, the timing reduction doesn’t seem to have hurt tremendously.

WPrime 32M

Memory Time (seconds) Percent Change
G.Skill Trident DDR3-2000 6.378 n/a
G.Skill Pi DDR3-2400 6.390 0.19%
Kingston HyperX DDR3-2000 6.391 0.20%

WPrime 1024M

Memory Time (seconds) Percent Change
G.Skill Trident DDR3-2000 193.383 n/a
G.Skill Pi DDR3-2400 193.398 +0.01%
Kingston HyperX DDR3-2000 193.234 -0.08%

The difference here was somewhat of an enigma. As expected, it lost just a little bit in the 32M test, but it somehow managed to gain 0.08% in the 1024M test. Only losing 0.20% in the quick test and gaining a hair in the longer one isn’t bad at all!

Synthetic Bandwidth and Latency Results

Since synthetic bandwidth & latency tests tend to better highlight the differences in memory, I added MaxxMEM to the rotation. It has been added to the list of HWBot benches (though it still isn’t scored for hwboints), so it’s a good one to add to our testing.

Everest

Everest Read

Memory Speed (MB/s) Percent Change
G.Skill Trident DDR3-2000 19414 n/a
G.Skill Pi DDR3-2400 20869 +7.49%
Kingston HyperX DDR3-2000 19336 -0.40%

Everest Write

Memory Speed (MB/s) Percent Change
G.Skill Trident DDR3-2000 15787 n/a
G.Skill Pi DDR3-2400 15827 +0.25%
Kingston HyperX DDR3-2000 15794 +0.04%

Everest Copy

Memory Speed (MB/s) Percent Change
G.Skill Trident DDR3-2000 22662 n/a
G.Skill Pi DDR3-2400 23285 +2.75%
Kingston HyperX DDR3-2000 22503 -0.70%

Everest Latency

Memory Latency (ns) Percent Change
G.Skill Trident DDR3-2000 37.7 n/a
G.Skill Pi DDR3-2400 35.5 -5.84%
Kingston HyperX DDR3-2000 38.0 +0.80%

Not a horrible loss really. Compared to the baseline, it didn’t lose more than 1% anywhere across the board. It even gained a little in write speeds. The largest hit was on latency, but it is still a respectable showing.

MaxxMEM

MaxxMEM Read

Memory Speed (MB/s) Percent Change
G.Skill Trident DDR3-2000 17694 n/a
G.Skill Pi DDR3-2400 n/a n/a
Kingston HyperX DDR3-2000 17640 -0.31%

MaxxMEM Write

Memory Speed (MB/s) Percent Change
G.Skill Trident DDR3-2000 15715 n/a
G.Skill Pi DDR3-2400 n/a n/a
Kingston HyperX DDR3-2000 15730 +0.10%

MaxxMEM Copy

Memory Speed (MB/s) Percent Change
G.Skill Trident DDR3-2000 19745 n/a
G.Skill Pi DDR3-2400 n/a n/a
Kingston HyperX DDR3-2000 19663 -0.42%

MaxxMEM Latency

Memory Latency (ns) Percent Change
G.Skill Trident DDR3-2000 43.5 n/a
G.Skill Pi DDR3-2400 n/a n/a
Kingston HyperX DDR3-2000 44.6 +2.53%

MaxxMEM showed similar gains and losses for read, write and copy, but the latency result was a surprise, losing over a nanosecond. It’s not a horrible loss, at 2.53%, but it does speak to the importance of those sub-timings.

The results above were obtained from running the individual tests three times & taking the best, but some people like the Everest quick bench, so here they are side by side.

Trident Cache & Mem Bench

Trident Cache & Mem Bench

Pi Cache & Mem Bench

Pi Cache & Mem Bench

Kingston Hyper X

Hyper X Cache & Mem Bench

Overclocking

The testing at stock is a very necessary step to see what you’re getting when you pay for the rated timings and voltage, but now we get to the fun stuff!

As I’ve said in the past, you can spend hours and hours tuning memory to get the best timing and MHz combination. The results you see below are only a few hours worth of toying around – quick and dirty, if you will. They were the best I could get in a relatively short amount of time. You may do better (or worse). As always with overclocking, your mileage may vary.

To start, we’ll see what happens when you raise vDIMM to 1.68 V, drop the memory ratio & crank down the timings. Here’s the best I was able to get:

HyperX Cranked Timings

HyperX Cranked Timings

DDR3-1800 at 7-9-8-24 and able to run all of the tests in the screenshot thrown at them – not too shabby!

Now the testing I have the most fun with. Bump vDIMM up to 1.72 V and see just how far you can push the MHz at the rated timings and run SuperPi1M without crashing. Unfortunately, the result wasn’t necessarily exciting.

SuperPi1M Max MHz

SuperPi1M Max MHz

Ok, so they’re not stellar clockers at the rated timings – 1044 Mhz (DDR3-2088). No one guarantees overclockability (which is probably not a word anyway). You get what you paid for, anything else is just icing on the cake.

I didn’t feel like being done yet though. Let’s see what happens when you loosen them just a hair to 9-11-9-27, shall we?

SP1M Max MHz Round 2

SuperPi1M Max MHz Round 2

Now we’re cooking with gas! Just a small bit lost in timings let them get to a respectable DDR3-2304. Here’s the validation for kicks.

CPUz Validation DDR3-2304

CPUz Validation DDR3-2304

Conclusions

Let’s start with the list and go from there.

Pros

Cons

  • Timings are a bit loose.
  • Overclocking headroom had to be unlocked by slacking timings.

This kit is a strong contender, especially at its price point. For $144.99 with free shipping, it’s only $5 more than the cheapest 4GB DDR3-2000 kit on Newegg. They definitely can clock, and clock well. Exceeding DDR3-2300 is not a bad speed at all, even taking the timings into consideration.

While you won’t set any bandwidth records with them, you also can’t expect to. They’re a mid-level kit, but their overclocking potential is approaching the high-end. For these reasons, I’m happy to say the Kingston HyperX KHX2000C9AD3T1K2/4GX kit is for sure, Overclockers Approved!

-Jeremy Vaughan (hokiealumnus)

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>