With the movement to second or even third systems in the home (primarily den and living room) for the purpose of entertainment and some retro gaming – SFF (Small Form Factor) has gained a bit of ground. Thin cases of less than 4 inches and mini/thin mini ITX (the thin mini ITX generally uses SODIMM) make great rack mount components to go along side your Receiver, DVD/Blu-Ray player and CD Changer. Having the lowest profile memory in these systems allows unhampered airflow and more space for wiring.
Crucial has provided the Ballistix Sport VLP for testing and evaluation. This memory is about as low as you can go and it is at the speed sweet spot of 1600 with latencies 9-9-9 and a cycle time of 24 running on 1.35v. These specs make it near the top for 1.35v memory and the very low profile makes one wonder why other companies have to use such large heat spreaders with their 1.35v memory.
For comparison the same setup that was used for the Crucial DDR3-1600 CL8 VLP UDIMM Tactical as well as Intel Celeron (1155 SB core) ITX and AMD E450 Based ITX systems.
Speed DDR3 1600 (PC3 12800)
Cas Latency 9
Dual Channel Kit
Aluminum heat spreaders
Due to the tight spaces involved, some of the images could not come into complete focus; however, the advantages in a SFF system are quite visible. In the first two shots we see that, in a passively cooled system, airflow can be improved allowing for a quieter system or opening up some cooling headroom for overclocking. In the latter shots we again see advantage go to the lower profile Crucial VLP sport memory where simple CPU HSF connecting wires can hang up on memory modules.
Looking at this memory it is obvious that the heat spreaders (sinks) are not suited for dissipating great amounts of heat (it has a little more than half the effective heat dissipation area of the Tactical LP). This does not mean the memory should be avoided and thought of as low performance. Deciding how this memory would best be used led to the test methodology.
It was clear that this memory was not record setting but the question arises: how much performance does one really lose? Let’s take a good look at this in a few scenarios and see.
Looking at the SPD timings it was concluded that running in XMP 1.35 volt profile would be the easiest thing to do as it gives a better view of what should be expected of this memory. The first testing was done on an HTPC setup with a Sapphire H67 ITX board and Celeron G440 at the heart with 64 bit Windows. Here a realistic scenario is presented (my bedroom HTPC) which could apply to many systems: 4GB of ram on a 64 bit OS using only one stick of memory to keep heat and power usage down.
Memory speed is generally limited by the CPU on these SFF, low power CPU, setups so testing was limited – just to get a look at the potential advantages. Just three tests were run on this system (NovaBench, PCMark7 and STALKER COP) and gains that are attributable to the larger memory size were observed. It is notable that the STALKER COP benchmark actually was faster with the 4GB memory setup.
Testing for compatibility with an AMD Brazos(E450) system and a P55/i7(870) system was done as well to ensure cross platform/generation functionality. Everything went perfectly with no BIOS resets needed. A smoother boot was noticed on the AMD system and the i7 system did benefit in responsiveness from the move to 1N (1T) where 4 sticks of 4GB 2N (2T) were previously used to reach 16GB. This memory would be a worthwhile upgrade as tested on all three of these systems.
Final testing was done using an i5 3570 @ stock speeds on a Gigabyte UD5 z77 motherboard with a nVidia 465 for graphics. Here testing was done comparing to a baseline of 800 and testing from 1600 up to 2133. The Ballistix Sport sticks were clocked at 1600 and 1866 for the testing. 2133 did not play nice with these sticks and increasing the voltage was not an option for us. These sticks are purpose built and that is the way they will be tested. That is in no way saying that the sticks can not reach higher speeds; but, it is saying that it is not necessarily the most responsible thing to do.
3DMARK 11 Extreme 3DMARK Vantage Performance Aquamark 3 Cinebench 11.5 CPU Cinebench Open GL S.T.A.L.K.E.R. COPMafia II PCMARK 7 NovaBench SpecViewPerf The benchmark testing conclusively shows that though memory is capable of massive throughput the CPU does not take advantage of it. In encoding tests little (generally nothing) seems to be gained by moving to faster memory. In gaming tests 1600 seems to take care of the job as well. The 800MHz comes in only a few percentage points behind the 2133 MHz and in most cases it must be asked when will the CPUs catch up with the massive headroom that the faster memory has opened up?
Heat is the main reason we chose to run this memory in the 1.35v profile. With the advantages of faster clocked memory being non existent, in most cases, the memory was run as it was designed to be. LV and cool.
First the meter is set up and the system idles for about 20 minutes Running about 18º F above ambient room temperature is not bad at all, but the small heat sinks (spreaders) make one think that these chips will get way hot. That is not the case.
To get a baseline IBT was run at Maximum 1.5v XMP 1.5v and then at 1.35v The 1.35v had a 4 degree advantage which is not bad at all considering the small surface area of the heat spreaders. With a baseline set the final test would be run: At 1866 1.35v the memory runs a full degree cooler than 1.5v at stock speeds. Cool running and a little overclockable (2133 was not a great thing on these chips).
These memory modules seem to be the perfect solution for the SFF PC, or in cases where the Ballistix tactical LP are too large, snuggling up under the HSF. They run cool helping keep case temperatures down, they run low voltage which for the power miser is a plus and they are not hampered by the base 1600 speed.
Timings, one time maker or breaker of a system, were not focused on. Timings were everything on a performance system in the past and are still relevant in some cases. With the large caches, much improved prediction and flushing algorithms, timings simply play a lesser part in the performance metric on main stream processors. Conversely, on processors that lack l3 cache or have weak prediction and flush capabilities (Many AMD processors have no L3 benefit from faster [tighter, lower] timings) the work is done in a fast swap within the memory and the timings are critical to performance.
For bringing us (the enthusiast community) low profile memory that is lower than the clasps that hold it in, at a decent price, Crucial gets the gold For having the lowest profile memory out there with par performance Crucial wins the silicon crusher.
Disclaimer: The comments made about the AMD processors were not meant in a derogatory fashion. I do not play partisan electronics. I select components based on need. I do not recommend one brand over the other and I use both for the application I feel best suits the processor.