OCZ Platinum XTC Series - ELV Triple Channel DDR3 DDR3 PC3-12800 Extreme Low-Voltage / OCZ3P1600ELV6GK
By: Dan Durland Introduction
Today I have another fine product from OCZ on the test bench. OCZ, a leading manufacturer of high-speed memory modules, have supplied me with a 6 GB (3x2GB) kit of their Extreme Low-Voltage DDR3 1600 MHz memory. Designed to be more power-efficient, OCZ ELV memory requires far less voltage while still maintaining stability at high speeds. Specifications
- 1600 MHz DDR3
- CL 7-8-8-24 (CAS-TRCD-TRP-TRAS)
- Available in 6 GB Triple Channel Optimized kits
- Platinum Z3 XTC Heatspreader (see quote below)
- Lifetime Warranty
- 1.35 Volts
- 240 Pin DIMM
(Xtreme Thermal Convection) heatspreaders optimize the thermal management of memory modules by promoting greater airflow by means of micro-convection throughout what is usually the dead air space inside conventional heatspreader designs. In this manner, build-up of heat is avoided and thermal dissipation of the memory components is offloaded more efficiently through the honeycomb design. At the same time, mechanical stability is maintained.
I could literally write for hours trying to explain exactly how memory works. And it could take almost as long to explain how to manually setup your memory because of all the different motherboard and bios combinations available. So I'm only going to list some basic information, if you have any questions please ask them in the forum and we'll try our best to answer them. Memory Frequency / DRAM Frequency
This setting will be displayed as a Ratio (2:6, 2:8, 2:10, 2:12, etc...) or as a Multiplier (6, 8, 10, 12, etc...). If you are presented with a ratio then you simply take the second number and that is your Multiplier. At the default BLCK setting of 133 we see that each Multiplier corresponds to the following Speeds.
BLCK x Memory Multiplier = DRAM Frequency
- 133 x 6 = 798 (800 MHz)
- 133 x 8 = 1064 (1066 MHz)
- 133 x 10 = 1330 (1333 MHz)
- 133 x 12 = 1596 (1600 MHz)
- 133 x 14 = 1862 (1866 MHz)
- 133 x 16 = 2128 (2133 MHz)
So why do some manufacturers give us ratio's, just what is that all about? Using 2:10 as an example, for every (2 MHz of the BLCK) : the memory bus speed (increases by 10 MHz). So at the default BLCK speed of 133 MHz we have 66.5 sets of 2 MHz multiplied by 10 equaling 665 MHz.
(133 MHz/2) x 10 = DRAM Bus Speed
66.5 MHz x 10 = 665 MHz Bus Speed (This will match the memory speed shown by CPU-Z)
665 MHz x 2 = 1330 MHz (our memory is Double Data Rate, DDR, so we multiply by 2). Memory Latencies
When you bought your memory it was rated to run at a certain speed (1333, 1600, 2000) in MHz with a pre-defined set of latencies (for example, 7-7-7-24, 8-8-8-24, 9-9-9-30). These latencies represent the following bios settings:
- CL - CAS#Latency - refers to the column of physical memory in an array of capacitors ( a grid comprising of rows and columns) used in DRAM modules. The latency refers to the active amount of clock cycles that must be expended to take a request. Combined, CL sends data from the memory controller, has it read to the memory location, and output to the modules output pins.
- tRCD - RAS to CAS Delay - The memory controller selects a bank, followed by a row location (using the Row Address Strobe, or RAS) and a column location using CAS. It represents the time in cycles for issuing a command and active read \\ write commands.
- tRP - Row Precharge Time - Row Precharge Time represents the minimum allowable time taken between any active command and the read \ writes of the following bank on the memory module.
- tRAS - Min RAS Active Time - Represents the amount of time taken between a row being accessed and deactivated. A tRAS row must be allowed to complete before being deactivated, setting this option too low can result in data corruption as the row is closed down too soon.
Depending upon your specific bios and the settings provided you may have multiple options available to adjust. Of those options we will only be looking at two (2) more of these.
- tRFC - Row Refresh Cycle Time - This parameter determines the amount of cycles it takes to refresh a row in a memory bank, again if this is set too low it will cause corruption and set too high will result in a loss in available bandwidth, but increase stability.
- CR - Command Rate - When the MC (Memory Controller) first tries to access memory, it has to latch onto a memory bank, known as CS (Chip Select). Then it proceeds to find the column (CAS), the Row (RAS), and then return the data to the CPU. Now, 1T means it takes 1 clock cycle to find a memory bank, vs. 2T where it takes 2 clock cycles to find the memory bank.
There was a time not that long ago when adjusting your system for maximum performance meant aquiring the lowest possible latencies, even at the expense of a few MHz in cpu speed. This is no longer true. With the memory controller built into the cpu we no longer have the bottleneck associated with the older technology. Instead we have a wider direct path between the cpu and memory controller which results in fast access to a tremendous amount of bandwidth. So, while we still want the fastest possible latencies, we no longer have to sacifice any cpu cycles. I would rather run at higher latencies at 4.0 GHz than slightly lower latencies at 3.9 GHz. DIMM or DRAM Voltage
This is the main voltage to your memory. Raising this voltage may help when overclocking your memory speeds or at high BLCK values. This voltage must be kept within .5v of the CPU VTT or QPI/DRAM Voltage (see below). CPU VTT or QPI/DRAM Voltage
This is the QPI or cpu core to memory voltage (memory controller voltage). This voltage must always stay within .5v of the DIMM voltage. If you set the DIMM voltage to 1.65v then the CPU VTT must be set to a minimum of 1.15v, at 1.70v the CPU VTT must be at a minimum of 1.20v and so on. Try to keep this voltage below 1.45v unless you have added a direct cooling solution for the Vreg/Mosfets (Vreg = Voltage Regulator).
Increase the CPU VTT when:
BSOD error code ***STOP: 0x00000124 "general hardware failure"
LinX errors happen only after 10 min or more
LinX hangs but does not BSOD
LinX reboots system without BSOD Image Gallery
So lets take a look at this memory in a sexy photo shoot, but be warned, the modules are sleek, shiny and quite promiscuous. Here we have our modules dying to be stripped from their pre-formed plastic packaging.
The modules are not the least bit coy and they really like to put on a group show.
I managed to separate this beauty from the others for a private shot.
And the modules finished the session out by lining up for an action shot. Test Setup
I'll be testing the OCZ Platinum XLV against a kit of OCZ Platinum LV modules with my i7 920 at both stock and overclocked speeds. Peak Wattage was measured using a Kill A Watt EZ P4460 plugged into the wall outlet with my APC Back-UPS XS 800 plugged into the Kill A Watt EZ.
- Intel i7 920 at 2.66 GHz and at 4.0 GHz
- EVGA E757 X58 LE
- OCZ Platinum XLV(6x2GB) DDR3 1600 7-8-8-24
- OCZ Platinum LV(6x2GB) DDR3 1600 7-7-7-21
- EVGA GTX 280
- Corsair 750w PS
- OCZ Agility 2 60GB – Windows XP Pro x64
- Seagate Barracuda 250GB SATA II 7211.10 – Page File/Data
- Proview 2200W 22" LCD Monitor
- APC Back-UPS XS 800
- Windows XP Pro x64
I need to note that at 4.0 GHz the OCZ3P1600ELV6GK modules would not post with the DIMM voltage set to 1.35 volts. At 1.45 volts the system successfully posted but it was not stable. Full stability was achieved once the DIMM voltage to the OCZ3P1600ELV6GK modules was set to 1.5 volts. Software
- Everest Ultimate 5.5
- SiSoftware Sandra Lite 2011
- Passmark Performance Test 7.0
- LinX 0.6.4
- Super Pi Mod 1.5 XS
- PCMark 04
- CineBench 11.5
- 3DMark 06
Everest Ultimate is a PC diagnostics software utility that has many useful features for monitoring and testing the performance of a PC and its various subsystems. I used the Cache and Memory Benchmark Tool to measure the Read, Write, Copy and Latency times. I will also note that within days of completing these tests and generating the graphs, Everest was aquired by FinalWire and renamed AIDA64 Extreme Edition. I performed a retest and the results were so similar that I felt creating new graphs were a pointless waste of time.
The results show the ELV modules running marginally slower than the LV modules at both stock and overclocked speeds. This can be attributed to the ELV modules slightly slower latency timings.
- SiSoftware Sandra Lite 2011
SiSoftware Sandra Lite 2011 is another PC diagnostics software utility that has many useful features for monitoring and testing the performance of a PC and its various subsystems. I ran the Memory Bandwidth and Memory Latency benchmarks and recorded the following data.
Sandra seems to confirm the Everest results as the ELV modules are again just slightly slower than the LV modules.
- Passmark Performance Test 7.0 64-bit
PassMark Performance Test is a PC hardware benchmark utility that allows everybody to quickly assess the performance of their computer and compare it to a number of standard 'baseline' computer systems. I used the Memory Mark benchmark which tests the allocating, accessing, memory speed and efficiency of the modules.
Much like the other benchmarks, Passmark is showing the ELV and LV modules as being almost identical with the only visible difference in the 4 GHz Large Ram Test. Because the Large Ram test is measured in operations per second it comes as no surprise that the slightly slower timings of the ELV modules cause them to fall behind.
LinX is a benchmark used to measure a cpu's performance in GFlops with G standing for G
iga and Flops standing for FL
oating point OP
erations per S
econd. In order to test the cpu's performance, data has to be moved through the memory sub-system and I've set LinX to use all of my available memory.
I have to say I'm slightly surprised at the LinX results. While they still bear a direct resemblance to the previous benchmarks, the ELV modules take 1st prize in LinX.
Super Pi Mod 1.5 XS is a benchmark that times the cpu's ability to calculate Pi to a specified number of digits. I ran the benchmark through its complete range of digits at both stock and overclocked speeds.
The results seem to be back on track with the ELV and it's slightly slower latencies trailing the LV modules by the smallest of margins.
PCMark 04 is an older version of Futuremark's long line of PC Benchmarking utilies. PCMark Vantage does not support Windows XP and PCMark 05 does not support 64-bit operating systems so I used PCMark 04 which still effectively measures memory performance. Using data block sizes of 8MB, 4MB, 192K and 4K PCMark 04 measured the Raw Block Read, Write, Copy and Random Access speeds.
The results almost mirror each other once again.
CineBench is a real-world cross platform test suite that evaluates your computer's performance capabilities. CineBench is based on MAXON's animation software Cinema 4D, which is used extensively by studios and production houses worldwide for 3D content creation. MAXON software has been used in blockbuster movies such as Spider-Man, Star Wars, The Chronicles of Narnia and many more.
This test scenario uses all of your system's processing power to render a photorealistic 3D scene (from the viral 'No Keyframes' animation by AixSponza). This scene makes use of various different algorithms to stress all available processor cores.
Single core testing shows the ELV and LV modules performing equally with the LV modules taking the lead at 4 GHz by the slimmest of margins. In multi-core testing we see the ELV modules edging ahead at 2.66 GHz with the LV modules capturing the 4 GHz test.
A 3DMark score is an overall measure of your system’s 3D gaming capabilities, based on comprehensive real-time 3D graphics and processor tests. For this review I'm only looking at the cpu score results to see what effect our memory has on them.
As we've seen through out the testing the ELV modules are performing slightly slower than their LV counterparts. Power Consumption
Using a Kill A Watt EZ P4460 I made note of the highest wattage level the system reached when running LinX and 3DMark 06. The Idle value was noted after the system was idle for a 30 minute period.
I finally found a test that the ELV modules dominate. While the ELV modules do use less power at 2.66 Ghz, the difference is still rather minimal. At 4.0 Ghz, however, the difference becomes much more pronounced. In our earlier tests we saw the ELVs take the LinX crown; even more impressive when we realise it was done with 15 watts less power. Conclusion Pros
- Uses Less Power
- High Performance
- Lifetime Warranty
- Attractive Low Profile Heatspreaders
- Slower Latency Timings
- needed an extra .15 volts at 4.0 GHz (DIMM Voltage)
You can look at the graphs and see that while the ELV modules were almost identical to the LV modules in every test, they were slower the majority of the time. Were they slow enough to make a noticeable difference for the average user? I really don't think they are. What stands out the most to me, after all of this testing, is the power differences I observed. In this age of ever increasing power consumption, OCZ has actually reduced the power consumption while still maintaining an extremely high level of performance.
If the latencies were just slightly faster I would award the OCZ Platinum ELVs Tech-Forums' highest award. But with the performance just below the LV modules I'm awarding a 4.5 Star Rating.
I would like to thank OCZ Technology for allowing Tech-Forums to review another one of their outstanding products.