Original Link: https://www.anandtech.com/show/2359
The launch of the Intel X38 Express chipset occurred two weeks ago amidst a lot of early hype about its potential and performance improvements over previous chipsets. The X38 chipset is finally starting to show some promise after several BIOS releases over the past two weeks on the first retail boards ASUS and Gigabyte. However, we are still not ready to declare the chipset a winner. The chipset is not a disappointment at this stage in the game, but the whole launch event fiasco and lack of choices available in the marketplace is disappointing to us.
We are busily completing an extensive testing phase on this new chipset with the ASUS, Gigabyte, abit, and Foxconn boards that we currently have in the labs. We expect retail boards shortly from MSI and Biostar with Intel and DFI following up sometime in November with their products. In the meantime, we have received an avalanche of requests centered on which board performs best out of the available products currently for sale.
We are very close to answering those questions as testing is nearing completion now. However, we did notice a central theme in the majority of the requests and it revolved around the overclocking aspects of the boards and performance comparisons to the current chipsets. We will show the comparisons against other chipsets in the full board reviews and offer in-depth overclocking results with multiple CPU and memory configurations. The quick overview we are presenting today will help set the stage for those articles by answering our maximum FSB rates attained with the Intel Q6600 and E6550.
We can estimate now based upon the progress made in the past few days that the overclocking potential of the X38 chipset with current processor families should slightly exceed the P35 once it matures, much in the same way the P35 has exceeded that of the P965 over the course of the summer. At stock settings, the X38 is not any different really than the P35/680i - and for that matter the 975X chipset it replaces. This is disappointing but understandable as the X38 Express chipset is based upon the P35 core and is still saddled with an onboard memory controller.
This memory controller design from all indications has just about reached its optimum level of performance after years of refinement by Intel. In the case of the X38, it appears that additional attention and emphasis was placed on DDR3 performance at the expense of DDR2. Several of the X38 equipped DDR2 boards have been delayed as the manufacturers have tried to bring performance up to par with the P35. While being at DDR2 performance parity with the P35 in considered a success by the suppliers, it might not be for the users. However, this has been the goal from the outset as you will start to see the DDR2 X38 designs supplement and then replace the higher-end P35 products in the $150~$250 market sector.
This should not be surprising as Intel is rapidly moving forward with DDR3 as their memory of choice for enthusiast level products. The X38 is the first enthusiast product from Intel designed with this in mind. We expect to see additional performance benefits with the X38 DDR3 based offerings over other chipsets when Wolfdale and Yorkfield (Penryn) launch next month.
We originally reported in our X38 chipset overview that the X48 chipset would be making an appearance in February. That was true until today, as it now looks like the rush is on from Intel to push this updated chipset (X38 revision 2 in our book) up to a late November or early December release. The X48 is a basic drop in for the X38 chipset so board designs will not have to be changed for the most part.
The X48 will bring "official" 1600FSB and DDR3-1600 support, a few minor memory controller tweaks, and supposedly slightly better overclocking on the very high end, provided you have a capable processor. Also, expect a significant price increase that should buy you a few minor performance improvements that are only noticeable in certain benchmarks. The rush to bring this chipset to market quickly now tells us more about Intel's CPU release schedule than it does about any perceived benefits at this point.
Anyway, let's take a quick look at some initial overclocking results with boards from ASUS and Gigabyte. We had hoped to have initial results with the Foxconn and abit boards, but final BIOS releases just arrived and our efforts are concentrated on finishing the full reviews first. As a matter of fact, we just received BIOS releases from ASUS and Gigabyte to correct a few overclocking snafus we have noticed, especially with 4GB memory configurations.
Test Setup
ASUS P5E3 Deluxe / Gigabyte GA-X38T-DQ6 / GA-X38-DQ6 Test Bed | |
Processor | Intel Core 2 Quad Q6600, 2.4GHz, 2x4MB, 9x Multiplier, 1066FSB |
CPU Voltage | 1.1500V Stock - 1.375V ~ 1.4750V Overclocked |
Cooling | Tuniq Tower 120 |
Power Supply | OCZ 1000W |
Memory | Corsair CM3X2048-1800C7D DDR3 Corsair CM2x1024 10000C5D DDR2 |
Memory Settings | Various |
Video Cards | MSI HD X2900 XT 512MB |
Video Drivers | ATI Catalyst 7.10 |
Hard Drive | Western Digital 7200RPM 750GB SATA 3/Gbps 16MB Buffer |
Optical Drives | Plextor PX-B900A, Toshiba SD-H802A |
Case | Cooler Master Stacker 830 Evo |
BIOS | ASUS P5E3 - 0504 Gigabyte X38T - F3I Gigabyte X38 - F5I |
Operating System | Windows Vista Ultimate 64-bit |
. |
Test conditions were maintained the same, as much as possible, over the platforms tested. We selected the Intel Core 2 Quad Q6600 as our processor of choice since it represents one of the better price to performance values in the midrange processor market and is the CPU we will concentrate on in future reviews.
Our E6550 is an engineering sample that was hand picked to help us determine maximum FSB levels. We have seen this CPU reach 570FSB levels on the DFI P35 board with air cooling and it potentially has the capability to reach 600FSB with the right cooling methods. Our two retail E6550 processors both reach 500FSB and then basically have a hard stop at that point when overclocking. We are utilizing Microsoft Vista Ultimate 64-bit as our operating system along with 4GB and 2GB memory configurations.
ASUS P5E3 Deluxe Overclocking
Q6600 8x475 2GB
Q6600 9x400 4GB
E6550 6x550 2GB
We were able to reach a stable 8x475 setting with the Q6600 and the 2GB memory configuration at DDR3-1900 8-8-8-24 1N. Stability was superb with the CPU requiring just 1.4500V and memory set to 1.95V. Memory Timings were set fairly aggressive at this speed considering the front side bus and strap settings, although we were still able to tune performance levels (tRD) to 6 with Memset and complete our test suite.
Once we moved to a 4GB configuration, the board would not POST with the memory set to DDR3-1680 or higher. In fact, we had a choice of running at 8-7-7-24 2N at DDR3-1640 or dropping down to DDR3-1600 but with timings set to 7-6-6-18 2N. We chose the slightly slower memory/CPU speeds as the benefit in improved timings made up for the differences in speed. Our full range of testing with 4GB was quite disappointing, especially when overclocking.
The most serious problems occurred at the 9x333 setting with the memory at DDR3-1333. The timings had to be set to 8-7-7-18 2N for any type of stability. We were able to run at 7-7-7-15 2N at DDR3-1333 but had to increase memory voltage to 2.10V and MCH to 1.8V, both settings being unacceptable for at this speed. However, ASUS has delivered a new BIOS that in preliminary testing has solved most of the problems, although we still have not gone above DDR3-1760 at this time.
Our E6550 reached an impressive 550FSB on this board. However, we had to run our memory at a 1:1 ratio for system stability. This was with 2GB or 4GB configurations and our best settings were at 6-6-6-15 1N but with performance level (tRD) set to 12. The other two memory ratios available with the 400 strap resulted in constant lockups when running our test suite. Hopefully, this problem has been addressed in the latest BIOS release.
Gigabyte GA-X38T-DQ6 Overclocking
Q6600 8x480 2GB
Q6600 8x450 4GB
E6550 6x550 2GB
We were able to reach a stable 8x480 setting with the Q6600 and the 2GB memory configuration at DDR3-1536 7-7-7-18 1N. Stability was excellent with the CPU requiring 1.4625V and memory set to 1.95V. Unlike the ASUS board, we could not run our memory higher than DDR3-1800 with 2GB on this board without really loose timings. In fact, the DDR3-1920 settings at this FSB speed had to be set at 10-10-10-30 2N with tRD set to 10 for stability. We found the DDR3-1536 timings offered better performance. Gigabyte is currently addressing this problem and the latest BIOS update has allowed our timings to improve to 9-9-9-24 but with tRD (performance level) increased to 11.
After installing a 4GB configuration, the board actually reversed course and allowed us to run at DDR3-1800 settings with 8-8-8-18 at 1N which greatly improved memory performance over the ASUS board. However, our maximum FSB dropped to 450, although in application testing the board performed similarly to the 2GB configuration in most cases. We also did not notice any of the problems at DDR3-1066 or 1333 that the ASUS board exhibited with the 4GB memory configuration. In fact, our timings at DDR3-1333 with 4GB were 6-6-6-15 1N and performance level (tRD) set to 4 which matched the 2GB timings.
Our E6550 reached the same 550FSB level as the ASUS board, but we were able to run our 2GB memory configuration at DDR3-1467 7-7-7-18 1N compared to DDR3-1110 6-6-6-15 1N on the ASUS board. Our application results are showing a slight advantage for the Gigabyte board, although the effort to tune the board properly was significantly more involved than the ASUS board. The one advantage we had with the Gigabyte board was that our CPU was set to 1.425V to reach the 3330MHz rating and the ASUS board required 1.475V for the same stability.
Gigabyte GA-X38-DQ6 Overclocking
Q6600 8x485 2GB
Q6600 8x450 4GB
E6550 6x550 2GB
The DDR2 based Gigabyte board reached an impressive 8x485 setting with the Q6600 and 2GB memory configuration at DDR2-1164 with 5-4-4-12 2N settings. Stability was superb with the CPU requiring just 1.4750V and memory set to 2.2V. We played with the memory settings to improve latencies, but the board was extremely particular when changing sub-timings. tRD settings were very sensitive and just about any setting other than auto resulted in lockups during testing. We also found the F3 BIOS offered the best memory performance although FSB overclocking was limited with the quad-cores when compared to F5i.
After installing our 4GB configuration, the board would not POST with the memory set to DDR2-1100 or higher. Our 485 FSB was still attainable with 4GB, but only after changing to the 333 strap and 1:1 memory ratio. Our timings had to remain fairly lax with settings at 4-4-4-12 and performance level (tRD) at 9. Gigabyte still has a way to go on tuning the board to run at its potential with 4GB.
The E6550 reached the same 550FSB level as the DDR3 boards and we were able to run the memory at DDR2-1100 with 5-5-5-15 timings. However, our sub-timings on the 400 strap were still loose in order to gain stability. Our DFI P35 board will run this processor up to 570 FSB with the memory timings at 5-4-4-9 and significantly better sub-timings. Our initial 4GB testing resulted in us lowering the FSB setting to 540 in order to gain memory stability at DDR2-1080.
Quick Thoughts
We still have significant testing to complete on our current inventory of X38 boards that also includes the Maximus lineup from ASUS in both DDR2 and DDR3 versions. The continued problems with 4GB memory compatibility leaves us concerned at this time. However, from all indications, the X38 boards are starting to mature with the latest BIOS releases. In fact, the number of BIOS releases we have received for the seven boards we have on-hand has left our heads spinning the past few days. We continue to believe this chipset has great potential and only needs further BIOS maturity; certainly, it is the chipset to have for CrossFire performance based on current test results. We continue to feel that the launch of this chipset occurred too early based upon our continued frustrations with the motherboards, but at least we see real progress finally being made. We will be back shortly with full reviews of the motherboards, extensive overclocking coverage, and head to head comparisons against the P35, 680i, and 975X chipsets.