UPDATE [26-Jul-07]: Added details about max FSB, Corsair overclocking & timings and pictures of my Ninja in the motherboard.
__________________________________________________________________________
INTRODUCTION
I?ve just put together a new E6750 system and I thought I?d share my overclocking experiences for the public interest. This is the first time I?ve done something like this, but its is the sort of write up I?d like to see so here it goes.
Here are the new components (prices in Australian dollars):
- E6750 ($239)
- Gigabyte P35-DS4 ($269)
- Corsair TWIN2X1024-6400C4 (4-4-4-12 @2.1V) ($202)
Here is what I kept from my last system:
- P180
- Scythe Ninja (Rev. A) with Nexus 120mm
- G15 Keyboard
- Leadtek 7800 GT
- Tagan 420-U02 (22A +12V)
Here is what I upgraded from:
- DFI Lanparty UT Ultra-D 939 (just sold for $142)
- AMD Athlon X2 4600+ @ 2.6Ghz (just sold for $354)
- 1.5GB Corsair Value Select DDR PC3200 RAM (selling now?probably get around $80-$100)
As I side note, I?m running Vista Home Premium (I upgraded from Windows 2000) and as any silent PC aficionados may note, my old setup was designed to be fairly quiet.
THOUGHTS ON THE UPGRADE
The Q6600 was very tempting but in the end I decided to go the dual-core route. Why?
1) I probably would have required a new power supply as I?m near the limit of the 22A with the current setup and I didn?t want to cough up the extra cash.
2) As I?m primarily a gamer, the potential higher clock was more attractive and while there are games that are coming out soon that are truly multi-threaded (uses more than 2 cores) I?m not sure whether there will be a huge difference between dual-core and quad-core (e.g. the particle stuff demoed for HL engine isn?t coming in EP2) performance in the next 6-9 months. But I could be completely wrong?we?ll have to wait and see.
3) Cooling an overclocked 65nm quad core isn?t easy (even with a Ninja), especially when don?t want your PC to sound like a plane taking off.
4) With the P35 purchase, I can upgrade to a cooler 45nm Penryn quad core next year when games start to make more use of them.
5) I can use $100 I saved to help buy an 8800 or 8900 later this year.
I got the P35-DS4 for good passive cooling so it?s quiet, great overclocking, better stability, and to a small extent better resale value (don?t laugh?I just sold my DFI Lanparty UT Ultra-D 939 for $142 on eBay when I paid $199 two years ago). As I work from home a software developer I can tax deduct a high proportion of all this so the board effectively cost me ~$210.
I got the Corsair 6400 with tighter timings to deal with the main problem of OCing the E6750?the low multiplier.
SYSTEM SETUP
No problems setting up. The Scythe Ninja fit perfectly into the P35-DS4 and is sited close to the top back of the case so its gets lots of extra airflow from the top and back P180 exhaust fans. Here are some pics of the Ninja in the board:
http://img517.imageshack.us/im...njarevaongap35sce9.jpg
http://img339.imageshack.us/im...njarevaongap35szl7.jpg
Only tricky thing was stretching the ATX_12V_2X cable from the power supply on the bottom of the case to the top left of the motherboard. Definitely mount the CPU and cooler before you put the motherboard in the case. The DS4 set the RAM voltage at the standard 1.85V so I had to manually set it to the correct 2.1V designed for the RAM.
As the P35 only has one IDE connector, and I had three IDE hard-drives (and one SATA drive) and one IDE DVD in the previous system, I bought a SATA DVD and dumped a 60GB IDE hard drive. But the placement of the IDE connector on the bottom of the motherboard worked perfectly with the P180, as I placed the two IDE drives in the lower bay, reducing heat and increasing airflow by eliminating those nasty IDE cables from the air passage of the main case body.
MOTHERBOARD IMPRESSIONS
For those interested in the P35-DS4, my overall impression is a very good one. The build is great and the passive cooling system works brilliantly (especially in the P180) where my chipset temps don?t deviate from 40-41c. I?ve even bumped the chipset voltage bit during the OV process and it didn?t make any difference. The layout is very good with lots of room around the ram and the connectors are all in a good spot.
The only thing that stops me from raving about the board is the BIOS. I?ve just from the DFI Lanparty NF4 board and in comparison it needs a lot of work. System monitoring just gives you ?OK? rather than voltages running though the components. In the OCing options, you can only set voltage increments (i.e. +0.05 V) rather than actual voltages on everything but the CPU. This means you?ve got to know what the correct voltage should be (see above re the RAM). At least you can set the processor voltage as a number rather than an increment.
On the bright side, Gigabyte can always improve the BIOS UI. At least they got the hardware right?And it?s really right.
NOTES ON MONITORING
I?m using Orthos and TAT to load the processor and SpeedFan to monitor the temperatures. A quick rundown for those who don?t know - C2D have three temp probes: one inside each core and one in between the cores. The in-core temps will always be hotter than the between-core temp. Speedfan reads all three. TAT and CoreTemp show the in-core temps. For a full discussion of Core 2 Duo temps see:
http://forumz.tomshardware.com...ide-ftopict221745.html
I?ve noticed a lot of people on the web talking about the really low E6750 core (Tjuntion) readings of under ambient temps (i.e. 18-19C) at idle from TAT, CoreTemp and SpeedFan. Just so you know?.This is bollocks. Yes, you get these readings but they are wrong. The Tjunction offset is incorrect and should be 100W not 85W so you need to offset the in-core temps +15C in SpeedFan to get proper readings. You can tell it?s wrong because the in-core temps are cooler then the between core reading. Now that?s out of the way, on to the good stuff.
Temperature readings I give below are taken from the max value observed in the between-core probe for a 10 minute TAT pass with 100% on both cores. I don?t need to run longer than that as I don?t get heat build up in my case due to good ventilation. If you want in-core temps, add 11C to 12C to the quoted in-between core temp.
OVERCLOCKING
Note ? A stable overclock to me is one that passes both a 10+ hour Blend (CPU and RAM) and 10+ hour CPU stress test in Orthos. I also like to run a few rounds of 3DMark06 in the background to test with an extra power load.
I could give you a detailed description of my early testing, but suffice to say: Set the FSB to 400, drop the memory multiplier to x2, stock voltage for everything, and hey presto... a stable 3.2GHz and it hasn?t even broken a sweat. It was the easiest 20% OV I?ve ever made in my life. Pushing higher than that fell into more usual OC technique. Lift the FSB with a dropped multiplier. Here's what I got:
FSB Overclocking
The upper FSB limit was 515 for me, but I can't tell if that's the memory or the motherboard. I'd say memory as it was pushing DDR 800 at 1030. Increasing chipset voltage, memory voltage or memory timings didn't make a difference.
Memory Overclocking
The Corsair 6400C4 could keep its 4-4-4-12 timings at 2.1V until I hit a 465 FSB. Then I dropped to 5-5-5-18 to climb to a 515 FSB.
Something I wasn?t used to coming from the 939 was the lack a memory divider below 2. This means that I couldn?t test the FSB independently of the memory as the mem was already past its rated 800 spec.
CPU Overclocking
Right now, I?m sitting at 3.4Ghz (425 FSB) with a CPU voltage of 1.3875 (stock voltage for a E6750 is 1.35). If I drop the voltage to 1.38125 (the next setting down) it?s not stable (see above for what I call stable?not these pansy 40 minute tests that people post).
Here are the temps:
E6750 @2.66 (1.35V) - 47C
E6750 @3.40 (1.3875V) - 55C
Can the CPU go higher? Absolutely?but it needs more voltage. To give you an idea, I?ve established that 1.425V isn?t enough to run stable at 3520 Mhz (440 FSB) so 3.4 is where I?m sitting now.
I?ve still got some headroom in temps alone without getting uncomfortable (~60 in-between core readings for TAT is my limit) and I have a few other things to do. My Ninja has heavy mill marks on the base that I?ve never got around to lapping so I?ve got the sandpaper, I just need to sit down and do it.
I also have the option of increasing the fan speed. I have four 120mm fans in my P180: one intake at the front (Low), one back outlet (Med), one top outet (Med) and the Nexus on the Ninja itself. Increasing the two outlet fans to high drops peak TAT temps by 2C but it?s very loud. Using a stronger fan than the Nexus would probably improve temps too.
SYSTEM PERFORMANCE & BENCHMARKS
Benchmarks (Vista) X2 4600+@ 2.6 E6750@2.66 E6750@3.4
3DMark06 3825 3850 3953
3DMark06 - CPU 1831 2308 2984
HL2: LC - 640x480 0xAA 8XAF 89.89 164.79 164.14
HL2: LC - 1024x768 0xAA 8XAF 87.1 134.13 137.1
HL2: LC - 1024x768 4xAA 8XAF 87.11 110.82 111.26
HL2: LC - 1280x1024 0xAA 8XAF 83.02 100.51 102.94
HL2: LC - 1280x1024 4xAA 8XAF 72.91 78.05 77.76
Sandra - Dhrystone 18909 24508 31271
Sandra - Whetstone 16047 17133 21861
Sandra - Integer 49084 147073 187642
Sandra - Floating Point 54139 80086 102180
Sup Com - Sim - 1280x1024 8301 8905 9093
Sup Com - Render - 1280x1024 4164 5112 5391
(Sorry about the crappy layout...This forum doesn't preserve spacing).
I?ve included the X2 4600+ benchmarks for comparison. Note that a 4600+ stock clock is 2.4GHz, but I thought I?d give a rough comparison of clock for clock with the stock C2D using 2.6 (though we are talking DDR vs. DDR2 here as well).
I was surprised by the SupCom numbers, I thought the Sim numbers would be better. I think that it indicates the GPU state of the game for me now. Interestingly, my G15 keyboard LCD screen allows me to monitor CPU usage on each core as well as mem usage. While SupCom is meant to be well threaded, most of the benchmark one core is at about 90-100% while the other is at 20-50%. This holds up with that review (it?s online somewhere) where they found a quad made very little difference to SupCom performance.
The HL2: Lost Coast number show that the X2 was completely CPU bound at anything under 4xAA at 1280x1024. The C2D numbers show that its now GPU bound (even at 640x480). Interestingly, the G15 monitor shows that the Source engine makes very good use of both cores most of the time (well balanced with 60-80% usage on both cores).
The Sandra scores speak for themselves. Enjoy.
__________________________________________________________________________
INTRODUCTION
I?ve just put together a new E6750 system and I thought I?d share my overclocking experiences for the public interest. This is the first time I?ve done something like this, but its is the sort of write up I?d like to see so here it goes.
Here are the new components (prices in Australian dollars):
- E6750 ($239)
- Gigabyte P35-DS4 ($269)
- Corsair TWIN2X1024-6400C4 (4-4-4-12 @2.1V) ($202)
Here is what I kept from my last system:
- P180
- Scythe Ninja (Rev. A) with Nexus 120mm
- G15 Keyboard
- Leadtek 7800 GT
- Tagan 420-U02 (22A +12V)
Here is what I upgraded from:
- DFI Lanparty UT Ultra-D 939 (just sold for $142)
- AMD Athlon X2 4600+ @ 2.6Ghz (just sold for $354)
- 1.5GB Corsair Value Select DDR PC3200 RAM (selling now?probably get around $80-$100)
As I side note, I?m running Vista Home Premium (I upgraded from Windows 2000) and as any silent PC aficionados may note, my old setup was designed to be fairly quiet.
THOUGHTS ON THE UPGRADE
The Q6600 was very tempting but in the end I decided to go the dual-core route. Why?
1) I probably would have required a new power supply as I?m near the limit of the 22A with the current setup and I didn?t want to cough up the extra cash.
2) As I?m primarily a gamer, the potential higher clock was more attractive and while there are games that are coming out soon that are truly multi-threaded (uses more than 2 cores) I?m not sure whether there will be a huge difference between dual-core and quad-core (e.g. the particle stuff demoed for HL engine isn?t coming in EP2) performance in the next 6-9 months. But I could be completely wrong?we?ll have to wait and see.
3) Cooling an overclocked 65nm quad core isn?t easy (even with a Ninja), especially when don?t want your PC to sound like a plane taking off.
4) With the P35 purchase, I can upgrade to a cooler 45nm Penryn quad core next year when games start to make more use of them.
5) I can use $100 I saved to help buy an 8800 or 8900 later this year.
I got the P35-DS4 for good passive cooling so it?s quiet, great overclocking, better stability, and to a small extent better resale value (don?t laugh?I just sold my DFI Lanparty UT Ultra-D 939 for $142 on eBay when I paid $199 two years ago). As I work from home a software developer I can tax deduct a high proportion of all this so the board effectively cost me ~$210.
I got the Corsair 6400 with tighter timings to deal with the main problem of OCing the E6750?the low multiplier.
SYSTEM SETUP
No problems setting up. The Scythe Ninja fit perfectly into the P35-DS4 and is sited close to the top back of the case so its gets lots of extra airflow from the top and back P180 exhaust fans. Here are some pics of the Ninja in the board:
http://img517.imageshack.us/im...njarevaongap35sce9.jpg
http://img339.imageshack.us/im...njarevaongap35szl7.jpg
Only tricky thing was stretching the ATX_12V_2X cable from the power supply on the bottom of the case to the top left of the motherboard. Definitely mount the CPU and cooler before you put the motherboard in the case. The DS4 set the RAM voltage at the standard 1.85V so I had to manually set it to the correct 2.1V designed for the RAM.
As the P35 only has one IDE connector, and I had three IDE hard-drives (and one SATA drive) and one IDE DVD in the previous system, I bought a SATA DVD and dumped a 60GB IDE hard drive. But the placement of the IDE connector on the bottom of the motherboard worked perfectly with the P180, as I placed the two IDE drives in the lower bay, reducing heat and increasing airflow by eliminating those nasty IDE cables from the air passage of the main case body.
MOTHERBOARD IMPRESSIONS
For those interested in the P35-DS4, my overall impression is a very good one. The build is great and the passive cooling system works brilliantly (especially in the P180) where my chipset temps don?t deviate from 40-41c. I?ve even bumped the chipset voltage bit during the OV process and it didn?t make any difference. The layout is very good with lots of room around the ram and the connectors are all in a good spot.
The only thing that stops me from raving about the board is the BIOS. I?ve just from the DFI Lanparty NF4 board and in comparison it needs a lot of work. System monitoring just gives you ?OK? rather than voltages running though the components. In the OCing options, you can only set voltage increments (i.e. +0.05 V) rather than actual voltages on everything but the CPU. This means you?ve got to know what the correct voltage should be (see above re the RAM). At least you can set the processor voltage as a number rather than an increment.
On the bright side, Gigabyte can always improve the BIOS UI. At least they got the hardware right?And it?s really right.
NOTES ON MONITORING
I?m using Orthos and TAT to load the processor and SpeedFan to monitor the temperatures. A quick rundown for those who don?t know - C2D have three temp probes: one inside each core and one in between the cores. The in-core temps will always be hotter than the between-core temp. Speedfan reads all three. TAT and CoreTemp show the in-core temps. For a full discussion of Core 2 Duo temps see:
http://forumz.tomshardware.com...ide-ftopict221745.html
I?ve noticed a lot of people on the web talking about the really low E6750 core (Tjuntion) readings of under ambient temps (i.e. 18-19C) at idle from TAT, CoreTemp and SpeedFan. Just so you know?.This is bollocks. Yes, you get these readings but they are wrong. The Tjunction offset is incorrect and should be 100W not 85W so you need to offset the in-core temps +15C in SpeedFan to get proper readings. You can tell it?s wrong because the in-core temps are cooler then the between core reading. Now that?s out of the way, on to the good stuff.
Temperature readings I give below are taken from the max value observed in the between-core probe for a 10 minute TAT pass with 100% on both cores. I don?t need to run longer than that as I don?t get heat build up in my case due to good ventilation. If you want in-core temps, add 11C to 12C to the quoted in-between core temp.
OVERCLOCKING
Note ? A stable overclock to me is one that passes both a 10+ hour Blend (CPU and RAM) and 10+ hour CPU stress test in Orthos. I also like to run a few rounds of 3DMark06 in the background to test with an extra power load.
I could give you a detailed description of my early testing, but suffice to say: Set the FSB to 400, drop the memory multiplier to x2, stock voltage for everything, and hey presto... a stable 3.2GHz and it hasn?t even broken a sweat. It was the easiest 20% OV I?ve ever made in my life. Pushing higher than that fell into more usual OC technique. Lift the FSB with a dropped multiplier. Here's what I got:
FSB Overclocking
The upper FSB limit was 515 for me, but I can't tell if that's the memory or the motherboard. I'd say memory as it was pushing DDR 800 at 1030. Increasing chipset voltage, memory voltage or memory timings didn't make a difference.
Memory Overclocking
The Corsair 6400C4 could keep its 4-4-4-12 timings at 2.1V until I hit a 465 FSB. Then I dropped to 5-5-5-18 to climb to a 515 FSB.
Something I wasn?t used to coming from the 939 was the lack a memory divider below 2. This means that I couldn?t test the FSB independently of the memory as the mem was already past its rated 800 spec.
CPU Overclocking
Right now, I?m sitting at 3.4Ghz (425 FSB) with a CPU voltage of 1.3875 (stock voltage for a E6750 is 1.35). If I drop the voltage to 1.38125 (the next setting down) it?s not stable (see above for what I call stable?not these pansy 40 minute tests that people post).
Here are the temps:
E6750 @2.66 (1.35V) - 47C
E6750 @3.40 (1.3875V) - 55C
Can the CPU go higher? Absolutely?but it needs more voltage. To give you an idea, I?ve established that 1.425V isn?t enough to run stable at 3520 Mhz (440 FSB) so 3.4 is where I?m sitting now.
I?ve still got some headroom in temps alone without getting uncomfortable (~60 in-between core readings for TAT is my limit) and I have a few other things to do. My Ninja has heavy mill marks on the base that I?ve never got around to lapping so I?ve got the sandpaper, I just need to sit down and do it.
I also have the option of increasing the fan speed. I have four 120mm fans in my P180: one intake at the front (Low), one back outlet (Med), one top outet (Med) and the Nexus on the Ninja itself. Increasing the two outlet fans to high drops peak TAT temps by 2C but it?s very loud. Using a stronger fan than the Nexus would probably improve temps too.
SYSTEM PERFORMANCE & BENCHMARKS
Benchmarks (Vista) X2 4600+@ 2.6 E6750@2.66 E6750@3.4
3DMark06 3825 3850 3953
3DMark06 - CPU 1831 2308 2984
HL2: LC - 640x480 0xAA 8XAF 89.89 164.79 164.14
HL2: LC - 1024x768 0xAA 8XAF 87.1 134.13 137.1
HL2: LC - 1024x768 4xAA 8XAF 87.11 110.82 111.26
HL2: LC - 1280x1024 0xAA 8XAF 83.02 100.51 102.94
HL2: LC - 1280x1024 4xAA 8XAF 72.91 78.05 77.76
Sandra - Dhrystone 18909 24508 31271
Sandra - Whetstone 16047 17133 21861
Sandra - Integer 49084 147073 187642
Sandra - Floating Point 54139 80086 102180
Sup Com - Sim - 1280x1024 8301 8905 9093
Sup Com - Render - 1280x1024 4164 5112 5391
(Sorry about the crappy layout...This forum doesn't preserve spacing).
I?ve included the X2 4600+ benchmarks for comparison. Note that a 4600+ stock clock is 2.4GHz, but I thought I?d give a rough comparison of clock for clock with the stock C2D using 2.6 (though we are talking DDR vs. DDR2 here as well).
I was surprised by the SupCom numbers, I thought the Sim numbers would be better. I think that it indicates the GPU state of the game for me now. Interestingly, my G15 keyboard LCD screen allows me to monitor CPU usage on each core as well as mem usage. While SupCom is meant to be well threaded, most of the benchmark one core is at about 90-100% while the other is at 20-50%. This holds up with that review (it?s online somewhere) where they found a quad made very little difference to SupCom performance.
The HL2: Lost Coast number show that the X2 was completely CPU bound at anything under 4xAA at 1280x1024. The C2D numbers show that its now GPU bound (even at 640x480). Interestingly, the G15 monitor shows that the Source engine makes very good use of both cores most of the time (well balanced with 60-80% usage on both cores).
The Sandra scores speak for themselves. Enjoy.
you have a nice cool fast chip ! its called AMD PWNER!
Facebook
Twitter