Uncore Frequency Testing - preliminary findings

As to uncore testing, i only really did max uncore extensive test with W3520 D0 on DFI UT X58 around Uncore of 4600-4760mhz for 32M pi-validations. All only i7 920 C0/D0 cpus on DFI UT X58-T3EH8, I didn't push uncore past 4200-4400mhz. On P6T6, only used my poorest/worst i7 920 3836A756 C0 #2 cpu and that easily pushed to 4200mhz uncore for 32M and 4401mhz uncore for validation. This is all on water cooled cpu. So pretty nice the differences could be due to C0 vs D0 stepping or added NF200 on P6T6 ?.

P6T6 though gains performance in 32M pi up to 4200mhz. At 4300+ uncore 32M pi loops were slower than at 4200 and below uncore, so suggests RTL latency etc was too loose at 4300+ uncore. Still more investigating need

So for you viewing pleasure

System:

• i7 920 C0 3836A756 #2
• EK Supreme LT 1366 / FESER 480 / MCP655
• Asus P6T6 WS Revolution
• OCZ Platinum DDR3-1333Mhz 7-7-7-24 (OCZ3P1333LVA6GK 'Animal Series')
• 512MB Inno3D 8800GT OC Edition @742/1836/2000
• 750GB Samsung SATA
• Pioneer DVR-217 DVD-RW
• Corsair HX1000 PSU
• WinXP Pro SP3 (nLite fully updated) with Vista Theme

Super Pi 32M
Uncore Comparisons: 16x vs 18x vs 19x vs 20x vs 21x

Testing uncore frequency on i7 920 3836A756 C0 #2 cpu with P6T6 WS Revolution 0601 beta bios and pretty much the same key voltages needed for 32M Pi stability as with DFI UT X58-T3EH8 tests with W3520 D0 cpu - CPU VTT and VDIMM - as you push Uncore frequency higher you need more CPU VTT uncore volts (QPI/DRAM) and VDIMM



Max screenshot uncore for now is 4401Mhz

Everest Ultimate
Uncore and RTL A/B/C value comparisons

RTL A/B/C = 51/54/54


RTL A/B/C = 61/62/63


RTL A/B/C = 64/66/66


RTL A/B/C = 66/66/67


RTL A/B/C = 69/72/72

i7 920 3836A756 C0 #2 @4493Mhz HT Enabled

Managed 25Mhz clocks higher HT enabled Super Pi 32M clock than what I could manage with exact same cooling on DFI UT X58-T3EH8.

• Asus P6T6 WS Revolution HT Enabled 32M Pi @1.488v = 4493Mhz
• DFI UT X58-T3EH8 HT Enabled 32M Pi @1.488v = 4468Mhz

Click image for full screen shot

@4493Mhz


@4468Mhz


Some 3d benchmarking





System:

• i7 920 C0 3836A756 #2
• EK Supreme LT 1366 / FESER 480 / MCP655
• Asus P6T6 WS Revolution
• OCZ Platinum DDR3-1333Mhz 7-7-7-24 (OCZ3P1333LVA6GK 'Animal Series')
• 512MB Inno3D 8800GT OC Edition @742/1836/2000
• 750GB Samsung SATA
• Pioneer DVR-217 DVD-RW
• Corsair HX1000 PSU
• WinXP Pro SP3 (nLite fully updated) with Vista Theme

Just bought this cpu 3rd hand, nice clocker apparently - Xeon W3540 3845B010

On Asus P6T6 WS Revolution with 0007 beta bios. Default AUTO vcore in bios reports = 1.088v. Manually need to set vcore to 1.10625v to get 1.088v vcore reported in bios.


Very promising start indeed using 4Ghz Prime95 benchmark as a guide to a cpu's potential. This Xeon W3540 3845B010 gem batch managed to do 4Ghz at 1.160v idle / 1.168v load with 52-57C load temps which is the best LGA1366 cpu i have experienced just edging out my Xeon W3520 3845A935 gem batch which did 4Ghz at 1.184v load with 57-60C load temps with exact same cooling EK Supreme LT / TFC480 / MCP655 H20.

System:

• Intel Xeon W3540 3845B010
• EK Supreme LT 1366 / FESER 480 / MCP655
• Asus P6T6 WS Revolution
• OCZ Platinum DDR3-1333Mhz 7-7-7-24 (OCZ3P1333LVA6GK 'Animal Series')
• 512MB Inno3D 8800GT OC Edition 190.62WHQL
• 750GB Samsung SATA
• Pioneer DVR-217 DVD-RW
• Corsair HX1000 PSU
• WinXP Pro SP3 (nLite fully updated) with Vista Theme

Xeon W3540 3845B010
vs
Xeon W3520 3845A935

Promising start with oc'ing 4378Mhz at 1.272v and 4487mhz at 1.320v for Super Pi 32M.

@4378Mhz at 1.272v

@4487Mhz at 1.320v

@4593Mhz at 1.424v

@4637Mhz at 1.472v

Seem to have a problem with booting into windows at >4600Mhz HT enabled but seems okay booting under 4600Mhz and then using either TurboV or setFSB to bump BCLK and cpu clock up. Going to try other cpu multipliers to make sure.

Max Validation @4821Mhz at 1.488v

• Max CPUZ Screenshot = 4681Mhz at 1.472v
• Max CPUZ Validation = 4821Mhz at 1.488v

With 6GB Corsair Dominator GT 2000 7-8-7-20 kit of memory I never got to try looser 8-9-8-x timings much with DFI UT X58-T3EH8. So now on Asus P6T6 WS Revolution, will try 8-9-8-22 2T timings to see how much further I can push memory.

System:

• Intel Xeon W3540 3845B010
• EK Supreme LT 1366 / FESER 480 / MCP655
• Asus P6T6 WS Revolution
• 6GB Corsair Dominator GT 2000 7-8-7-20
• 512MB Inno3D 8800GT OC Edition 190.62WHQL
• 750GB Samsung SATA
• Pioneer DVR-217 DVD-RW
• Corsair HX1000 PSU
• WinXP Pro SP3 (nLite fully updated) with Vista Theme

Max Validation @DDR3-2280Mhz 8-9-8-22 2T at 1.90v

Used TurboV to push mem to DDR3-2320Mhz but couldn't save the CPUZ Validation files heh. Will try again for next session

QPI/DRAM (Uncore = CPU VTT) = 1.650v
Uncore frequency = 4560Mhz

Max Super Pi 1M @DDR3-2268Mhz 8-9-8-22 2T at 1.90v

QPI/DRAM (Uncore = CPU VTT) = 1.6375v

Max Super Pi 32M @DDR3-2220Mhz 8-9-8-22 2T at 1.90v

QPI/DRAM (Uncore = CPU VTT) = 1.6375v

Now trying to combine max memory overclocks with max cpu clocks with HT disabled, but seems I'm being held back by possibly 2 factors

1. Xeon W3540 3845B010 cpu temps which at 4660Mhz at 1.528-1.544v was already idling at same temps of 46-52C as my Super Pi 32M run with Xeon W3520 3845A935 was @4794Mhz at 1.552v with exact same H20 cooling.
2. Memory clocks didn't seem to be as stable as tested at lower cpu clocks, probably due to cpu heat affecting memory stability ? Going to work on the voltages etc to see.

In meantime, here's what I could pull off @4660Mhz HT Disabled 32M Pi at 1.528v and validation @4812Mhz at 1.544v (limited by mem clock).

System:

• Intel Xeon W3540 3845B010
• EK Supreme LT 1366 / FESER 480 / MCP655
• Asus P6T6 WS Revolution
• 6GB Corsair Dominator GT 2000 7-8-7-20
• 512MB Inno3D 8800GT OC Edition 190.62WHQL
• 750GB Samsung SATA
• Pioneer DVR-217 DVD-RW
• Corsair HX1000 PSU
• WinXP Pro SP3 (nLite fully updated) with Vista Theme

Max Validation

Limited by mem stability/clock due to cpu heat ?

Max Super Pi 32M @4660Mhz at 1.528v

Took a side by side screenshot of my 12x mem multi DDR3-2184Mhz 7-8-7-20 1T 32M run versus my 10x mem multi higher cpu clocked DDR3-2118Mhz 7-8-7-20 1T 32M run to see what cputweaker reported for my subtimings. See if any subtiming settings are a factor.



DDR3-2184Mhz (12x mem) vs DDR3-2118Mhz (10x mem)

• QPI/DRAM volts = 1.55v vs 1.6375v due to lower bclk needed on 12x mem config
• tRFC = 110 vs 88 could be a factor ?
• tRRD = 7 vs 7 both same
• tREFI = 955 vs 849
• tFAW = 37 vs 33
• tRTL A channel = 56 vs 57
• tWTP = 27 vs 26
• tRTP = 7 vs 6
• Write to Read (same rank) = 20 vs 19

Going to play with subtimings and see

Update: Dropped cpu multiplier to 19x and using 10x memory multiplier, will try to reproduce similar looser subtimings as used in my DDR3-2184Mhz 7-8-7-20 1T run with 12x memory multiplier. I managed to get as far as Super Pi 32M stable @DDR3-2188Mhz 7-8-7-20 1T at 1.84v vdimm

Manually set following subtiming values in bios:

• tRFC = 110
• tFAW = 37
• Back to Back CAS# Delay = 4 (tightened to try and offset loosened tRFC value)
• Write to Read (same rank) = 20

Super Pi 32M passed at:

• DDR3-2158Mhz 7-8-7-20 1T at 1.72v
• DDR3-2167Mhz 7-8-7-20 1T at 1.78v
• DDR3-2177Mhz 7-8-7-20 1T at 1.80v
• DDR3-2188Mhz 7-8-7-20 1T at 1.84v

DDR3-2158Mhz 7-8-7-20 1T at 1.72v

DDR3-2167Mhz 7-8-7-20 1T at 1.78v

DDR3-2177Mhz 7-8-7-20 1T at 1.80v

DDR3-2188Mhz 7-8-7-20 1T at 1.84v

Xeon W3540 HT Disabled
@4702Mhz - 4759Mhz at 1.568v

Was worth the effort in figuring out and applying 12x memory multiplier's subtimings for Corsair Dominator GT with 10x memory multiplier configuration. With manually set subtimings applied to 10x memory multiplier config, was able to get memory memtest86+ test #5 10x loop stable which was enough for me to pass Super Pi 32M @4702Mhz - 22x 214bclk with DDR3-2136Mhz 7-8-7-20 1T at 1.74v. That's a nice improvement over max 32M Pi @4660Mhz with DDR3-2118Mhz 7-8-7-20 1T at 1.84v

Super Pi 1M @4759Mhz

Super Pi 1M & 32M @4702Mhz

Swapped out single 8800GT for 2x HD4870x2 CrossfireX. I flashed back my HD4870x2 cards from Asus HD4870x2 TOP bios to their original ATI reference HD4870x2 bioses (plan to sell these cards soon). Full 6x2GB = 12GB DDR3 configuration with Windows 7 Ultimate 64bit and Catalyst 9.8 Hotfix drivers.

System:

• Intel Xeon W3540 3845B010
• EK Supreme LT 1366 / FESER 480 / MCP655
• Asus P6T6 WS Revolution
• 6GB Corsair Dominator GT 2000 7-8-7-20
• OCZ Platinum DDR3-1333Mhz 7-7-7-24 (OCZ3P1333LVA6GK 'Animal Series')
• 2x HD4870x2 CrossfireX (HIS + Gigabyte) + CAT 9.8 Hotfix
• 750GB Samsung SATA
• Pioneer DVR-217 DVD-RW
• Corsair HX1000 PSU
• Dual Boot - WinXP Pro SP3 (nLite fully updated) with Vista Theme & Win7 Ultimate 64bit RTM

W3540 @4550Mhz at 1.472v

W3540 @4400Mhz at 1.384v

W3540 @4000Mhz at 1.184v

My new personal best 3dmark06 score ever = 31,812

3dmark06 = 31,812
http://service.futuremark.com/compare?3dm06=11910938



System:

• Intel Xeon W3540 3845B010
• EK Supreme LT 1366 / FESER 480 / MCP655
• Asus P6T6 WS Revolution
• 6GB Corsair Dominator GT 2000 7-8-7-20
• OCZ Platinum DDR3-1333Mhz 7-7-7-24 (OCZ3P1333LVA6GK 'Animal Series')
• 2x HD4870x2 CrossfireX (HIS + Gigabyte) + CAT 9.8 Hotfix
• 750GB Samsung SATA
• Pioneer DVR-217 DVD-RW
• Corsair HX1000 PSU
• Dual Boot - WinXP Pro SP3 (nLite fully updated) with Vista Theme & Win7 Ultimate 64bit RTM

Pushing it further with Corsair Dominator GT with 6-7-6-18 timings Pretty tricky getting higher due to fluctuating 3dmark06 cpu scores between runs heh



Finally broke 32k on water cooled cpu and stock air cooled HD4870x2 GPUs!



New personal best for Aquamark3 as well HD4870x2 CF @865/935 = 337,153 !



System:

• Intel Xeon W3540 3845B010
• EK Supreme LT 1366 / FESER 480 / MCP655
• Asus P6T6 WS Revolution
• 6GB Corsair Dominator GT 2000 7-8-7-20
• 2x HD4870x2 CrossfireX (HIS + Gigabyte) + CAT 9.8 Hotfix
• 750GB Samsung SATA
• Pioneer DVR-217 DVD-RW
• Corsair HX1000 PSU
• Dual Boot - WinXP Pro SP3 (nLite fully updated) with Vista Theme & Win7 Ultimate 64bit RTM