Quad Core Overclocking and GTL Reference Voltage

[virtualrain]

Many believe that GTL Ref. Voltage is the #1 culprit in low quad-core overclocks on non P965 platforms.

Your article provides great insight into what it means and what it does, however, it raises even more questions...

Even after reading your article I'm still left wondering why P965 boards have much better success at overclocking quad-cores than other chipsets (of specific interest to me is 680i). Why is that?

Do the GTL Ref. Voltage settings in the DFI RD600 board actually allow it to catch up to a P965 in terms of quad-core overclocking? If not, what else is holding it back?

Besides GTL Ref. Voltage, what are chipset and motherboard manufacturers going to need to do to support better quad-core overclocking?

I look forward to your insights.

[Tony]

rememeber all cpu's react differently to GTL voltage tweaks, even individual cores respond differently.

Nico from OCZ has push up 50+fsb more on RD600 using the GTL tweaks, so there are major gains to be had if you tweak correctly

[eva2000]

Yes definitely on DFI ICFX3200 with GTL disabled 7x502FSB prime95 blend one core would fail immediately with my E6600 B1 cpu. Enabled GTL with default 48/48/128 and priming away

Using 12/22 official bios - high FSB clocks by dropping multi from 9x to 7x and clocking FSB up to 502FSB. My ICFX3200 is retail early sample, removed NB's white foam spacer, reapplied thin layer of AS5 under stock passive NB heatsink. Added 60x38mm Sunon 23.5cfm MagLev fan over the NB.

NB Load temp = 51-55C at 1.76v bios set NB volts in 31.8C room temps open desktop bench rig running Prime95 v25.1 blend.

Click image for full screenshot.

[NicoOCZ]

GTL + VTT are the key to higher FSBs with Kentsfield, no doubt about it. I did some testing tonight with my Xeon 3210. With the default GTL Values of 48/48/128, this was my maximum:



Now, with GTL increased to 166/166/166, I managed to get the FSB up to 400 from the previous 334. Even the P5B Deluxe, known as one of the best (if not the best) Kentsfield overclocking board, didn't get me more than 375 out of this Kenny:



I'm still finetuning the voltages, this is just a first impression of what GTL voltage is able to do to your Kentsfield

Besides the Xeon 3210 and the DFI RD600, I'm using 2GB of OCZ PC2-8000 @ 503 4-4-4-12 1T 2.2V, Geforce 8800GTX, OCZ Evostream 600W, CPU and NB watercooled.

[freecableguy]

Looking at Figure 1 from my original GTL reference voltage article, it's not too hard to see why Kentsfield responds so well to GTL adjustments. Because Cores 0/1 and Cores 2/3 communicate via the FSB there is definately a lot of gain to be had by tuning for data transfer over the bus. I'll be exploring this subject in more detail soon enough.

[eva2000]

Kris, bit-tech.net DFI ICFX3200 article touched on GTL ref volts but guess they mis-read some stuff can you explain it better from post #17 and #18 http://forums.bit-tech.net/showthread.php?t=127788

Also seems part of the reason for lower bandwidth/slow pi times is the independent buses as explained at http://www.bit-tech.net/hardware/200...00_t2rg/8.html

Compared to other boards we see that the RD600 chipset has a large latency problem due to the fact that both buses are independent from each other. This means the data has to navigate from one to the other at a complete different clock speed and the chipset has to negotiate all of this in a manner that causes the right data to arrive at the right time.

Linked memory buses that run with set ratios allow this to be done easier, but unlinked / asymmetric buses require the core logic to work harder. The difference between NVIDIA's nForce 680i linked and unlinked modes are only about 150MB/s, but the unlinked mode doesn't make both buses truly independent like on the ATI RD600.

Anyway to adjust in bios the different clock speeds that the 2 independent buses communicate at to get better bandwidth ?

[thorgal]

Seems I really need to get into this, for my bad axe 2 doesn't allow more than 330 FSB either with my Kents B0, whereas P5B-del also gets me around 375-380. Unfortunately I don't have an RD600 sample (yet), so that means I'll have to mod or wait

Edit : Kris, I am looking forward to your results from Bad Axe 2 on this

[dearthz]

Originally Posted by thorgal

Seems I really need to get into this, for my bad axe 2 doesn't allow more than 330 FSB either with my Kents B0, whereas P5B-del also gets me around 375-380. Unfortunately I don't have an RD600 sample (yet), so that means I'll have to mod or wait

Edit : Kris, I am looking forward to your results from Bad Axe 2 on this

i wonder whether it is a BIOs issue or a chipet issue than limit 975 based mobo to be struggling to hit 400 fsb with kentsfield?

[Tony]

Its GTL ref voltages....every cpu responds differently so you have to tune the board to suit the cpu

[NicoOCZ]

My best result so far, not (yet?) primestable, but good enough for a 32M shot:



I used GTL 176/176/176 here, also had to play with the FSB I/O strengths to get more stability. Once and if I manage to get this primestable, I'll post the detailed settings.

Tho, what Tony says is absolutely right. I know at least two QX6700ES that don't react to my GTL Settings like my X3210 does.