Venice Overclocking Issues w/ Neo4-F

[ToeJam13]

All,

I am having issues driving my Athlon64/3800 past a 7% overclock. I have read about 30 recent forum threads on the subject and have tried suggestions in them all with no luck. I am looking for a second opinion as to if I have reached the limit of my CPU or of there may be something else sinister at work.

Stock processor speed is 2400MHz
Max HTT freq mainboard will post with is 292MHz. (CPU-multi @ x4, MEM-spd @ HTT/2, HT @ x3)
Max MEM freq memory will pass stress with MemTest86 is 273MHz. (CPU-multi @ x4, MEM-spd @ HTT/1, HT @ x3, vRAM @ 2.7V)
Max CPU freq processor will pass stress with Prime95 L-FFTs is 2580MHz (CPU-multi @ x12, MEM-spd @ HTT/1, HT @ x3, HTT @ 215, vCore @ 1.593v)
Max CPU freq processor will pass stress with Prime95 L-FFTs (reduced MEM-spd) is 2784MHz (CPU-multi @ x12, MEM-spd @ HTT/2, HT @ x3, HTT @ 232, vCore @ 1.593v)

What I notice...
[*]vCore at both load and idle fluctuates wildly between 1.584V to 1.536V according to CPU-Z
[*]I can drive my vCore to 1.624V in BIOS, but it does not increase stability
[*]+12V at idle is 11.63V
[*]+12V at load is 11.58V
[*]CPU temp in OS while idle is 33C
[*]CPU temp in OS while under load is 50C
[*]PCI Express reads 100MHz
[*]Legacy PCI reads 33.33MHz

What I?ve been told?
The MSI Neo4 may not have a working PCI lock. Many people have suggested that what it?s reporting back to utilities is false. I?ve tried setting my PCIe bus from 100 to 101 to force the lock, but nothing so far. Several people have suggested that a broken PCI lock shows itself when systems fail to overclock past a 230MHz HTT. However, its not been stated if that failed lock would show during a post, or during a boot into Windows.

System:
[*]Athlon64/3800+ Venice (ADA3800BPBOX/CBBLE 0513CPBW)
[*]MSI Neo4-F (nForce4)
[*]Patriot PC3200 2-3-2 modules (512MB*2)
[*]Leadtek PX6600GT-THD (Geforce 6600GT, 128MB, PCIe)
[*]Silverstone Tech SST-ST460 (460W, Active PFC)
[*]Seagate Barracuda 7200.7 (200GB, ATA100, ST3200822A)
[*]NEW: Swiftech MCW6000-64 CPU waterblock
[*]NEW: Swiftech MCP 350 pump

 

[SNM]

PCI devices, so far as I can tell, aren't accessed until well after a post. Try shoving your CPU multiplier and mem dividers as low as they go and see how far you can crank the FSB and still load Windows.

 

[ToeJam13]

Hrm...

With a CPU multi of x4 and a MEM-spd at half of HTT-spd, I have my HTT currently set to 280MHz. Windows boots fine.

So I guess my Neo4-F isn't the problem since the PCI lock is working as expected for my video, USB, LAN and PATA devices. Sadly, that means that my Venice will not overclock past 7% reliably. That?s rather disappointing. I was hoping for at least 20%.

EDIT:
On a side note of thought, should my vCore numbers be jumping around as much as they are? Also, might a hotter +12VDC line squeeze more out or is this just wishful thinking?

 

[CheesePoofs]

It could be the board isn't a good overclocking board or that the CPU just won't run faster than that. What cooler are you using?

 

[ToeJam13]

Right now I am using the stock AMD retail cooler. My Swiftech waterblock is going on as soon as my radiator shows up in the mail.


I'm unsure as to the quality difference between the Neo4-F and Neo4-Platinum. Obvious chipset differences aside, the boards seem nearly identical.

The only thing I don't like when it comes to overclocking on this board is that the voltage selection is unusual. Rather than letting me go up to 1.7V in 0.05 increments, it will only allow me to go up to 1.45V and then apply a "bounus voltage" to it, starting at 3.33% and going up in increments of 3.33% up to 20% (effective 1.74 max).

 

[Zigrat]

I have a venice 3000+ and a Neo-4F. I am hitting 2400 MHZ which is more than enough for me so i am not pushing it further But I can ONLY do this with clock gen, not in bios because of the "219 bug", which apparently is in full effect for venice too.

Did you bump your ram voltage to 2.7 from 2.5? I recommend this.

 

[ToeJam13]

I don't think I'm suffering from the "219" bug. When I drop my MEM-div from HTT/1 to HTT/2, drop my HT-multi from x5 to x3 and drop my CPU-multi from x12 to x4, I can POST with my HTT set as high as 292MHz. WindowsXP will boot with my HTT set as high as 290MHz, and I have verified that my HTT is actually 290 with both CPU-Z and ClockGen.

When overclocking memory, 2.7V is usually what I use. When keeping my MEM locked to HTT and CPU-multi at x4, I can push my Patriot PC3200 2-3-2 memory and HTT as high as 273MHz using a 2.5-3-2-8-1T setting. At this speed, I can pass both MemTest86 and Prime95 "blend" tests for over an hour before I stop them.

When dropping my MEM-div from HTT/1 to HTT/2 and my HT-multi from x5 to x3, I can push my Venice to ~2784MHz [232*12] before things start getting upset. However, if I bump my MEM-div from HTT/2 to HTT/1.12, that top processor speed drops to ~2700 [230*12]. If I keep my memory locked to HTT, I'm lucky to get things stable beyond ~2580 [215*12].

What's also odd is that the board won't let me run my memory faster than HTT even though it has a HTT*1.08 and HTT*1.5 setting. When I choose either of those, it picks the slowest speed (HTT/2) instead.

If I try to work around the higher MEM-spd bugs by keeping my memory locked to HTT and drop my CPU-multi from x12 to x10, I can get my HTT up to about 240 before the BIOS goes beserk and brings up the NvRAID detection utility (which is disabled in BIOS) and gets stuck. So even though my MEM can handle 273MHz, the MSI mainboard will never let me utilize it. I can't get past the 240 HTT mark until I drop my CPU-multi down to x6. Blah!

So, I am thinking that I might be able to get more out of my processor and a LOT more out of my memory if I had a different mainboard. All of a sudden, DFI is starting to look good right now...

 

[ChakkaSol]

Hi ToeJam, welcome to the club because you have the bug...here is how to tell if you have it - it really should be called the 219 - 229 cold boot hypertransport bug. Here is some excepts from my original post 2 1/2 weeks ago:

http://forum.msi.com.tw/index.php?topic=80211.0

In this illustration (from me in the thread), I let the bios ver 3.1 autoset the memory timings. On each cold boot post, the bios autoset the vdimm at 2.70v and the memory timings were autoset at 2.5,3,3,8 at 1T (1T manually set).

1. fsb 218x9, htt 218x5, ok
2. fsb 219x9, htt 219x5, ok
3. fsb 220x9, htt 220x5, no cold boot post
4. fsb 220x9, htt 220x4, no cold boot post - this is the 219 bug...at higher fsb, it really starts to affect performance
5. fsb 220x9, htt 220x3, ok
6. fsb 221x9, htt 221x3, ok
7. fsb 222x9, htt 222x3, ok - highest stability level of my valueram at 1:1, prime95 and memtest stable 8 hours plus each

However, if I manually lower vdimm to 2.55v and manually raise the n4 chipset voltage to 1.55v, my rig cold boot posts 4 out of 5 times at 222x9 and 222x4...


And some more from me...

I just dont accept the arguements that this cold boot 219 - 229 hypertransport problem doesnt affect performance...it does...

My test scenario is setting the ref clock(fsb) to 250x9, ht frequency multiplier to 250x1 and memclock to 166. At this setting, I can cold boot my rig no problem. I did overclock my MSI 6600gt to 11% (555/1110) because if you use DOT, the average person can overclock to this level (using coolbits, I can get to 580/1160). Here are my test scores - I test twice just to be sure and cold boot post each time:

Test 1 - 250x9, 250x1, 166 - 3Dmark05=3748, PCmark04=4388
Test 2 - 250x9, 250x1, 166 - 3Dmark05=3744, PCmark04=4394

The change I made using clockgen is to initially set fsb to 200x9, ht freq mult to 200x4 and memclock to 166. After I cold boot post, I use clockgen to increase fsb to 250x9 and by default, the ht freq multiplier is at 250x4. Here are my test scores using clockgen:

Test 3 - 250x9, 250x4, 166 - 3Dmark05=3983, PCmark04=4531
Test 4 - 250x9, 250x4, 166 - 3Dmark05=3986, PCmark04=4529

As you can see in every test case, I take a performance HIT when I have to boot at ht freq multiplier at x1 versus using clockgen after you boot to windows xp (which freq mult is set at x4 at 250).

I just cant get over the fact that we have to accept lesser performance since I can get to 250x4 thru clockgen but not through the bios thru a cold boot. IMO this is a bios problem-period!!!!!!!!!! And I hope this is a bios problem because if its a problem with the motherboard, then were all SOL....


And finally what Wesley Fink and Kristopher Kubicki from Anandtech did when they confronted MSI at Computex this last week:

Anandtech has now formally announced the following on its website:

Unfortunately, we have recently discovered some performance issues when using the board with 90nm processors, so until MSI can release a BIOS update that addresses the issue, we would avoid the board.

Update: After this Guide was initially written, it came to light that the MSI boards are having problems with 90nm parts. The MSI Neo4 SLI was our original high-end pick, but we changed that out after the new information. Sorry for the confusion!

Will MSI care to notice this and get this bios fixed soon?

http://www.anandtech.com/guides/showdoc.aspx?i=2426&p=2

 

[ToeJam13]

Drats!

I was messing around with ClockGen tonight, and I noticed that when I cold boot with a HTT of 200 and try to push past ~216-219MHz with ClockGen, my system freezes. In the past, I've always tried using ClockGen after cold booting at my HTT at 220MHz or higher, and it never froze.

Oh well, thanks for the info. Sadly when the rest of my water cooling system arrives next week, it looks like it won't help any. Hopefully MSI will get this solved soon.

 

[bjc112]

Your PSU rails seem rather low as well..

 

[ToeJam13]

Yeah, I noticed that too.

I swapped out my Silverstone 460W supply for my spare Coolmax 400W supply. +12V rail went from 11.58V under load to 11.65V. Still seems low. Oh well.

 

[furballi]

My Enermax 495 V2.0 went from 12.23 to 12.06 when I increase CPU Vcore from 1.4V to 1.7V (full load in Prime95 with 3000 Winchester).

-use 5:6 memory divider.
-use 3x LDT multiplier.
-bump up the chipset voltage by two notches.

Use CPUz to double check your memory speed. You're okay if under 220 MHz. Also set memory timing to 2.5-3-3-8.

 

[KayKay]

hmm, this is not good news. luckily i have another 4 months or so before I pick up my parts. C'mon MSI fix the problem!

 

[ChakkaSol]

The Problem Gets Worse! Well, the thread was locked on the MSI forum boards and a few users have been banned regarding the 219 cold boot hypertransport bug - actually 2 threads on the issue. The mods refuse to acknowledge the problem!

http://forum.msi.com.tw/index.php?topic=80974.0
http://forum.msi.com.tw/index.php?topic=80211.240

A new thread was started a week ago on Rebelshaven for those interested in this 219 bug and any updates to the problem - THANK GOD FOR ANANDTECH FOR HELPING TO GET TO THE BOTTOM OF THIS PROBLEM WITH MSI!!!!!!!!!:

http://www.rhcf.com/sisubb/ultimate...c/21/242/3.html

 

[ToeJam13]

Okay, the bad news.

I just upgraded my mainboard with the new 1.5 BIOS revision that is supposed to fix several issues with Rev-E Venice/San Diego chips.

The results:
1) My system still becomes unstable beyond a HTT of 228MHz.
2) New Rev-E memory dividers (HTT*1.165, HTT*1.33, HTT*1.5) still cause the system to boot at lowest mem SPD (HTT*0.5)

In short, nothing with regards to overclocking ability seems to have changed.

UPDATE:
What has changed is that Cool & Quiet support is now functional. Not that it has anything to do with overclocking... its just a "good thing" to know.