[Ret Sticky]Overclocking sndbx for A64 939 systems with Winchester, Opteron dual core

Without doing further research on the internet, this is from a post in the sticky
A64 CPUs, chipsets, motherboards
in case you have not seen it.

Originally posted by hitechjb1
About Rev E and SSE3 instructions

Rumor is a new revision E 90 nm 939 (Venice ?) with SSE3 support is coming (2005 ?).
http://anandtech.com/mb/showdoc.aspx?i=2264&p=3


From this article which reported AMD's chief Hammer Architect's (Kevin McGrath) presentation, at Stanford University
http://techreport.com/onearticle.x/6363

from above article said:

...
The enhancements include power reductions (gained by using slow but less leaky transistors in non-critical paths) and speed improvements (by using fast but leaky transistors in critical paths). Also, the processor halt and stopclock states have been improved, reducing some unnecessary work previously conducted during these states, resulting in a savings of several hundred milliwatts. Like the Pentium 4, future Hammer chips will feature on-die thermal throttling to cool themselves down if certain temperature limits are reached.

Performance-wise, the big news is the addition of SSE3 instructions, which accelerate a number of different types of computation, including video encoding, scientific computing, and software graphics vertex shaders. (For more on SSE3, see our Prescott review.) Beyond SSE3, the updated Hammer core will convert the LEA instruction, under certain circumstances, into an ADD instruction, which has only a single cycle of latency. AMD's design mavens have also added additional write-combining buffers to the chip, so it can combine up to four streams of non-cacheable writes, up from two. Hammer's data prefetch has been improved, as well.

...

Click to expand...

Rumor is there will be additional metal layers added for the E core. In general terms, extra metal layers are used to
- improve power distribution, i.e. less voltage drop for a given current or more current to more devices
- improve connectivity to package I/O, i.e. more package pins
- reduce signal RC delay, hence potentially higher clock frequency
- reduce clock skew, hence potentially higher clock frequency
- provide more flexible communication paths between functional units and multiple cores
- improve signal integrity
...
So the net is to enable more complex architecture and logic functions, multiple cores, higher socket pin count, higher clock, more devices, ....

About SSE3 Instructions:
http://www.intel.com/technology/itj/2004/volume08issue01/art01_microarchitecture/p06_sse.htm

Click to expand...

Thank you c627627 and hitech, that is exactly what I was looking for - knew it existed, but assumed there was more than one person who could show me exactly where to look... Nevermind the fact that I know more people should have been thinking the same thing I was thinking so its good to bring that info into this thread also.

im still curious about the HT bus...

lets say that i buy the 3000 and my memory (a-data hyperram pc4200) tops out at their default settings: 266MHz Bus and 3 4 4 8 timings... with a 3x multiplier id be left out with a crap HT bus of 798MHz... would that hinder performance? or what if i try 4x thus 1064MHz HT, would such a neglible HT oc affect stability?

im just watching at your excelent a64 3000 and thinking about saving the money of a 3200, then if my ram doesnt cut it i might get some ocz vx.

also would ZZF carry out a descent a64 3000 stepping?

your thoughts/suggestions anybody?

PhobMX said:

im still curious about the HT bus...

lets say that i buy the 3000 and my memory (a-data hyperram pc4200) tops out at their default settings: 266MHz Bus and 3 4 4 8 timings... with a 3x multiplier id be left out with a crap HT bus of 798MHz... would that hinder performance? or what if i try 4x thus 1064MHz HT, would such a neglible HT oc affect stability?

im just watching at your excelent a64 3000 and thinking about saving the money of a 3200, then if my ram doesnt cut it i might get some ocz vx.

also would ZZF carry out a descent a64 3000 stepping?

your thoughts/suggestions anybody?

Click to expand...

Well I wouldn't worry about underclocking the HT bus since its theoretical bandwidth is so high that it is almost impossible to saturate it, thus underclocking it has very little performance effect. If you decide to go 3000+ route, then I would say that what you see here is the exception not the rule. If you have nice TCCD ram (VX won't go 300+ htt, although it will do about 270 with really tight timings), then 3000+ is fine since 300x9 still gives you about 2.7ghz. I got a 3200+ retail from ZZF a week ago and it was 0451 so they do have fairly recent steppings.

PhobMX said:

im still curious about the HT bus...

lets say that i buy the 3000 and my memory (a-data hyperram pc4200) tops out at their default settings: 266MHz Bus and 3 4 4 8 timings... with a 3x multiplier id be left out with a crap HT bus of 798MHz... would that hinder performance? or what if i try 4x thus 1064MHz HT, would such a neglible HT oc affect stability?

im just watching at your excelent a64 3000 and thinking about saving the money of a 3200, then if my ram doesnt cut it i might get some ocz vx.

also would ZZF carry out a descent a64 3000 stepping?

your thoughts/suggestions anybody?

Click to expand...

The 939 HT system bus has a high bandwidth rated at 8 GB/s, whereas the PCI-e x16 video bandwidth is 5 GB/s.

If the HTT tops at 266 MHz (your concern), the 798 MHz HT bus frequency and 798 x 2 MT/s (million transfer/s) should give a bandwidth of around 6.4 GB/s between the CPU and the NF4 chipset. About 80% of the nominal bandwidth for 939 which is far from perfect.

It would be better if the LDT_multiplier of x4 can be used giving 1064 MHz which is only slightly above the 1000 MHz spec. I think probably it should be OK.

The effect of HT bus on video performance has to be studied. HT bus should not affect CPU and memory performance as they are in separate data paths.

If indeed the low HT (80% of spec) poses video performance issue, either the HTT has to be higher (by overclocking the memory) to use x3, or lower the HTT (hence lower the memory bus also) to use x4. Anyway, there is no much you can do until you test out the video performance with the memory, HTT and HT in combination after setting up the new system.


Further, with a low HTT and memory bus in the mid 200 MHz, a 3200+ with x10 is preferred to give more setup flexibility. A 3000+ can be used if the HTT and memory can reach 280 - 300 MHz range.

none of the cards currently can max out the AGP bandwith yet, so i wouldn't worry about that

I have a question.

That 92 mm Enermax fan you list is the very same one that we discovered has a major down side of shutting down Socket A systems when the adjustable fan setting is set too low? Does it do the same for this mobo? Even if it does not, warning should be posted for people to stay away from it...


since they are being directed to the setup you're using... and also may think Antec is the top choice above Fortron and OCZ power supply models you listed and it isn't (right?) so maybe that should be accentuated as well.

Didn't he mention in the very start of the thread that he had a boot/post problem, due to the fan shutting down:

2. When the CPU fan is connected to the CPU fan header, the bios has an auto speed control that the fan would not spin if CPU temperature is below certain temperature (25 C). When the system was first started, the CPU fan span momentarily and stopped while the GPU fan and maybe also the chipset fan were still ON, I immediately shut down the system. In combination with the two broken switches, I thought the motherboard was DOA. After a while, I figured out that it was due to the auto fan control, and eventually the system booted up without problem.

Click to expand...

I expect that with a different fan he would not have experienced this problem.

Thats a pretty serious issue though... Hopefully fixable with a BIOS upgrade.

Oh, its momentary? I thought it didn't spin at all. That momentary lapse is pretty common... This stopped you from posting??

Were you stopping and starting the machine??? or is it from cold boot?

Wow, just spent about 30 minutes reading through this whole thing and I have to say I am VERY impressed hitechjb1!! Thats an awesome overclock and your thread is extremely well organized and a pleasure to read.
Boy, I wanted a winchester before reading this thread, but now I REALLY want one, lol.

Thank you for sharing your overclocking results with us, I know I appreciate it!

c627627:
The 92 mm fan is a minor issue only when its fan speed knob position is set to too low. Further discussing this may create unnecessary confusion. That 92 mm is taken off the list.

IMOG:
It was not due to that 92 mm fan with speed adjustment. A fixed speed fan was used at that time. Further discussing this may create unnecessary confusion.

nicknomo:
It posts without any problem. The CPU fan may stop if auto-fan control is used and CPU temperature is low enough (say < 25 C).


About the CPU auto-fan control

When the CPU fan is connected to the CPU fan header and the motherboard controls the fan speed automatically based on CPU temperature. It is good this way for regular 24/7 usage as fan noise can be cut down significantly and I find that the system is much quieter unless the system is used for Prime95 the whole day .

It is actually a non-issue but rather an auto-fan speed control feature of the motherboard. It may be scary at the very first time since during posting the CPU fan stops after spinning momentarily if the CPU is below certain temperature (say < 25 C) at cold start. Chipset fan has that auto speed control feature too.

To bypass such feature, the fan can be connected directly to a molex connector and the fan speed wire (single wire) to the CPU fan header for fan speed monitoring. There should/may be a bios option to disable the auto fan speed control.


c627627:
I added the following to that post:

Originally posted by hitechjb1

It should be pointed out that the above hardware list is what was used to do the setup testing. They should be used as a reference only and result may vary due to the intrinsic electrical variation in each component even for the same part number.

The parts to achieve good result is not limited to the above list.
E.g.
- Other heat sink such as XP-120 can be used, other fans can be used.
- Other cooling compound such as Arctic Silver can be used, I use Arctic Ceramique due to its non-electrical property and ease for cleaning.
- The PSU list for Antec Neo Power 480W or Fortron Blue Storm 500W is not exclusive and the order does not imply preference.
- The exact model/manufacturer of the video card is not crucial for CPU, memory and motherboard overclocking. Choose one based on need and level of gaming. 6600/6800 can be configured to dual video card mode using NF4 Ultra/SLI motherboards, ATI cards may not.

CPU, memory and motherboard are the key components and the overclocking results may vary greatly, say 10-20% from what has been achieved. Based on recent results reported, they should have a good chance to achieve
- Winchester to 2.5 - 2.7+ GHz stable on air, ~ 1.55 V
- G. Skill TCCD 4400 to 280 - 310+ MHz, ~ 2.8 V
- DFI LP NF4 Ultra-D/SLI-DR to 300+ MHz on HTT

Click to expand...

hitechjb1 said:

HTT = 318 MHz
CPU = 2862 MHz (with x9)

memory_HTT_ratio = 1:1, memory = CPU / 9 = 318 MHz, efficiency = 81%
memory_HTT_ratio = 9:10, memory = CPU / 10 = 286 MHz, efficiency = 85%
memory_HTT_ratio = 5:6, memory = CPU / 11 = 260 MHz, efficiency = 90%
etc

Possible explanation:

When the CPU is clocked faster (consuming faster), the memory controller is not fast enough to keep pace and provide enough data I/O with the L2 cache running in sync with the processor clock, hence resulting in more cache wait states relative to the memory controller and in turn lower efficiency (the actual bandwidth is still higher, but efficiency which is bandwidth per memory clock is reduced).

Will the revision E0 correct/improve this?

Click to expand...

That's the first explanation I've seen for this and it sure seems like a good one. Increased latency, but not reduced raw bandwidth.


I always tended to think of the on-die controller as a seperate entity from the CPU, but it is in fact very dependent on the CPU speed it seems. Lots of difficult concepts to grasp with these architectural changes...it just never ends.

Nice work.

Hi guys,

This thread is very interesting, especially since I have the same MB on the way in the mail. My current OC on my 3500+ NC caps at 2530Mhz but I MUST get better.. (OCing is like crack). I plan on ebaying my CPU and getting a 90nm core processor when the DFI arrives. My goal is to hit 3000Mhz. I saw hitechjb1 got near that with a 3000 +, very impressive indeed. My question is which processor has the best chance of hitting 3000Mhz, 3000+, 3200+, 3400+, or 3500+? Looks like default clock doesnt mean much anymore these days so I am trying to figure out which will OC the best. I originally planned on getting a 3500 90nm >week46 but would like some opinions before purchasing so I don't make a mistake like I did with a 130nm core. I have decent water cooling for the CPU so temps shouldnt be an issue. Thanks in advance!

3 Ghz is a lofty goal...

I don't think you can reach that unless you get very lucky.

Eldonko said:

Hi guys,

This thread is very interesting, especially since I have the same MB on the way in the mail. My current OC on my 3500+ NC caps at 2530Mhz but I MUST get better.. (OCing is like crack). I plan on ebaying my CPU and getting a 90nm core processor when the DFI arrives. My goal is to hit 3000Mhz. I saw hitechjb1 got near that with a 3000 +, very impressive indeed. My question is which processor has the best chance of hitting 3000Mhz, 3000+, 3200+, 3400+, or 3500+? Looks like default clock doesnt mean much anymore these days so I am trying to figure out which will OC the best. I originally planned on getting a 3500 90nm >week46 but would like some opinions before purchasing so I don't make a mistake like I did with a 130nm core. I have decent water cooling for the CPU so temps shouldnt be an issue. Thanks in advance!

Click to expand...

To achieve 3000 MHz, realistic or not, let do some math first.

For 3000+,
HTT has to reach 3000 / 9 = 333.3 MHz (possible but not likely)
memory would run at 9:10 ratio, memory = CPU / 10 = 300 MHz (doable with G. Skill 4400 LE with luck)

For 3200+,
HTT has to reach 3000 / 10 = 300 MHz (this is more likely)
memory would run at 1:1 ratio, memory = CPU / 10 = 300 MHz (doable with G. Skill 4400 LE with some luck)
memory would run at 9:10 ratio, memory = CPU / 12 = 250 MHz (not good for performance)

For 3500+ running x11 multiplier (if using x10, would be same as 3200+),
HTT has to reach 3000 / 11 = 273 MHz (doable)
memory would run at 1:1 ratio, memory = CPU / 11 = 273 MHz (doable)

Assuming you can find a 3000 MHz CPU, you would need at least a 3200+ with some luck. 3500+ would be easier.


Currently, unless you can test some samples and pick CPU, for 90 nm Winchester 2700 MHz is a more realistic goal for air, maybe 2800 MHz for water. 3000 MHz is very unlikely, even just for booting into Windows.

For 2700-2800 MHz goal, if you are sure the motherboard and memory can handle 312 MHz stably, then a 3000+ would be OK. But if you are not sure about that, a 3200+ would be easier for setting up and tunning the system, as HTT and memory of 270 - 280 MHz would be sufficient.

To maximize the chance, I would try to get CBBHD 0447 or after (it is not a guarantee). CBBID is too new and I haven't seen much result as of today.

I picked a 3000+ since my goal was 2700 MHz and I "assumed" the DFI LP UT Ultra-D and G. Skill 4400 Le would do 300 MHz.


The underpinning argument is in the section "939 Winchester 3000+ vs 3200+" of this post or post 65 of this thread:
Low PR 90 nm 939 Winchester (Sept 2004)

Thanks for the reply.

I realize that in theory a 3500 should have the best chance to hit 3000Mhz because of the higher default clock speed and an option of 11 multi, but this may not be the case. OCing results for 3000+ and 3200+ 90nm CPUs seem to be much better percentage wise than 3500s. I was just wondering if it is worth the extra hundred or so bucks for a 3500+ if a 3200 may OC even further. I didn’t really “expect” to hit 3000Mhz, I am just hoping to. I plan on choosing my processor very carefully to improve my chances. The MB I will be using is a DFI LP UT Ultra-D and I have OCZ pc3200EL. In Wesley Fink’s review of the DFI boards the 3200EL hit > DDR600 so I am hoping the RAM can keep up.

Test your current 3500+ NewCastle with the DFI and memory first, then you can get an idea how well the board and memory can perform by using a lower CPU multiplier and pushing the HTT and memory frequency. After finding that out, you can then decide whether to keep the 3500+ NewCastle or not. If you decide to get a Winchester, those HTT and memory number will help you to decide between a 3000+ and a 3200+.

Since you already have a NewCastle 3500+, may well keep it and wait for Venice.

Yeah you are right. Sometimes money burns a hole in my pocket when new stoofs comes out heh heh. I will test the NC when my board arrives and post results. Thanks again.