AMD64 Overclocking Guide

Status
Not open for further replies.

Nubius

Golden Master
Messages
11,599
http://i4memory.com/showthread.php?t=327

That is a very thorough and up-to-date overclocking guide for the AMD64 computers......hopefully we can make this a sticky as kangaroos guide is important for the basics, however it's getting outdated for the AMD64 platforms.

Update - Another very well written guide can be found here, many people may find this one easier to follow as it includes numerous BIOS screenshots for first time navigators...thanks to F6Hawk for this contribution ;)

Another good site for learning the basic theory and process behind overclock from PZEROFGH
 
AMD Athlon 64 Core Explanation

Note: No processor is guranteed to do anything apart from operate safely at advertised specifications. Although some cores have reputations of high overclockability, chips of the same core vary based on speedbinning, fab location, and week and stepping numbers.

AMD Roadmap - Submitted by DJ-CHRIS

Update - It seems that the trend for all cores released by AMD are the same. The early cores produced during earlier weeks seem to yield better results than the same cores produced at a later date. I suspect that this is the result of AMD perfecting the production of the core over time, and testing them through tighter speedbins. If a new core is released, expect to yield better results when it is newer.

A common question which seems to be asked is the difference between the numerous cores which AMD offers on the Athlon 64 lineup. Each core offers a variation of improvements over its predicessor, and this will explain the relations between the current cores offered for Socket 939 Athlon 64s.

One of the large differences between AMD cores is the process size. The process size is measured in nanometers (1 nm = 0.000000001 meters). Smaller process size is in theory a superior core, simply because the circuits consume less power, and uses less current as it can charge faster.

Consider a 130nm core, and a 90nm core. Imagine the 90nm core is a 200mL cup, and the 130nm core is a 300mL cup. Not only can the 200mL cup fill faster, it fills up with far less liquid or voltage, therefore it uses less power and produces a smaller heat output.

The only problem with smaller process size is the fact that as transitors get smaller, the insulation also gets smaller and current can have a tendency to leak, or give out excess heat output. AMD has not reached this problem as of yet and their 90nm cores are still energy and heat conservative, however the Intel "Prescott" suffers from this problem.

AMD currently has five single cores availible to the socket 939 family; Newcastle, Clawhammer, Winchester, Venice, San Diego. AMD also offers multicores: Manchester and Toledo.

Newcastle
Newcastle is one of the older AMD Athlon 64 cores, being brought to the socket 939 family from socket 754. Newcastle chips are based on the 130nm process, and have the largest power consumption and heat dump overall next to the Clawhammer core. They've got the normal Athlon 64 cache size of 512KB.

Newcastle is availible in chips such as the 3000+, 3200+, 3500+, and 3800+. It's a dated core and as far as overclockability goes, don't expect much more than 400MHz on air. I really would stay away from this core as I don't think there's any difference in price between it and the more power conservative 90nm cores.

Clawhammer
This core is another one that migrated from the socket 754 family, and assumed the role of dealing with the high end processors in the family, such as 4000+ and FX series processors. It is also avalible in 3400+ format however. It's nearly identical to the Newcastle except it pumps out more heat, consumes more power, and has a larger 1MB L2 cache.

Like the Newcastle, this core has bitten the dust recently, and is outshown in both overclockability and overall performance by newer 90nm cores.

Winchester
This is the first native socket 939 core that AMD introduced, and is also the first core to boast the 90nm process size. The Winchester is near identical to the Newcastle, however it has a smaller heat dump and consumes less power. Again, a modest 512KB L2 cache.

These cores scored some big points in the overclocking catergory early on, you should expect about a 400-600MHz increase on these cores. Unfortunetely however, the 90nm process wasn't perfected and couldn't sustain higher clock speeds, and was only availible up to 3500+. A Winchester generally caps at around 2.6-2.8GHz since the silicon couldn't support higher clock speeds. The Winchester has been revised, and replaced with the "Venice".

Note: Some Motherboards require an updated BIOS in order to use 90nm cores.

Venice
The Venice is notorious for it's ridiculous power and voltage conservation, reaching clock speeds as high as 3GHz with little overall vcore increase. The Venice uses an improved intergrated memory controller which can account for all 4 DIMM slots running at 400MHz. Previous memory controllers would underclock the RAM to 333MHz if all four slots were in use.

The Venice specifically fixes the silicon problem found in the Winchester and can handle higher clock speeds. It's also the first AMD core to add SSE3 instruction sets, the third iteration of the SSE instruction set for the IA-32 architecture. It is a SIMD instruction set. If you are purchasing a midrange processor, get a Venice based chip.

Note: Some Motherboards require an updated BIOS in order to use 90nm cores.

San Diego
Once the silicon problem found within the Winchester was fixed, the 90nm process could support higher clock speeds. AMDs high end cores such as the 4000+ and FX series were stuck on the 130nm Clawhammer core since the 90nm silicon could not support them, however, using the Venice architechture, the San Diego, with a larger 1MB L2 cache, was born.

The San Diego is identical to the Venice, same instruction sets and memory controller, as well as same notorious overclockability. The introduction of the San Diego effectively migrated all Athlon 64 cores to 90nm process technologies. If you are buying a high end core, the San Diego is for you.

Note: Some Motherboards require an updated BIOS in order to use 90nm cores.

Manchester
This core is the lesser of the two multicores avalible, with a 512KB L2 cache on each of its cores. The Manchester consumes a lot of power and gives off quite a bit of heat despite running on a 90nm process, however, still achieves impressive overclocks ranging between 2.6-3GHz.

Note: Some motherboards may require a BIOS update in order to recognise both cores.

Toledo
Toledo, in theory, offers the fastest processor on the market at the moment, the 4800+. Toledo offers a 1MB L2 cache on each of its cores, and uses a 90nm process. It consumes more power than the Manchester and gives off quite a bit of heat dump, but still provides impressive overclockability.

Note: Some motherboards may require a BIOS update in order to recognise both cores.

To summarize, in my personal opinion the only two cores worth considering right now are the Venice, and the San Diego. The previous cores are outdated , and price is relatively the same when comparing to their predicessors.

As far as multicore processors go, multithreaded applications to take advantage of Toledo/Manchester are a rarity. Performance gains over single core processors are neglible, and the price is still too much for what they are offering. I suggest waiting 6-12 months before purchasing a multicore processor.
 
Socket 939 Opteron FAQ

Due to the tremendous increase in popularity with the new AMD Opteron processors availible in the socket 939 package, and the recent increase in repetively asked and answered questions, for your convenience we introduce the Opteron FAQ...with up to date stepping listings and tweaking tips

Opterwhat?
The Opteron is a strand from the K8 family of AMD processors, the most reknowned from this family being the Athlon 64. K8 includes Athlon 64, Athlon 64 FX, Athlon 64 X2 and Opteron processors, which all share the similar architecture of intergrated memory controllers...and as a result, incredible memory bandwidth and system performance via the Hypertransport bus

But what seperates the Opteron from other K8 processors? The original difference lay in the fact that the Opteron was seperated on the socket 940 package which used ECC memory, whereas the Athlon strand followed from socket 754 to 939 with unbuffered memory. Because of these differences, the Opteron was previously not considered for home PC owner use due to expensive ECC memory

However, AMD has since migrated the Opteron to the 939 package which no longer requires (or rather supports) ECC memory. This makes the Opteron a much more affordable and attractive option to small business enviroments looking for cheap workstations. However, since the Opteron is literally almost identical to other K8 architecture, the Opteron now became an option for the average PC owner

The Opterons are specifically extremely popular due to their high probability of achieving an above-average overclock, as well as the fact that all Opterons, even entry level ones comparable to 3000+ strand processors, feature 1MB of L2 Cache for much less than the entry level 3700+ San Diego. Opterons are generally considered to have a better price/performance ratio as an Opteron is generally cheaper than the Athlon it performs roughly the same as, this has been demonstrated as people are selling their previous Athlon cores to be replaced with Opterons

Opteron vs. Athlon 64
As mentioned, both the Opteron and Athlon 64 are derived of the same architecture, therefore they perform roughly the same as each other in a variety of enviroments. If this is the case, why still label it as an Opteron and not simply an Athlon

The large noticable difference between the Opteron and Athlon 64 is that the Opteron is built with the capability of running numerous HTT links at once, whereas the Athlon 64 only uses one. On a 939 platform, this feature is relatively useless as the extra HTT links are used to communicate between two or more Opterons running in a dual socket configuration, however since socket 939 can only support one simuteanously operating CPU the other HTT will never be used

The other difference between the Opteron and Athlon 64 is that it is usually agreed on that the Opteron is tested more intensely than the Athlon because the Opteron is intended to undergo a more intensive workload, therefore stability is a key concern in the production of the Opteron. This generally translates into "better" quality silicon on the Opteron which should in theory yield higher overclocks than an average Athlon. It is also speculated that the more intense testing leads to better memory controllers which will lead to better overclocking results and latencies on not only your processor, but your memory as well

Opteron Model Numbers
Unlike the Athlon strand of processors, the Opterons do not use a P-rating but rather a numbered system for determining a CPUs power. Below are the current 939 Opterons availible as well as the Athlon they roughly perform at the same level at

Single Core Processors
Opteron 144 - 1.8GHz - 1MB L2 Cache - Venus Core - 3200+
Opteron 146 - 2.0GHz - 1MB L2 Cache - Venus Core - 3500+
Opteron 148 - 2.2GHz - 1MB L2 Cache - Venus Core - 3700+
Opteron 150 - 2.4GHz - 1MB L2 Cache - Venus Core - 4000+
Opteron 152 - 2.6GHz - 1MB L2 Cache - Venus Core - FX-55*
Opteron 154 - 2.8GHz - 1MB L2 Cache - Venus Core - FX-57*

Dual Core Processors
Opteron 165 - 1.8GHz - 1MB L2 Cache - Denmark Core - 3800+
Opteron 170 - 2.0GHz - 1MB L2 Cache - Denmark Core - 4200+
Opteron 175 - 2.2GHz - 1MB L2 Cache - Denmark Core - 4400+
Opteron 180 - 2.4GHz - 1Mb L2 Cache - Denmark Core - 4800+

*Note - The Opteron 152 and 154 strand processors will perform roughly the same as the FX-55 and FX-57 processors respectively, however the Opterons do not have unlocked multipliers but the FX processors do

Opteron Stepping Numbers
Single Core Stepping

1. CABNE 05xx (excluding 0540)
2. CABYE 0540
3, CABNE 0540 / CABYE 05xx (excluding 0540)
4. CAB2E 05xx
5. CABGE 05xx

The golden weeks to look for in regards to the Single Opterons are 0528 and 0530, specifically from CABNE and CABYE stepping. The potential of the single opterons are generally in almost every case at least 2.8GHz with a good aftermarket cooler except in rare cases.

Note: It should be noted that the CAB2E stepping cores are known as to be either hit or miss, some are yielding overall good overclocks while others struggle beyond 2.75GHz

Dual Core Stepping Numbers
The only common stepping numbers from the Opteron dual core family are CCBWE and CCB1E. The CCB1E cores are almost always guranteed to hit 3GHz under the correct conditions, however they are cold bugged meaning if you wish to run LN2 or Phase Change or Dry Ice, you are limited to a CCBWE. The CCBWE stepping is hit or miss, there are several good weeks out there and no doubt there will be more in the future. Most Dual Opterons will do 2.6Ghz almost guranteed, however the 3GHz range is limited to (in order):

1. CCB1E 0550
2. CCBWE 0550
3. CCB1E 05**/CCBWE 0530

Opteron Temperatures
The Opteron strand of processors generally run hotter than their Athlon counterparts, with maximum acceptable temperatures on single core Opterons being between 65-71, whereas the dual cores are rated a maximum temperature of between 49 and 65. Generally good aftermarket air cooling should be appropriate for hitting 3GHz on the correct stepping, however water will generally yield better results with the astromical 3.4GHz results happening on phase change coolers

For those looking to dramatically cool down temperatures, one can remove the IHS from the core. The IHS on the Opterons do not make good contact between the core and the heatsink therefore thermal dissipation is degraded. Removing the IHS will void your processor warranty and increase the risk of damage therefore it is suggested the only those with experience go about removing it, however it is worth the trouble if you are willing to mod your heatsink and use a bit of patience

In addition, specifically on DFI boards, the dual core Opterons appear to giving incorrect temperature readings, although this may be the case on any other board. Many people are reporting multicore Opterons of the same vcore of single core Opterons to be running much cooler, which should not be the case. Since DFI and many other motherboard manufacturers do not officially support the 939 Opterons, it may be a while before this bug is fixed. It is recommended that you overclock the multicores very slowly, especially if you are on air, and it would be a good idea to use a third party device such as a temeperature probe to get a more accurate temperature reading

I am currently using BIOS version 704-2bta BETA by bigtoe and after crossreferencing with a probe I found the temperature reading for the core was fairly accurate. On stock cooling under load at 1.5vcore the core was reaching 50-55C under load which appeared normal.

Memory Controller Issues/Tips
The Opterons do not seem to like using memory dividers, specifically 180 and/or 166, therefore it is recommended that you try and maintain 1:1 ratio with your HTT/RAM...the memory controller yields excellent results with TCCD/BH-5 chips due to better silicon production therefore a 1:1 is a huge draw to the Opteron

It is also widely agreed that the Opteron struggles with both read preamble and max asynchronous latency found within A64 tweaker, therefore if you are having problems with RAM stability it is recommended you tweak those values. From my personal experience with both TCC5 and TCCD, when I was having issues getting past 300HTT with 1:1 memory, loosening max async lantency to 8ns and even 9ns yielded much more headroom with almost no bandwidth loss. Read preamble didn't seem to help my results much, however loosening it with BH-5 chips seems to help with memory frequencies especially at really high voltages (3.7v+).

Compatibility
The Opteron should be able to work in any existing socket 939 motherboard regardless of manufacturer and/or chipset. You will probably be best updating to the most recent BIOS version of your motherboard before using an Opteron in order to ensure you have no problems booting with or detecting your processor

It should be noted that AMD has mandated that only specified socket 939 motherboards officially support Opterons. Most consumer grade motherboards produced by companies such as DFI and MSI will not officially support Opterons meaning you will void your processor warranty and you will not recieve tech support from the company in question
 
Status
Not open for further replies.
Back
Top Bottom