Adding A Blue Orb & Removing BIOS Chip From The KT7
PC maintenance article - 10.8.2014
Many people are not content with the KT7's Northbridge fan. It's a brush based fan, which means it makes noise -particularly when it's cold (at bootup). Some also want to replace it to get better cooling. The most common way of doing this (for most chipsets) is to use a Blue Orb chipset cooler.
To start the mod, I removed the old fan. You have to be careful when taking it off, because the back of the board has a lot of leads which can be scratched easily. Just use a needle nose pliers to clamp the plastic and push it through the hole. The fan will then remove with ease.
Be careful not to scratch the leads on the back of the board
I was surprised to find that Abit actually used thermal compound (although a very small amount) between the chipset fan and the Northbridge. I didn't think there would be any kind of material there at all.
Not much thermal compound
Then I attached the Blue Orb. I soon found out that the Blue Orb does not fit on the KT7. drivers we have recently refreshed are
The stock heatsink/fan on the Northbridge is a tad longer than the Blue Orb. Because of this, the Blue Orb is short by about half a centimeter. I tried bending it and forcing it on the board for about thirty minutes to no avail. There is simply no possible way to get it to fit. Of course you can juse use thermal tape on the bottom without clamping the unit down, but this defeats the purpose of getting the added cooling of the Blorb.
I then tried to mount the Blue Orb's fan on the KT7's heatsink, since it was obvious the Blorb's fan was a tad better. This also was impossible to accomplish (without some serious modification). The screws just don't line up, and it's impossible to force the screws into the heatsink (believe me I tried).
Mounting the fan on the KT7's heatsink won't work
It is a good idea to take off the current heatsink and fan on the KT7's Northbridge. Abit didn't use very much thermal paste at all. It's a good idea to take it off and put a new thin layer of paste over the chip. It's possible this could further increase stability of the board, possibly allowing you to raise the FSB higher.
Since users have asked exactly how to remove the BIOS chip from the board, I have put together a short little guide. The BIOS chip is located on the bottom left of the board. The writing on the chip (not the sticker) reads from the LEFT corner of the board - remember this when you put it back in!
The chip is in it's socket rather securely. It's not possible to just "lift" the chip out with your fingers. You will need to use a flathead screwdriver to pry the chip off. You should NOT NEED TO USE MUCH FORCE AT ALL. If you get a lot of resistance, chances are you are trying to pry the BIOS chip's socket off the motherboard, and not the BIOS chip out of it's socket. The chip will pop right out. Be careful when prying because if you pry a little too far the end pins will start to bend. If this happens, just bend them back. It's obviously better to avoid the situation all together by simply prying the chip off from both ends evenly.
When you're ready to reinsert the chip just place it on the BIOS socket and push it down. It will take a little force, but again don't apply too much. If it isn't going in make sure all the pins are properly lined up. Like I mentioned above, make sure the writing on the chip reads from the left side of the motherboard, because it WILL fit in both ways, and it WILL fry if you put it in the wrong way.
If you need some more assistance with this or any other KT133 motherboard, check out out Abit KT7 FAQ. You might also check out our gigantic KT133 forum if you have any questions or concerns.
Choosing Memory For Your Athlon Motherboard
PC maintenance article - 6.8.2014
When Corsair Micro sent me two 128MB sticks of their new PC150 SDRAM to review I wondered how I could write yet another memory review and still keep the reader interested? After all, what can be said a stick of SDRAM that hasn't already been said so many times before? But looking at the posts on our forums here, in my own Inbox and on alt.comp.periphs.mainboard.abit, it is clear that what many people lack is an understanding of the benefits of quality RAM and how much they need. This article attempts to address this. In the process, we'll also discover what Corsair Micro's PC150 SDRAM is capable of...
Why worry about memory?
Many people are happy to concentrate on getting the latest and greatest processor, a GeForce 5 graphics card (review coming soon!) and a massive hard disk to store all their MP3 files. Memory - if it is considered at all - is thought of in terms of quantity rather than quality. RAM is RAM is RAM. Or is it?
In fact, poor quality RAM is probably the single biggest cause of unstable systems we see reported on the forums and newsgroups. It is almost guaranteed that a user suffering frequent system crashes will have "generic PC100 SDRAM" on their list of system components somewhere. Cheap RAM, along with small power supplies, are the two greatest false economies a system constructor can make. Good quality RAM, although more expensive, buys both stability and performance.
By the way - "generic" SDRAM refers to the anonymous DIMMs found in less specialist memory shops. They are generally poorly designed modules with dubious stability. Would you buy a processor or motherboard from an unknown company?!
To start with, we need to consider a little memory theory to understand the choices available to us.
When VIA released the Apollo KX133 chipset for the Slot A Athlons and then subsequently the Apollo KT133 and KT133A chipsets for the Socket A "Thunderbird" and Duron processors, the user was presented with a variety of memory options. A unique feature of these chipsets is the asynchronous memory bus, which means that the memory bus can run at a different speed to the front side bus (FSB). This means that whilst the FSB is running at 100MHz, the memory bus can run at either 100MHz or 133MHz through a simple BIOS setting. On the KT133A the FSB can run at either 100MHz or 133MHz, and so can the memory. The earlier KX133 chipset also supports a 66MHz memory bus. And to make matters more confusing, manufacturers such as Corsair Micro and Mushkin have recently announced 150MHz RAM. Can this be used on a 133MHz memory bus and is it worth the extra money?
Memory designed to run at 100MHz is called PC100 and conforms to a stringent standard defined by Intel. Memory designed to run at 133MHz is called PC133 and conforms to extension of the PC100 standard promoted by VIA. 150MHz SDRAM is occasionally referred to as PC150 SDRAM, but this is really a misnomer. There is no agreed standard for 150MHz SDRAM and in fact this RAM tends to be high quality PC133 SDRAM that is guaranteed to be stable at 150MHz.
An important consideration for SDRAM is the rate at which it can transfer data to and from the processor. This is not done directly, but through a controller chip known as the Northbridge, which on the KT133 chipset is the VT8363 controller (and VT8371 and VT8363A on the KX133 and KT133A respectively). This is illustrated in the figure below. As the memory bus is 64-bits (or 8 bytes) wide, the memory speed directly determines the theoretical peak rate at which data can be passed to and from the Northbridge. For 100MHz SDRAM this is 100M x 8 Bytes = 800 MB/s, for 133MHz SDRAM this is 1.064GB/s and for 150MHz SDRAM this is 1.2GB/s. The data is passed between the Northbridge and processor on the FSB, which uses Digital™ Alpha™'s EV6 double data rate (DDR) technology to achieve an effective rate of 200MHz on the KX133 and KT133, and 200MHz or 266MHz on the KT133A - also with a 64-bit bus. This means that the FSB supports a peak data rate of 1.6GB/s (or 2.1GB/s on the KT133A). The memory bus is clearly the theoretical bottleneck, even on the slower KX133/KT133 chipsets.
VIA KT133 block diagram (from VIA's website)
In practice, memory benchmarks show that is difficult to achieve an average data throughput that is even half the theoretical peak transfer rate on the memory bus. This due to the memory timings which are explained in some detail in Sharky Extreme's excellent Memory Guide Part 4, Part 1, Part 2 and Part 3 of the Memory Guides at 3Drage.and also RAM Guide: Part I DRAM and SRAM Basics and Part II: DRAM and SDRAM at Ars Technica. In simple terms, however, the memory cannot simply instantly provide data with each clock cycle. There is an initial "overhead" or latency associated with a request to read or write data from a given memory address, followed by the data appearing on each clock cycle. It is hard to describe these overheads without repeating the detailed descriptions of SDRAM operations in the references above, but if for simplicity we simply regard the memory as a table with rows and columns, then the principle delays are:
- CAS latency: the delay, in clock cycles, between issuing a read command and the availability of the first piece of output data
- RAS to CAS delay (tRCD): the delay necessary between defining the row and then defining the column in which the desired data lies
- RAS precharge delay (tRP): the delay necessary between finishing one read operation and beginning the next
Where CAS and RAS mean column address strobe and row address strobe respectively - the mechanism by which the "entries in the table" are addressed. When you buy SDRAM, these latencies will be mentioned on the datasheet - eg. PC133 x-y-z. PC133 3-2-2, for example, implies a CAS latency of 3 clock cycles, RAS-to-CAS delay of 2 clock cycles and a RAS precharge delay of 2 clock cycles. Although measured in clock cycles, the underlying reasons for these delays are due to the actual time taken for signals to settle within the SDRAM chip. The timings are therefore specified for a given memory speed, and will change with bus frequency. For example, RAM that has a CAS latency of 2 at 133MHz may require a CAS latency of 3 at 150MHz. It all depends on the quality of the RAM.
To summarise all this timing technobabble, the point to realise is that the performance of the memory depends on not just the "MHz" but also on how efficiently the RAM uses each clock cycle. The best SDRAM is typically rated as 2-2-2 at 133MHz, but you may find RAM with timings of 3-2-2 or slower. To make matters more confusing, suppliers often abbreviate these timings to just the first number - the CAS latency, eg. PC133 CL2 or PC133 CL3. It is up to you, as the buyer, to do your research! If a supplier doesn't provide this information, choose one who does.
What does this all mean to me?
Unless you are a computer geek, you don't really care what the MHz is or the meaning of the SDRAM timings. You want to know whether your computer will be more stable or faster, if you invest in some quality RAM. Will PC133 SDRAM make your computer 33% faster than PC100 SDRAM? How does the CAS latency affect things? How far can you overclock the RAM and how does this help? This is what this article attempts to address.
Memory Under Test: Corsair 150MHz SDRAM
The tests in this article were performed using Corsair's 150MHz SDRAM, part number CM654S128A-150. For those of you who like looking at pictures, here it is:
128MB Corsair 150MHz SDRAM
As the part number suggests, this memory module is actually based on Corsair's popular PC133 CAS2 module, but is guaranteed to run at 150MHz with a CAS Latency of 3. The module comprises eight 16MB SDRAM chips as shown above. What more is there to say? It looks like any other DIMM!
The rest of my system comprises an ABIT KT7-RAID motherboard, Thunderbird 1GHz CPU, Creative Labs GeForce Annihilator Pro 256 DDR graphics card and a Creative Labs Soundblaster 1024 Live! Player soundcard. I have two ATA/66 8GB Seagate disks configured as a RAID-0 array on the Highpoint controller. The motherboard is using BIOS WZb01.
The most popular memory benchmark in use today is the memory benchmark within the popular SiSoft Sandra software. This benchmark measures the sustained data rate of the memory using a respected algorithm called STREAM. Details of the Sandra benchmark are available at the Sandra memory benchmark FAQ. More information on the detailed STREAM algorithm are available at www.streambench.org although it should be noted that SiSoft have improved the benchmark somewhat in their implementation. Sandra reports two figures - the bandwidth achieved to the integer Arithmetic Logic Unit (ALU) of the processor and the bandwidth achieved to the Floating Point Unit (FPU).
A popular gaming benchmark is 3DMark2000 version 1.1 from MadOnion. I used this to see what impact memory would have on gaming. However, this benchmark gives a somewhat unscientific score with no units, and it is not at all clear how this figure is derived. Higher is "better", but that's about all I know.
Finally, continuing the gaming theme, I used the timedemo facility within Quake III to see what impact SDRAM would have on a real-life game. I didn't fiddle around trying to set optimum Quake III settings for the benchmark - sound was on, the resolution was set as 1024x768 and the graphics settings were all on high.
In retrospect I think 3DMark2000 and QuakeIII must be limited by the graphics card on my machine - both benchmarks are very graphics intensive and both showed very little difference from 100MHz CAS3 to 150MHz CAS2. The message here - which is important - is that you should always identify the bottleneck in your system! For gaming, mine is the graphics card not the CPU or memory!
PC100 versus PC133
The first question is whether there is any benefit to buying PC133 SDRAM over the cheaper PC100 SDRAM. What benefit does PC133 actually give? To test this I used the BIOS settings on the KT7-RAID to set the memory bus at 100MHz and to set the SDRAM timing to 3-3-3 (by setting the CAS latency as 3 and using the misnamed 8/10ns DRAM Bank Timing Setting, which uses a tRP=3, tRD=3 and tRAS=6 under the WZ BIOS). This gave the following Sandra score:
A mere 366/391 MB/s - some way short of the theoretical 800MB/s supported by 100MHz SDRAM! This nicely illustrates the issue of RAM timings described earlier.
If we increase the bus speed to 133MHz but leave the timings alone (ie. PC133 3-3-3), representing the low-end of the PC133 market, we increase the score to 441/517 MB/s. These represent increases of about 20% and 30% respectively in the throughput for the ALU and FPU. Not bad.
3DMark2000 scores increased from 4834 to 4900, and Quake III frames per second (FPS) from 80.5 to 82. Not so dramatic - but most likely due to a graphics card bottleneck as described earlier.
What couldn't be reproduced here is instability, as the test was simulating PC100 CAS3 and PC133 CAS3 RAM using the much better Corsair 150MHz SDRAM. However, anecdotal evidence from Icrontic's forums and Usenet suggests that PC100 CAS3 SDRAM is the cause of many instability problems on Athlon motherboards and really should be avoided at all costs. As SDRAM prices continue to fall, this should not be difficult.
CAS2 versus CAS3
The principal difference between "cheap" PC133 SDRAM and "quality" PC133 SDRAM are the timings at which they can run. We saw above that PC133 3-3-3 SDRAM provides an ALU/FPU bandwidth of 441/517 MB/s. By setting modifying the BIOS settings to Turbo and CAS latency of 2, we can set the SDRAM to run at a timing of PC133 2-2-2. This increased the ALU/FPU bandwidths by a further 10-13% to 484/587 MB/s - both very respectable scores. Of course I have done more than just change the CAS latency here; I also decreased the RAS-to-CAS delay and RAS precharge delay by one cycle. This therefore represents the maximum change in timings.
The 3DMark2000 and QuakeIII scores were again less dramatic: from 4900 to 4933 and 82 to 82.8 respectively. Same excuse as before.
Overclocking the memory
Overclocking refers to running a component at a speed greater than it was designed for. This gives enhanced performance but often at the expense of decreased stability. I normally run my 1GHz Thunderbird at 1.1GHz by changing its multiplier to x11 in the BIOS. This is 100% stable, but I have difficulty overclocking the CPU much beyond this. To test the overclocking potential of this memory, I changed the multiplier back to its default of x10 and instead started increasing the FSB. Remember that the KT133 chipset does not have an excellent reputation for a greatly overclocked FSB.
I found I was able to gradually increase the FSB to 112 MHz, with the machine finally hanging during the 3DMark2000 test when the FSB was set at 113MHz. At 112MHz, the memory bus is running at 149MHz. Not only was the Corsair 150MHz 100% stable at this frequency - but I was running at the aggressive CAS latency 2 timing rather than the specified CAS latency 3! The Sandra scores were 546/661 MB/s - a full 12% better bandwidth (as you might expect with a 12% overclock of the memory bus).
Realising that the FSB of 112MHz represented a CPU speed of 1120MHz, which was at the limits of what I knew my CPU is capable of, I decided to underclock the CPU multiplier to x9.5 and continue increasing the FSB. I managed to successfully achieve an FSB of 116MHz and successfully boot into Windows and conduct the benchmarks. This showed that the Corsair 150MHz SDRAM could successfully run at 154MHz at a timing of 2-2-2! Well beyond its specification of 150MHz 3-2-2. At this setting the Sandra memory scores reached 543/674 MB/s - the highest I achieved. As proof, I attach the score below. Note that the scores are significantly higher than the reference AMD Athlon 1GHz on a KT133 chipset.
Given that the RAM was running well beyond its specification, I reduced the CAS latency to 3, and continued to increase the FSB to 119MHz. This meant running the Corsair SDRAM at 158MHz with 3-2-2 timing. Doing this I was able to successfully boot into Windows and conduct the benchmarking tests and achieved a score of 544/665 MB/s. This is a little lower than the 543/674 achieved with 154MHz at 2-2-2, and indicates the tradeoff between clock speed and timing settings.
I was unable to go beyond an FSB of 119MHz, although I suspect this was due to limitations in the KT133 chipset rather than the SDRAM. A 19% overclock is a rare achievement with this chipset.
All in all I was very impressed with the performance of this Corsair 150MHz SDRAM. It easily surpassed its 150MHz 3-2-2 specification and ran at 154MHz at 2-2-2. In the end, the motherboard chipset appeared to limit the performance, not the SDRAM.
In simple terms, this test illustrated one of the benefits of quality RAM. By simply modifying the BIOS settings I was able to achieve an almost 50% greater ALU memory bandwidth and 70% greater FPU bandwidth compared with that achievable with PC100 3-3-3 SDRAM. I would not hesitate to recommend Corsair Micro's 150MHz SDRAM to the overclocking community.
Quantity of RAM
The above tests have established the importance of quality, but what of quantity? It is certainly easy to define the lower limit. Any machine running Windows will need at least 64MB of memory to avoid swapping with one or two applications running. For most home users, 128MB of RAM is sufficient to support Windows, games and activities such as web browsing and word processing. More than 128MB is only necessary for memory intensive activities such as photo editing, video capture or running the computer with several applications open simultaneously. It is hard to conceive of situations where a home user would ever require more than 256MB.
In these tests, I investigated the effects of adding additional SDRAM to the second and third DIMM sockets on the motherboard. Some users experience stability issues when adding additional DIMMs to their motherboard, but this did not prove a problem for me. This is probably due to the quality of the RAM. Note that on the KT7 at least, there are two BIOS settings that can be used to improve stability with multiple DIMMs: Delay DRAM Read Latch (increase the delay to improve stability) and MD Driving Strength (set to Hi).
I did notice, however, a small but noticeable drop in memory bandwidth as additional DIMMs were added, although the ability to overclock the system was not affected. With two 128MB sticks of 150MHz Corsair RAM, in DIMM sockets 0/1 and 2/3, the memory score at 100MHz FSB fell from 484/587 to 474/568. When a third 128MB DIMM (Mushkin PC133 Rev 2) was added to socket 4/5 then the score fell further to 470/557. I assume this reduction in throughput is due to the additional overhead placed on the memory controller when additional DIMMs are used, although I have no evidence for this.
It is therefore worth noting that you should only add enough memory to ensure your machine doesn't need to access virtual memory (the disk). Additional memory above and beyond this may actual reduce performance slightly.
My RAM is better than your RAM...
Although the importance of quality RAM has been established, does it make any difference which brand is used? To test this, I compared the bandwidth achieved with 128MB of Corsair 150MHz at 133MHz with 2-2-2 timings to that achieved with 128MB of Mushkin PC133 Rev.2 at 133MHz with 2-2-2 timings. Both scored an identical 484/587 MB/s in Sandra. This is really to be expected: both were running at the same speed using the same memory timings.
Assuming the memory is a quality brand, and hence stable, the difference is therefore one of overclockability (which is determined by the design of the DIMM circuitry and choice of RAM chip) and price.
These tests have shown a number of things:
- PC133 SDRAM offers up to 33% better bandwidth than PC100 SDRAM. Furthermore, PC100 is often the cause of stability problems on Athlon motherboards.
- CAS latency 2 SDRAM offers an increase in memory bandwidth of some 10-13% over CAS latency 3.
- Good quality SDRAM such as Corsair 150MHz SDRAM allows the FSB to be pushed to well beyond its specification, resulting in a potential increase in bandwidth of up to 70% over PC100 CAS3.
- Underclocking the processor's multiplier can allow the FSB to be increased further, resulting in better SDRAM performance
- Adding additional sticks of SDRAM can have a small negative effect on memory bandwidth. Therefore only ever add as much memory as you need.
- The difference between RAM types is fundamentally that of cost and stability. All SDRAM provides equal bandwidth at a given setting - it's just not guaranteed that all makes can be stable!
- Corsair 150MHz CAS3 SDRAM is a most capable memory, running in these tests well beyond its specification at 154MHz CAS2.