266 fsb ?

[Sky Viper]

Is this a characteristic that is only inside the chip?  Or is there a setting that I should look for to ensure that I am using my new 1.33 G proc. fully.

I just upgraded from my 750 Duron and went from 50 fps to 75 in the tower at A1 Baltic (off line).

my specs:
Asus A7V133
Athlon T-Bird 1.33 G
256M PC133 SDRAM
Visiontek GeForce2 GTS 64M AGP (12.41 Drivers)
10G IDE ATA Hard Drive
Soundblaster Live!
Creative Modem Blaster 56k
DX8.1

Viper

[Defiance]

Hiya,
I think although i maybe wrong but the 1.33 came in 2 flavours ie 200 or 266

support in AMD® Thunderbird™ / Duron™ 550MHz ~ 1.3GHz+ processors. Such new chipset is ASUS's first Socket A version to incorporate 266/200MHz FSB and 1.5GB PC133 / PC100 / VCM133 SDRAM capability in the x86 platforms. This marvelous mainboard features support for Stepless Frequency Selection, a new technology which allows CPU frequency setting to be set in 1MHz-increments and also Ultra-Fast DMA/100 data transfer, and AGP Pro/4X, etc. The additional upgrades include RAID 0 or 1 support, 4 USB ports, and PromiseR PCI-ATA100 controller for UltraDMA/100.

Stats from asus site

Man i gotta sober up lol

Sorry for edits i been celebrating englands win

Have Fun

Def

[bloom25]

The 266 Mhz fsb is DDR, which means that it actually runs at a clockrate of 133 Mhz.  If you go into the bios and look at the CPU FSB (front side bus) frequency if you see 133 MHz, then it's running at full speed.

I can tell you that you are at 133 Mhz (266 Mhz) though, because if you were at 100 mhz your CPU would only be detected as a 1Ghz processor.  Just look at boot up time, if it says 1333 mhz, then you are ok.

[AKWarp]

The T-Birds came in two flavors -- 200mhz FSB and 266mhz FSB.  Which you use depends primarily on your mobo.  

If you are using 133mhz SDRAM, then you should be running your FSB at 133mhz.  If you are, and your CPU is detected at the correct speed, then you have a 266mhz FSB CPU, however you are NOT utilizing the DDR capability.  The Asus A7V133 uses only standard PC100 or PC133 SDRAM memory, it does not have the capability to use DDR memory.  

Don't get confused about the 266mhz bus thing.  The T-Birds only come in 200 and 266 flavors (which means the CPU has the capability to use that speed and is completely independent, to some degree, about the memory).  The reason is that even DDR memory is only 100 or 133 mhz, it's just that it is a "double pumped".  This means that PC2100 memory is actually 133mhz SDRAM that allows two read or write operations per clock cycle instead of one.  In real world use DDR memory actually sees only about a 15%-20% increase in performance over SDR memory....still an increase, but not double the performance as many would assume.  That's because your system isn't always performing 2 reads or writes to memory all the time.

From what you have posted, it sounds to me like your system is fine and is running at the correct speeds.  However, that A7V133 is getting a litte long in the tooth by today's standards.

[beet1e]

However, that A7V133 is getting a litte long in the tooth by today's standards.

EEk!  I've had mine barely a year, and it's already long in the tooth?

Looks like I've come to the right place to ask the following question. I have that A7V133 board, with a single stick of 256MB SDRAM. The FSB is set to 100MHz, and the CPU clock multiplier is set to 12, giving 1.2GHz which is the non-overclocked speed of my Athlon processor. Q: Can I alter either of these parameters to increase CPU performance, without overclocking it?

See the attached system spec - .MHT file opens with browser.

By the way, I've been damn impressed with this Asus mobo, and the support provided on the excellent website. Since I upgraded to this config a year ago, things have run as smooth as silk.

[MrLars]

Originally posted by AKWarp


The reason is that even DDR memory is only 100 or 133 mhz, it's just that it is a "double pumped".  This means that PC2100 memory is actually 133mhz SDRAM that allows two read or write operations per clock cycle instead of one.

Just finished my new box and found this out myself...but in my case the ram is pumped 4X giving me a 400mhz FSB with a 100mhz clock cycle.

Asus P4T-E
2gig Northwood
512 RDRAM

I'm extreemly happy with this rig and the cost for the upgraded components was a paltry $400  

[bloom25]

AKWarp, I can't tell for sure from your post, but you may be a little bit confused on some things here.  (I wish I could draw a picture to explain what I'm about to say here, it would help a lot.)

You seem to indicate that using PC133 Sdram somehow means "not utilizing the DDR capability."  I think you might be confusing the memory bus with the processor bus here, they are not the same.

The AMD Athlon processors use a bus protocol known as EV6.  This bus links the processor to a portion of the motherboard chipset known as the "northbridge".  The northbridge is directly responsible for communicating with system memory (RAM) and the AGP bus.  The northbridge also connects with a component known as the "southbridge."  Historically this connection was done via the PCI bus, but the newest chipsets typically use their own dedicated bus for this. (Vlink, Hypertransport, etc)  The southbridge controls devices such as the PS2 ports, USB, COM (serial) ports, LPT (printer) ports, and controls access to cards on the PCI bus (sound cards etc.).  I mention this because it helps to understand how components talk to each other on the motherboard.

The EV6 bus links Athlon (and Duron) series CPUs to the northbridge.  This bus operates at either 100 MHz or 133 MHz.  It transfers data at both the rising and falling edges of the clock.  This means data is transfered two times every clock cycle, thus the term DDR (double data rate).  Athlon processors claiming a "200 MHz" FSB are thus 100 MHz DDR and "266 MHz" is 133 MHz DDR.

The system memory bus is IN NO WAY directly related to the processor FSB (front side bus).  (The number of wires in the bus isn't even the same... )  Older SDRAM memory transfered data at the rising edge of the clock only.  DDR SDRAM transfers data on both edges of the clock.  The northbridge functions as a type of translator between memory and CPU.  (That's a terrible analogy, but it will suffice.  There are some peculiarties of SDRAM memory that I'm not going to discuss here... )  To access information in system memory the CPU sends out a request.  The northbridge then selects and activates a particular row in memory, which is then transfered into the northbridge.  It is this transfer of data to the northbridge which is faster with DDR Sdram.  When the data has finally arrived at the northbridge, the CPU is signaled and the data is placed on the CPU FSB and transfered into the CPU.  The data transfer speed from PC2100 DDR Sdram happens to occur at the same rate as the Athlons with the "266 MHz" FSB, but it takes quite a long time between CPU request and the time the CPU actually receives the data, this is generally known as latency.  To make this discussion complete, I'd have to discuss how DRAM memory operates and some of the tricks involved in speeding up the process.  (In general terms it must be charged before being read from, and reading from a particular cell discharges that cell, which must then be recharged before being read from again.  Terms known as CAS (column address strobe), RAS (row address strobe), and precharge all come into play here.  You may be familiar with the term CAS, as in CAS 2, CAS 3 memory.  It's also worth noting that SDRAM transfers the entire contents of a row, even though the CPU only requests a particular portion of that row.  This is a trick to minimize latency, as often the next request from the CPU will have already been transferred.)

AMDs upcoming "Hammer" line of CPUs includes the memory controller on the CPU itself to minimize latency.

I'm going to have to cut this short, as I've got some work to do.  I hope I've helped you understand this a little bit better.  I'll probably add more later on tonight.

[AKWarp]

MrLars, yours is 400mhz because you use RDRAM memory...completely different animal than DDR memory :-)

Bloom, I believe what you just said is the long techincal explaination of what I said ;-)

My concern was that he didn't get confused about 200/266mhz CPU speed, and only using 133mhz SDRAM.  Many people don't understand the difference between the CPU and memory bus speeds, especially since so much emphasis is placed on the FSB.


Beetle, altering the FSB and/or cpu multiplier IS overclocking ;-)
That's how it's done.  Your CPU's rated "speed" is the FSB speed multiplied by the CPU multiplier (that's why it's called the cpu multiplier :0 ).  If you increase the FSB from 100mhz to say, 110mhz and keep the CPU mult of 12, then your CPU speed is 1.32ghz (110mhz x 12 = 1.32ghz).  Ideally, the best overclock would be to control both the FSB and CPU mult, keep the CPU mult down and up the FSB as much as you can.  The reason is that with higher bus speeds, everything is faster (peripherals, memory, etc) which attributes a lot to increased real world performance.  There is a downfall to all this though.  The PCI bus, IDE bus, etc are designed to run at a spec speed (33.3mhz for PCI, for example).  So, in addition to the FSB and CPU multiplier setting, you also have what's called a "clock divider".  On most motherboards, this clock divider has 3 settings:  1/2, 1/3, or 1/4.  This divider "divides" the clock rate of the FSB so things are in spec for the PCI bus, etc.  So, at a 100mhz FSB speed, you use a 1/3 clock divider (1/3 x 100mhz = 33mhz), so the PCI bus is happy.  The kicker here is you may have a lot of leeway in upping your FSB speed and the CPU itself might be just fine with it, but your peripherals, dives, etc may not.  Most industry specs allow a certain percentage of error margin and your peripherals may even be happy up to 40-42mhz on the PCI bus.  Then again, they may not...but the only way to know is to try it....gradually.  Don't go for a mongo overclock straight up, you might end up frying something (DAMHIKT!).  

Most folks looking for the mega-overclock make the mistake of using the 1/4 clock divider if their FSB speeds and 1/3 divider are too much for the peripherals.  If your FSB overclock is successful at very high increases, then this may be fine since you may still obtain spec speeds on your PCI bus while gaining a huge bump in your CPU, but generally those kinds of successes are in the days past with the old celerons and the like, or with high dollar cooling solutions (such as refridgeration).  If you end up lowering your PCI or IDE bus speeds below spec, then chances are your overall system performance will suffer instead of increase (even though you are techincally getting a higher speed from your CPU).   This is because you are starving your CPU for data...it might be able to crank out more speed, but it can't do it if the rest of the system can't feed it.

In the case of the Athlons, it is possible to unlock the CPU multiplier.   Early Athlon unlocking involved nothing more than a pencil ;-)  Seriously, you could scratch the pencil from pin to pin on a certain area of the chip and unlock the multiplier.  Basically, the pencil lead conducts electricity enough to make the connection and it works.  On newer Athlons, AMD started using a new manufacturing method to try to discourage this technique.  AMD started using a laser to cut small pits in front of the pins on the CPU die so you could not bridge them with the pencil lead.  But, leave it to hardcore overclockers...someone figured out  away to bypas even this.  The trick now is to fill the pits with a non-conducting substance (usually some kind of glue or epoxy), and then connect the pins with conductive silver lacquer.  Most overclocking websites now offer "Athlon unlocking kits" for about $10 specifically for this purpose.

[MrLars]

Originally posted by AKWarp
MrLars, yours is 400mhz because you use RDRAM memory...completely different animal than DDR memory :-)

You piqued my interest and, after reading about the 16 bit v 64 bit bus config, I understand now what you meen by different animals. Fortunatly my box has excelent cooling across both sticks <4 including the RIMS> so I have no problems having both active so latency is no problem for me. The reviews I've read are dated to say the least and list the 2 main drawbacks of RDRAM as being heat and cost. Both are not a problem since heat disipation can be improved with better circulation in the box and the cost is inline with quality DDR ram these days.

[AKWarp]

Exactly, and on the surface anyone would just drool at the thought of 400 and 800mhz memory speeds.  Unfortunately, as shown in tests, the P4 systems equipped with RDRAM really weren't performing much higher than DDR based systems (RDRAM idiosyncracies aside).  The combination of the early high RDRAM prices and no major speed advantage in real world use lead to many board makers to release DDR based P4 mobos.  However, with price no longer a factor, if I were to build a P4 system today, then an RDRAM based board would be a given because although slight, it is faster.

Personally, I think the lack of overwhelming performance of the RDRAM based systems has more to do with the P4 architecture (specifically that monster pipeline and the inherent lag in flushing that beotch) than with high speed memory.  I can only salivate at the thought of an Athlon XP with 800mhz capability  ;-)  However, with 64 bit CPU's and OS's about to hit the mainstream, who knows what we'll be seeing in terms of bandwidth ;-P

[bloom25]

While it is an interesting technology, do not expect Rambus to stick around much longer.  Intel themselves plan to transistion all but the Xeon platforms to DDR SDRAM eventually.

To be honest, I don't see Rambus ever replacing DDR Sdram.  It's purely a question of economics.  It's cheaper to make SDRAM modules.  Rambus, as a company, is also not well liked by the major memory manufacturers in general.  (It could be because Rambus sued them all earlier this year... )

(Rambus ram does suffer from it's share of problems btw.   All memory modules in a Rambus architecture setup are connected serially.  That means data being read or written in a Rambus memory architecture must pass through all the other memory modules ahead of it on the bus.  Rambus memory also requires bus termination, unlike SDRAM.)

The Athlon XP would actually lose performance going to Rambus memory.  (Just like the P3 did.)  The K7 architecture is more sensitive to memory latency than raw bandwidth.  The Athlon series of processors is capable of executing multiple instructions in parallel and has a relatively short pipeline.  Processors of this type would benefit primarily by increasing the FSB clock frequency and increasing the size of the L2 cache.  Data prefetching and branch prediction are also very helpful.  (AMD does actually have a license from Rambus btw.)  If you guys go and take a look at the Throughbred core XP 2200 + reviews popping up around the web today you can see that AMD changed the Athlon floorplan (moved the pieces around inside the CPU).  Specifically they moved the L2 cache to the bottom edge of the chip, which could mean they plan on increasing the L2 cache size for their upcoming Barton core Athlon XPs.  Throughbred looks to me to be nothing more than a quick patch until Barton and Hammer are released.  It's actually quite impressive how quickly AMD managed to get Throughbred up to speed considering just how much the layout has changed.  

The Hammer would actually be an ideal candidate for Rambus memory.  Its integrated memory controller would reduce the latency greatly, and the Sledgehammer's planned 8 way setup (AMD Opteron) would benefit from the additional memory bandwidth Rambus offers.  (That's probably why Intel is keeping Rambus around for its Xeons.)  AMD seems to have choosen a dual channel DDR Sdram setup (as has Intel in its Granite Bay chipset) for it's next generation CPU.  A dual channel DDR Sdram setup offers low latency, low cost, and high bandwidth.  One HUGE disadvantage is the shear number of pins necessary on the Hammer chip.  If memory serves, Athlon XP 64 will have 753 pins and Opteron 903 pins.  This is going to result in very expensive chip packages.  (The package holds the die and connects all the pins to contact points along the edge of the die.) Using Rambus ram, the number of pins could be reduced.

There are some Q3 benchmarks of an 800 MHz Hammer available on the web now.  It's shown outperforming a 1.6 Northwood P4.  If it truely launches on time in late August, early October, at approx 2 Ghz true clockspeed it will be an impressive performer indeed.  I was interested in that the CPU was of A0 stepping, which indicates that it is of the first batch of CPUs AMD produced.  Since AMD began sampling Hammer in January, I'd expect them to have increased Hammer's clockspeed considerably since then.

[bloom25]

Sorry to take this thread so far off topic, but I didn't feel like starting a new one...

Take a look at the die photos of the Palomino core and Throughbred cores on this page:  http://www.anandtech.com/cpu/showdoc.html?i=1635&p=2

(There might be better ones around, but I haven't had time to search for them.)

The Palomino L2 cache is located on the right side of the CPU.  Notice in the Throughbred that that L2 cache has moved to the bottom of the die.  I can't for sure tell because the picture is just too small, but it looks like the L2 cache interface logic has also been expanded.  (The rest of the processor looks to be unchanged.)  That indicates to me that AMD plans on expanding the L2 cache in the future.

It appears by reading the rest of the article that AMD has not yet got their .13 micron production line up to speed.  The fact that the XP 2200s are not overclocking well at all indicates there may be production problems yet to be worked out.  (The Palomino based XP 2100+ are better overclockers than the current batch of Throughbreds.)

I don't think Throughbred is going to have more than a 3 month lifetime before being replaced...  At least they are compatible with nearly all Socket A boards, so Athlon users still have an upgrade path.  (You all are probably aware that Intel's P4 2533 Mhz, 2.4 Ghz "B", and 2.26 GHz "B" cannot be used on the existing i850 and i845D boards.)  The chances appear good that current Socket A boards may accept Barton core Athlon XPs if they are released as planned in Q3.