Can a KT133A support Athlon XP?

[Bob43]

I have an ECS K7VZA which uses the Via KT133A chipset, along with a Duron 900 MHz. Can that chipset support an upgrade to an Athlon XP 1600?

thanks.

[Sauron_Daz]

Check the mobomaker's website to see if a BIOSupgrade is available that allows you to use an XPxxxx+. The chipset can, but it depends on the mobomaker to make the nessesary adjustments to the BIOSfile.

[Peter M]

If you're sure your K7VZA has KT133A already (meaning it's rev. 3.x), then put the latest BIOS in and use any Athlon-XP as you please - ECS lists up to 2200+, so even the Thoroughbreds work.

Mind you, you're going to need adequate power supply and CPU cooling. Don't skimp on those. And your SDRAM must be PC133 too, since you can't run PC100 RAM with 133 MHz bus CPUs.

[gibsinep]

I had one running an XP for a few days.

However remember that you will be held back by your RAM and old board. So don't expect performance from a 2000+ with your setup to be the same as performance with a newer board. Kt 266a-333-400 :t

[causticVapor]

Exactly; bandwidth will have been halved, and thus performance will be 75-85% of its optimal.

[Peter M]

Depends ... on boards that have both DDR and SDR slots, you can do an apples-to-apples comparison. The performance increase from DDR RAM typically is from zero to 10 percent at best. Athlon doesn't depend on memory performance as dramatically as Pentium-4 does. Synthetic memory throughput benchmark results are of course more drastic, but we want to improve real life, don't we?
Sure, newer boards have other nice things, more I/O mainly, but if you don't need these, you can just as well stay with what you have.
regards, Peter

[Rugor]

The other thing to remember when considering using a KT133A board with an Athlon XP is that the performance penalty from using SDRAM does scale with clock speed. An Athlon XP 1600+ at 1.4GHz will perform similarly on KT133A to a DDR chipset, while a 2.0GHZ XP 2400+ will suffer more from the reduced memory bandwidth. Remember it's trying to feed some 45% more clock ticks with the same amount of memory bandwidth.

Anandtech did a comparison between KT133A and various first generation DDR chipsets and found that in many cases the low latency memory controller in KT133A outperformed DDR systems. They also determined that when dealing with sub-GHz CPUs there didn't seem to be any advantage from DDR.

[deadkenny]

Just reinforcing the previous couple posts, DDR does not equal double the performance. Obviously the slower the memory the more of a limiting factor it is and conversely the faster the less of a limit. However, even infinitely fast memory wouldn't increase performance beyond a certain limit, since the other components (bus speeds on the mobo, CPU, HDD) would then be responsible for limiting performance. Furthermore, DDR is not really 'twice' as fast as regular SDRAM anyway. Two signals per cycle is not the same as twice as many cycles per second. Sometimes DDR is referred to as being '266' or '333', but that's not really accurate. The underlying clock speed is still just 133 or 166MHz. As Peter said, the performance difference in practice is usually only 10%, maybe not even that. It might be as high as 15%, but the only way it would approach twice as fast would be some artifical benchmark.

[rizboy]

I have a ASUS A7V133 motherboard, and I checked ASUS website, and A7V133 motherboards use VIA KT133A chipset.

Anywho, can my ASUS motherboard support AthlonXP processors? Since I have 900mhz Athlon ThunderBird, I was thinking of upgrading my CPU to about AthlonXP 1600+ or 1800+.

My BIOS version is 1001C and I can download the update version from the ASUS website (1009). How do I update my BIOS anyway, I heard you use a floppy or something like that...

[deadkenny]

Peter will probably crucify me if I get this wrong, but I believe the processor support is a function of the particular board, not the chipset per se. The ASUS site should specify which processors are supported, although if it's a processor that came out after the board, you will have to find recent documentation. The BIOS revisions sometimes have notes, so if you're lucky there might be one that explicitly mentions adding XP support.

[Rugor]

It's both, but chipset first.

If the chipset does not support the processor, then no board based on that chipset can. However, not all boards based on a given chipset can support all processors. This can be a matter of either not supporting the proper voltages, or not having the proper BIOS support. Although if the voltages are supported you may be able to use the processor, despite it not being correctly recognized by the bios.

As an example, all KT133A boards use the Socket A (or Socket 462) infrastructure. Because AMD has kept this socket and its pinout stable, any Socket A processor can be physically installled in it. Some boards have the requisite power circuitry to support the XP, others do not. However, no KX133 motherboards support the XP, since that chipset's timings were such that it was incompatible with socket A.

If I can borrow some terms from my college philosophy courses, chipset support is a necessary but not a sufficient condition. If you don't have chipset support then there is no way that the processor can talk to the System, and you're dead in the water. If you have chipset support it may be possible to put some sort of modification to the board to provide the proper voltages for an unsupported processor. That's not a mod I would care to do but it's theoretically possible.

[deadkenny]

OK, and maybe this is a dumb question, but is there any intrinsic reason why a chipset couldn't be implemented in another format? Usually, when a new format comes along, there are new chipsets to go with it. But, for example, the BX chipset started out on slot 1 boards, but was later also used on socket 370 boards.

[Rugor]

No intrinsic reason, it just so happens there were trace length and timing issues with KX133 that prevented it from being implemented in Socket A and properly supporting the Thunderbirds.

[Peter M]

Chipset support for a certain CPU changes only when there have been changes to the CPU front side bus. From Athlon T-Bird to Athlon XP, there weren't any.
Mainboard support is about providing the right voltages and, in case the new CPU draws more power, having strong enough CPU voltage regulators. Again, neither changed from TB to XP.
BIOS support then is the last thing that must be there - if BIOS doesn't know what to do with a certain CPU make and model, then the thing won't work.

KX133 didn't suit Thunderbirds because from the original Slot-A Athlon to the TB, the electrical properties of the CPU bus did change. Enter KT133. From Slot-1 to initial 66/100 MHz socket-370, nothing changed, so the chipsets remained the same.

[causticVapor]

About SDR and DDR -- why is it that so many hardware sites then talk about massive performance gains on the K7 platform with it? It would seem that even the bandwidth-efficient K7 processor would benefit more than 10% with it.

I recently pumped up the FSB on a 2400+ from 266MHz to 400MHz (synchronous with the memory), and boot times have been halved. I realize that these gains in FSB clock speed do not amount to 3GB/s of real transfer, but doesn't a 400MHz FSB help a 2GHz CPU perform significantly better?

I do agree that there is not that much of a performance difference between a T-Bird 1GHz with DDR and a Pentium III at 1GHz with DDR. However, I always thought that if the FSB was synchronous with the memory then a bottleneck would be alleviated.

[Gripsz]

"Chipset support for a certain CPU changes only when there have been changes to the CPU front side bus. From Athlon T-Bird to Athlon XP, there weren't any.
Mainboard support is about providing the right voltages and, in case the new CPU draws more power, having strong enough CPU voltage regulators. Again, neither changed from TB to XP.
BIOS support then is the last thing that must be there - if BIOS doesn't know what to do with a certain CPU make and model, then the thing won't work.

KX133 didn't suit Thunderbirds because from the original Slot-A Athlon to the TB, the electrical properties of the CPU bus did change. Enter KT133. From Slot-1 to initial 66/100 MHz socket-370, nothing changed, so the chipsets remained the same."

from what you say Peter, then would most KT133 boards run say an XP 2100 @ 1300 ? if not, then why not- (beyond "BIOS doesn't know what to do") ?

[Peter M]

Frankly, I wouldn't expect any KT133 board's BIOS to support a CPU type that isn't really usable in there. But true, if there were 100 MHz bus Athlon XP, they'd electrically run in that board. There are btw, the lower "mobile" XP speed grades were 100 MHz bus.

[deadkenny]

I would expect that it would run, even if it's a 133MHz FSB processor on a board limited to 100MHz FSB. If you unlocked the processor, you could probably up the multiplier to the point where it's running near spec clock speed (of course lower FSB and higher multiplier is not spec). The only problem might be that SSE on the XP won't be 'recognized' so you'll lose some of the advantage of the XP chip.

Regarding the DDR question, DDR and async are two different issues. DDR and regular SDRAM are based on underlying clock cycles of 100, 133 or 166MHz (well, there's no official 166 regular SDRAM though). Whether one or the other is double data rate doesn't matter. If the FSB and memory bus are both 133 then it's synchronous, regardless of whether or not it's DDR.

[Rugor]

Caustic Vapor,

yes a faster FSB will help a processor perform better, and not just by running synchronously with DDR 400 memory. It also improves data transfers from CPU to northbridge.

It will speed boot times because the faster FSB lets everything communicate faster.

Total system speed is always improved by increasing the FSB, since it decreases total system latency by reducing the multiplier and hence the number of clock cycles the CPU has to wait for fresh data.

One reason the "333MHz" Athlons remain competitive with the higher clocked "533MHz" P4's is that the higher base rate of the FSB, combined with the lower clock speed allow for a much lower multiplier so the CPU wastes fewer cycles waiting for the FSB.

[ALS]

It does support it, but just 266 FSB. Try looking for bios update to see if it does.

[Peter M]

Again, the XP models (Palomino/Thoroughbred) are different from T-Bird, as far as BIOS support goes. If the board's BIOS doesn't at least have support for "Morgan" Durons (which are similar to Palominos), then I wouldn't even bother.

[ALS]

Take for example Chaintech 7AJA2. It has KTA133 and it supports AMD Athlon XP 2000.

[ALS]

VIA KT133A supports 266 FSB DDR
VIA KT133E supports 200 FSB DDR

[causticVapor]

Originally posted by Rugor
Caustic Vapor,

yes a faster FSB will help a processor perform better, and not just by running synchronously with DDR 400 memory. It also improves data transfers from CPU to northbridge.

It will speed boot times because the faster FSB lets everything communicate faster.

Total system speed is always improved by increasing the FSB, since it decreases total system latency by reducing the multiplier and hence the number of clock cycles the CPU has to wait for fresh data.

One reason the "333MHz" Athlons remain competitive with the higher clocked "533MHz" P4's is that the higher base rate of the FSB, combined with the lower clock speed allow for a much lower multiplier so the CPU wastes fewer cycles waiting for the FSB.




Thanks aplenty, but I'm still a bit confused on why DDR does not double throughput quite as well as it should, and the fact that the 133*4 bus of the P4 looks too good to be true. Isn't latency halved by double-pumping the FSB/mem or quartered through quad-pumping? I've seen some apps display an XP/DDR CPU, say an XP 1800+, at 5.75*266 instead of 11.5*133. Does this really happen or is it fake? I would imagine that the complexities involved with implementing double and quad-pumping take away a lot of the bandwidth available, and the latencies would still be the same because of the lower base clock rate. If I follow that correctly, then DDR400 at 200MHz is able to achieve and exceed the theoretical bandwidth of PC1600 (100*2) DDR in real-life scenarious.

This would all make sense then... the reason why intel is going to "667" and "800MHz" FSB...



EDIT -- now I see -- the latency stays the SAME at the SAME clock speed, no matter how much it's pumped. In order to get closer to those asymptotes the actual frequency must be increased.

[Rugor]

There are two key things to remember when dealing with bandwidth and latency with multiply pumped memory buses.

The first is that DDR SDRAM is still SDRAM and as such works exactly the same way. The only real difference is that it carries data signals on both the rise and fall of the clock rather than just one.

The second is that when it comes to electronics there are two ways to measure time, relative (as a function of clock cycles) and absolute (as a function of nanoseconds).

Let's take PC 133 SDRAM, and PC 2100 (DDR 266) DDR SDRAM for our examples. In both cases the chips work at 133MHz or with a cycle time of 7.5 nanoseconds. Given a CAS 2 latency rating they both have a CAS latency of two clock cycles or 15 nanoseconds. So far, so good, but look at it another way. Since this is related to the base clock, those two cycles of latency correspond to <i><b>four</i></b> data ticks for the DDR as opposed to two for the SDRAM.

This means that latencies have a greater effect on Bandwidth on DDR than it would on SDR memory. Let's compare hypothetical 400MHz SDR with DDR 400, and for the purposes of the discussion assume CAS latency is the only timing that matters and that both work at a 2 cycle CAS latency. In this situation the 400MHz SDR would have 5 nanoseconds of latency in comparison to the ten nanoseconds of the DDR. When dealing with large numbers of small data accesses these delays add up.

Higher base rates do have their own problems. For one they require better signal integrity and tighter timings just to work. For another, faster memory speeds are often limited to less aggresive memory timings because of the limits of how fast things can work in real time.

It's always a trade-off.

[Gripsz]

i guess motherboards aren't your forte

[Rugor]

who's forte?

[gibsinep]

I think he was complimenting you. :confused:

[Grentarc]

I have a KT133a Abit KT7A-RAID.. . (rev 1.3) and it is fine for the new T-Bred XPs even... although i am not happy with it at the moment, because it cant go higher that 152MHz FSB without corrupting my HDD, so its a slight problem there...
I have my KT133a at 1.9GHz (12.5x 152) and ther is not much increase in speed as you would expect from a 433MHz OC, but that is mainly due to the SDRAM... ahh well...

[causticVapor]

I see what you're saying Rugor, the rising and falling pumping takes up time, and decreases the efficiency of DDR... thanks.

[Gripsz]

forgive my response Peter. an inadvertant trial at rendering your shield of a response as lack of knowledge. why is what i asked and i expected more. perhaps i learned something.

Gripsz