Amd64 2800+

[Nico3k]

how much was that whole setup?
i think my XP-M is still a tiny bit faster. superpi 28sec, but 10610 MIPS on sandra.......... these 64bit setups are becomming more and more attractive every day.

the day someone reports a mobile 64bit chip running at 3GHz on watercooling is when I purchase

interesting to see those benchmarks though.

 

[hitechjb1]

2t, I updated my info in the collective A64 resource. 4:3 divider is what was being ran at the time of that last update as well, rather than the usual 6:5

2T would make the bandwidth efficiency lower than 1T.

Do you mean 2:3 memory:FSB ratio? Is there a 3:4 memory:FSB ratio in the bios?

 

[Buhammot]

The whole setup, as far as mobo and cpu, was $273USD, not too bad I think. But as far as the ratio, it isnt specific as to what the exact divider is, so I take the base HTT speed, 200, and the selected mem speed, 133 (theoretical limit they call it) and divide them. And some how.. my math was way off... It is a 2:3 divider.

 

[Jess1313]

could you not get the 150:200 diviser working w/ a64 tweaker. ?

 

[Buhammot]

They do work, just that it is kinda irritating to have to set them in the bios one way.. then set it with a64tweaker, then set it with clockgen...

 

[magnummafia]

was wondering if you have used a northwood setup recently, like a 2.8c overclocked to 3.2-3.5 range? if so how does the 2800 at 2.5ghz compare?

 

[Jess1313]

They do work, just that it is kinda irritating to have to set them in the bios one way.. then set it with a64tweaker, then set it with clockgen...

yeah a64 treaker can apply at start up, and if you feel like doing it, clockgen can to,(but you sill end up having to set bios/a64 TW/and Cg) its just a pain to set up, but when your done, all to do is go into this note pad file, and when the Multipier, and Htt. i dont know if many people use this or not, but if you need help setting it up to load at start up just PM.

 

[hitechjb1]

I noticed some interesting things:

In the first two CPU-Z, the family, model, stepping was F, E, 0 respectively, but
in the 3rd CPU-Z, they became F, C, 0 respectively.
FC0 is the representation for NewCastle.

Further, the CPU-Z version used were the same version 1.23.

 

[The Coolest]

hitechjb1, there was a problem with some of the NCs having "E" under model instead of "C", AMD issued an erreta, and basically said that any chip that actually is being read as "E" should be displayed as "C".

 

[hitechjb1]

hitechjb1, there was a problem with some of the NCs having "E" under model instead of "C", AMD issued an erreta, and basically said that any chip that actually is being read as "E" should be displayed as "C".

I am aware of that errata. But the question is, the same version 1.23 of CPU-Z run at different time gives different results, sometimes FE0 and sometime FC0, .... So what have changed to make the results difference.

Buhammot: Were the different CPU-Z runs using the exact same version of CPU-Z?

 

[Buhammot]

I only have the 1.23.. so yea

Also, I am runnin the 1t timings now, and my efficiency is 93% as long as i can keep it stable. I also ordered the Epox board, and gonna keep the best of the 2...

 

[The Coolest]

Well... It could always be a bug in the program.

 

[hawtrawkr]

was wondering if you have used a northwood setup recently, like a 2.8c overclocked to 3.2-3.5 range? if so how does the 2800 at 2.5ghz compare?

im running a similar setup with similar results to Buhammot so far the 2800 @ 2.5 has added 3k to my 2001 scores havent run it on 03 yet and still need to tweak some things out since this is my first amd system and i just threw it toghether yesterday

 

[magnummafia]

hey where did u get your 2800 from? i want to make sure i get a cg stepping
thanks

 

[The Coolest]

look for ADA2800AXBOX on newegg, the "AX" stands for a NewCastle.

 

[Buhammot]

I ordered mine through newegg, but they tried to pawn a clawhammer on me... unfortunately, i had the idea of fraud if they were unwilling to accept my rma request.. but they sent the NC as the rma.. so i was happy.

 

[BeerHunter]

The biggest thing about having the 2800+ is that you have to have some super memory, that can stand some massive mhz. Otherwise, it is an awesome chip from what we have experienced so far.

No you don't for example set 133 mem at 295HTT, 3x ldt, with cheap value ram which is 190 Mhz ram. Course you need a chip 2655 Mhz capable but their are literally dozens of permutations to get cheap ram to work even at the highest overclocks. I bet your chip will work at this speed with such a high OC on stock voltage

Edit: I forgot to compliment you on you alls overclocks. Nice
This thread is proof the mobile XP's are indeed a dead platform, even thier price/performance superiorty is about all gone now.

 

[BeerHunter]

look for ADA2800AXBOX on newegg, the "AX" stands for a NewCastle.

Unfortunatly newegg is only listing the clawhammer dirivative and worse the C0 stepping, ADA2800AEP4AP , so I would recommend doing a pricewatch.com search for ADA2800AXBOX where there are many vendors.

magnummafia,
I just answered your question as well.

 

[BeerHunter]

no if he had a 166:200 ration in bios, and was runnig 9x250, then the ram would be at "Htt x (cpu multiplier/11)", or in this case, HTT x (9/11)

250 x (9/11)= 204mhz ram speed

so you cant have 250HTT and 240 ram would be nice though

and now he is running 9x269htt, w/ 166/200

so thats 269 x (9/11) =220mhz ram speed

BTW hitechjb1, awsome sticky...

According to nvidia hitechjb1's calculations are incorrect.(edit but for 100 and 200 mem settings)

The way the do it is say you set 133 mem this gives a ratio of 133/200 or .665 reguadless of mulitplier or processor. Similarly for mem 166 it would be a ratio 0.83.



Notice in the 133 mem example below, the 370.609 DDR or 185.3 memory bus frequency.

If we work backwards 2507 Mhz / 9 = 278.65mhz HTT(FSB) x .665 ratio = 185.3 or 370.6 DDR, which is exactly what it's reading.

If we were to use hitechjb1's calculations it should read 278.65mhz HTT x (9/14) = 179.13 or 358.27 DDR

It is'nt.

The async mem ratio calc is simple:
200:200 = 200/200 = 1
166:200 = 166/200 = .83
133:200 = 133/200 = .665
100:200 = 100/200 = .5
And so on...

To calculate your memory bus frequency = HTT(FSB) x one of the ratios above depending how you have it set in the bios.

Nvidia's tool will verify this for any CPUany mutiplier. Try it.

I guarantee you Buhammot is running 223.27 Mhz not 220 mhz as in your example.

 

[hitechjb1]

According to nvidia hitechjb1's calculations are incorrect.(edit but for 100 and 200 mem settings)

The way the do it is say you set 133 mem this gives a ratio of 133/200 or .665 reguadless of mulitplier or processor. Similarly for mem 166 it would be a ratio 0.83.

Notice in the 133 mem example below, the 370.609 DDR or 185.3 memory bus frequency.

If we work backwards 2507 Mhz / 9 = 278.65mhz HTT(FSB) x .665 ratio = 185.3 or 370.6 DDR, which is exactly what it's reading.

If we were to use hitechjb1's calculations it should read 278.65mhz HTT x (9/14) = 179.13 or 358.27 DDR

It is'nt.

...

The async mem ratio calc is simple:
200:200 = 200/200 = 1
166:200 = 166/200 = .83
133:200 = 133/200 = .665
100:200 = 100/200 = .5
And so on...

To calculate your memory bus frequency = HTT(FSB) x one of the ratios above depending how you have it set in the bios.

Nvidia's tool will verify this for any CPUany mutiplier. Try it.

I guarantee you Buhammot is running 223.27 Mhz not 220 mhz as in your example.

Thanks for finding it.

I think the problem is the 1/2 memory divider, which is working as shown in the above Nvidia ultility.

CPU_multiplier = 9
memory_FSB_ratio = 2:3
so memory_divider = CPU_multiplier / memory_FSB_ratio = 9 / (2/3) = 13.5

For integer rounding, the memory_divider would be 14 instead of 13.5. This is the discrepancy.

My previous calculation is based on rounding the divider to integer value, it looks like 1/2 multiplier is now possible, in both CPU multiplier and memory divider. Don't know whether it is just for software tweaker or also for bios.

In the past, both CPU and memory divider are rounded to integer (from what I saw). Recently, don't know whether it is due to new bios and/or software tweaker, 1/2 multiplier is possible.

I'll look at it further and make correction to them.