What would a 2700mhz A64 winchester be equal

[RussianSensation]

Strider I see what you are saying. But then it gets pretty confusing since for P4 motherboards the ratios actually do mean what they say. For instance, I am running 235FSB and ram Ratio set to 5:4 on Abit IS7. Essentially my Ram is running at 188mhz or 376 effective. So you are saying that on A64 motherboard, the ratio doesnt tell you exact memory speed...hmm....does this affect all A64 motherboards? I guess this is misleading for users since CPU:RAM should mean specifically that.

I am curious, then if you buy A64 and run it at stock speeds (ie. 200FSB) then your memory isn't running at 400 mhz or PC3200? ....even if you set it to 1:1? Then what is it running at?

 

[StriderGT]

Originally posted by: Zebo

Originally posted by: nick1985
test system

athlon64 socket 939
epox 9nda3+
gig mushkin 3500 level II


270FSB with memory settings at 133FSB 1t-2-2-2-10
benchies


267FSB with memory settings at 166FSB 1t-2.5-3-3-10
benchies


as you can see, even with slightly slower timings, the memory set at 166 FSB Destroys the system at slightly higher FSB over all, and with memory timings faster. this is because the test that used 133FSB for the ram runs at a much crappier ratio for a divider, which kills the memory bandwidth...over 1000 points less in sandra.



the CPU scores are the same, but the memory bandwidth is 18% higher on the system with 166FSB memory, even with slightly lower FSB overall and loosened memory timings.

Yup bandwitdh is where it's at. Look at the recent anandtech A64 memory benches to confirm too.
http://www.anandtech.com/memor...oc.aspx?i=2226&p=9

Apples vs Oranges vs Bananas:

I am starting to loose faith in human sanity. Anand uses an unlocked FX, no wonder he doesn't need any memory dividers, he always runs 1:1!!!

- Nick has a 3000+ a64, max cpu multiplier 9x, so in order to overclock his CPU he NEEDS to go async (there is no performance hit by doing so as you already know)
- In doing so he must choose the correct memory divider (200,166,133,100) AND a cpu multiplier (9x,8x,etc).
- These TWO settings will define his CPU memory divider and then he can calculate his memory frequency.
- The only scenario that things ain't this complicated is when he chooses 1:1 (200) where the above math is still valid but it produces the obvious FSB=MEMORY
- So the comparison that Nick did involved:
1. CPU: 9x267=2403
Memory: 2403/11=218Mhz (not 267*5/6=222,5) this generates tremendous confusion!
and
2. CPU:9x270=2430
Memory:2430/14=173Mhz (not 270*2/3=180)

You need the math type to calculate the 11 and 14 dividers. (OCA64 calculates all these automatically)

So his memory is running 45Mhz slower the second time killing his performance.

PS1 Hasn't anybody understood how to use the calculator yet?!?!?!?!?!?:Q
PS2 This analysis answers your question too, Russian Sensation

 

[Antoneo]

I have been reading this thread and can't help but ask about the following information posted by StriderGT:

- Nick has a 3000+ a64, max cpu multiplier 9x, so in order to overclock his CPU he NEEDS to go async (there is no performance hit by doing so as you already know)
__________________________________________________________________

I'm not very familiar with Athlon 64's but I guess you are saying that memory bandwidth plays less of a role with performance than Intel's P4?

 

[DAPUNISHER]

There is a good brief explanation of the on-die memory controller and a chart showing how the memory frequency is derived here This part is particularly interesting

Of course, the integrated memory controller does not improve the memory bandwidth, but it allows to drastically reduce the request time. The measured latency is very low, as we'll see further. Moreover, unlike an external memory controller, the performances of the integrated controller of the K8 increase as the CPU speed increase ; consequently, so does the requests speed.

[/b] This explains partly why the A64 scales so well with clockspeed.

 

[DAPUNISHER]

Originally posted by: Zebo
Yup bandwitdh is where it's at. Look at the recent anandtech A64 memory benches to confirm too.
http://www.anandtech.com/memor...oc.aspx?i=2226&p=9

OK brudda, you know I'm not the sharpest knife on the magnet rack sometimes so explain to me why the differences between 10x240 480DDR and 9x267 533DDR are nominal but a smaller difference of 9x267 533DDR and 9x278 556DDR is much larger if it is the bandwidth and not the extra 100mhz clockspeed doing the trick? A full 3 seconds faster in SuperPi 2M with less of a HTT/LDT and DDR boost over the previous 2 settings@2.4ghz which resulted in the same superPi score and closer FPS points to the extra 100mhz clockspeed having the most impact doesn't it?

EDIT: I imagine you were indicating the difference between stock 400DDR 200HTT and 267HTT and 533DDR so I see your point about that when clockspeed doesn't change.

 

[StriderGT]

Originally posted by: DAPUNISHER

Originally posted by: Zebo
Yup bandwitdh is where it's at. Look at the recent anandtech A64 memory benches to confirm too.
http://www.anandtech.com/memor...oc.aspx?i=2226&p=9

OK brudda, you know I'm not the sharpest knife on the magnet rack sometimes so explain to me why the differences between 10x240 480DDR and 9x267 533DDR are nominal but a smaller difference of 9x267 533DDR and 9x278 556DDR is much larger if it is the bandwidth and not the extra 100mhz clockspeed doing the trick? A full 3 seconds faster in SuperPi 2M with less of a HTT/LDT and DDR boost over the previous 2 settings@2.4ghz which resulted in the same superPi score and closer FPS points to the extra 100mhz clockspeed having the most impact doesn't it?

EDIT: I imagine you were indicating the difference between stock 400DDR 200HTT and 267HTT and 533DDR so I see your point about that when clockspeed doesn't change.

Precisely the Mem increase % impacts around 1/10th to 2/10th of the same CPU increase % in most real life benchmarking scenarios (mem bandwith benchmarks excluded - which are the mosts inconsistent with real life performance benchmarks eg. P4...)
This is why I believe that the obssession with low latency, high bandwidth Memory is clearly for record breaking purposes and nothing more. Max real impact of best DDR200 module to the worst one all else equal: 3-5%...

 

[RussianSensation]

Originally posted by: StriderGT

PS1 Hasn't anybody understood how to use the calculator yet?!?!?!?!?!?:Q
PS2 This analysis answers your question too, Russian Sensation

I think the reason why people are consufed, including me, is that for the life of computer existence, ratios meant just that.

If P4 ran at 300FSB and multiplier of 12 your CPU speed was 3600, with a FSB of 300 and if you set CPU:RaM ratio to 3:2, your memory would run at 300FSB/3*2 = 200FSB (or 400mhz effective). If you set it to 5:4 ratio, your memory would run at 300/5*4 = 240mhz.

Now the confusing part is that you are saying that all A64 motherboards do not work that way....and that the ratios don't really mean what they mean....because you are now using the multiplier to figure out the memory speed, instead of using the FSB setting.

Strider, have you used CPU-Z to figure out if this is the case? Because if you just set 9x270 and say a ratio, you'll know what speed the memory is running at to test this.

 

[Zebo]

Originally posted by: DAPUNISHER

Originally posted by: Zebo
Yup bandwitdh is where it's at. Look at the recent anandtech A64 memory benches to confirm too.
http://www.anandtech.com/memor...oc.aspx?i=2226&p=9

OK brudda, you know I'm not the sharpest knife on the magnet rack sometimes so explain to me why the differences between 10x240 480DDR and 9x267 533DDR are nominal but a smaller difference of 9x267 533DDR and 9x278 556DDR is much larger if it is the bandwidth and not the extra 100mhz clockspeed doing the trick? A full 3 seconds faster in SuperPi 2M with less of a HTT/LDT and DDR boost over the previous 2 settings@2.4ghz which resulted in the same superPi score and closer FPS points to the extra 100mhz clockspeed having the most impact doesn't it?

EDIT: I imagine you were indicating the difference between stock 400DDR 200HTT and 267HTT and 533DDR so I see your point about that when clockspeed doesn't change.

😉 Read what always I say in all those memory reviews.

Posted on Oct 11, 2004 at 11:07 PM by Zebo Reply
Would you guys *PLEASE* test some budget ram like crucial 8T to show what a ripoff this boutique stuff is price/performance wise?

Posted on Oct 11, 2004 at 11:27 PM by Zebo Reply


It's almost like AT only reviewing FX's and EE's on the processor side.

I really feel AT is doing a diservice to the community by continually pimping this overpriced RAM in every review. Even for overclcokers this holds true, since much budget ram scales the same as the boutique ram when pushed..albeit with mybe a little looser timings and a little slower.

But sure as heck ain't 100% slower to justify boutiques ram 100% price premium. Especially when most users are on fixed budget and thier money is better spent on a better video card, more HD space or something else.

But comming to AT, as a builder, you'd think this overpriced RAM is your only choice since that's all they like present and are getting your budget jammed on the front end for almost nothing in return.🙁

Posted on Oct 1, 2004 at 4:49 AM by Zebo Reply
spensive!🙁

p/p is horrendous for this stuff. It's too bad you don't include micron/crucial 8t in there

etc ditto StriderGT

It' all about latency and given the Athlon 64 architecture and the integrated memory controller which is so low in the first place there is no statisticlly signifigant advantage to super high bandwitdh over low with decent timmings. Which is also why 939 (double bandwidth) I always laugh at peoples facination with too, in rubukeing 754.

 

[StriderGT]

Originally posted by: RussianSensation

Now the confusing part is that you are saying that all A64 motherboards do not work that way....and that the ratios don't really mean what they mean....because you are now using the multiplier to figure out the memory speed, instead of using the FSB setting.

Strider, have you used CPU-Z to figure out if this is the case? Because if you just set 9x270 and say a ratio, you'll know what speed the memory is running at to test this.

Not only does CPU-Z agree with me (and vice versa 🙂) but under the memory tab you get to see the infamous by now CPU/Mem Divider.

Pretty much agreed on everything Zebo but even DDR3200 at the holy 2-2-5-2 is as I said before at BEST 3-5% better than the vanila 3-3-8-3 in real world applications. You can not justify the low latency price premium either.

 

[beatle]

I'll have to check CPU-Z's memory tab for a divider value. I have no idea how to set the speed of my memory on my A8V accurately. I just know that it's running below 200mhz.

 

[Spikesoldier]

jack up the vdimm and loosen the latencies and choose a different divider to see if you can increase your memory bandwith.

or buy new memory, 1:1 would be screaming

you have a good chip i envy you

 

[Duvie]

I tried cas 2.5 at lowest possible settings at 2600mhz for 480ddr and no go...errors in superpi...also tried 2.85 vdimm and no go...Also tried slots 3&4 and they were worse....

I think I would have to go back to like 2500 to get stable memory speeds with 166....

My goal is now to get my temps donw and see if I can find a way to get 1.6v and not 1.63v and try to get 2700mhz prime95 stable....At 1.57v seen in link it ran for 1 hr before error (rounding)..It needs a touch more vcore but it does a big leap in between...maybe flash to 1.41 will help me get those in the middle vcores...


2700mhz cpu_z

 

[Dethfrumbelo]

Originally posted by: Zebo

It' all about latency and given the Athlon 64 architecture and the integrated memory controller which is so low in the first place there is no statisticlly signifigant advantage to super high bandwitdh over low with decent timmings. Which is also why 939 (double bandwidth) I always laugh at peoples facination with too, in rubukeing 754.

After re-reading the article it sure as Hell looks like memory bandwidth is nearly irrelevant to the A64 except in benches like Sandra. You're right - dual channel yields basically no real world improvement over single channel.

It seems there's not much point to making a fuss over memory when it comes to the A64. I'd like to see how low memory bandwidth would have to go before it actually started to hurt performance.

 

[divide by zero]

This is a fantastic thread. I'm getting ready to build a system and was trying to figure out which memory and power supply etc to buy.

From this and other posts & reviews it seems that as long as I get memory that will run 1T I'm good. I can loosen the timings and increase the memory divider as much as I need to support as high an OC as my A64 can handle.

In one of the recent memory reviews it mentioned that a really high memory OC was not possible until they upgraded the power supply to one that had more amps on the 12v rail. I may get one of those anyway for future proofing and overall stability but if I don't care about OCing my mem to insane levels then it is not a neccessity.

For future upgradability I'll stick with the 939, the MSI Neo2 Platinum.

Also, if I'm not worried about OCing the memory then I'm less hesitant to go with 2x1G or with 4x512M, whichever is cheaper.

So, on a board with PCI and AGP locked and the memory controller on the CPU, what else would hold back how high you could set the FSB? Would that only affect the CPU and the chipset?

 

[divide by zero]

Does anyone know what might limit raising the FSB on a motherboard using an A 64?

 

[StriderGT]

Originally posted by: divide by zero
Also, if I'm not worried about OCing the memory then I'm less hesitant to go with 2x1G or with 4x512M, whichever is cheaper.

Stick with 2x1G, its going to be much easier to keep it stable when you OC... (4x does not officially-according to th MSI K8N Neo2 Plat. manual- support 2x200FSB only 2x166=333)

Originally posted by: divide by zero
So, on a board with PCI and AGP locked and the memory controller on the CPU, what else would hold back how high you could set the FSB? Would that only affect the CPU and the chipset?

A bit of luck, all chips are not created equal, although by taking into account all these parameters you certainly maximize your potential 🙂

Originally posted by: divide by zero
Does anyone know what might limit raising the FSB on a motherboard using an A 64?

You can get an idea by looking at the 1T BIOS Tests worksheet of the calculator, but because the hardware configurations, even the board revisions change your mileage may vary...

 

[divide by zero]

Thanks for the reply. I'll have to play with it when I get my new rig.

With a perhaps unlimited FSB who knows what can be achieved!

 

[chilled]

Originally posted by: divide by zero
This is a fantastic thread. I'm getting ready to build a system and was trying to figure out which memory and power supply etc to buy.

I agree 100% with that!

I too am looking to buy the K8N Neo2 Platinum for use with a 3200+ CPU. However, with me being a student (and upgrading from a XP 2700+ as it is), I would like to reuse my Crucial 512MB PC3200, bought a few months ago. I plan to buy another such stick from Crucial to get 1GB total. I have a limited budget (£350 is already a stretch for me!).

So let me get this straight - I wanna OC and so by keeping the ram to say 240Mhz wouldn't hurt me too much right (as much as other PCU architectures), even if it's not at a 1:1 divider?

 

[chilled]

^ for a quality thread