INTEL - Synchronous is not important

[Jeff7181]

If you're getting similar performance by running asynchronously, why not just run it synchronously so you don't have to overclock your CPU as much?

Because more CPU intensive tasks can use the extra Mhz and not be hurt by memory bandwidth. But you're right, my point is that his conclusion is wrong.

 

[Tom]

Originally posted by: Jeff7181

Originally posted by: Dead Parrot Sketch
Stevejst- I agree with your conclusions pretty much. For instance my system is using cheap Samsung pc2700 ram and a 3:2 ratio, but because it's running at 280fsb it gets benchmark scores better than a stock 3.2g system using a 1:1 ratio and pc3200.

And it's a helluva lot cheaper.

A 3.36 Ghz processor is faster than a 3.2 Ghz processor... imagine that.

Hey, if I slow down to 266FSB, which is 3.2gig, it is still faster than a 3.2g running cas2 pc3200. even in memory bandwidth tests. And in other apllications the performance difference is even more pronounced.

And it's humongously less expensive.

 

[Jeff7181]

Originally posted by: Dead Parrot Sketch

Originally posted by: Jeff7181

Originally posted by: Dead Parrot Sketch
Stevejst- I agree with your conclusions pretty much. For instance my system is using cheap Samsung pc2700 ram and a 3:2 ratio, but because it's running at 280fsb it gets benchmark scores better than a stock 3.2g system using a 1:1 ratio and pc3200.

And it's a helluva lot cheaper.

A 3.36 Ghz processor is faster than a 3.2 Ghz processor... imagine that.

Hey, if I slow down to 266FSB, which is 3.2gig, it is still faster than a 3.2g running cas2 pc3200. even in memory bandwidth tests. And in other apllications the performance difference is even more pronounced.

And it's humongously less expensive.

Right, but you're using a 266 Mhz FSB rather than 200. You're not even considering the topic of this thread. Synchronous is important. I also don't give a flying cow turd about memory bandwidth numbers. Look at the performance of the system as a whole, don't single out one number. That's why I asked steve to do some real world benchmarks like with UT2k3. And he won't.

 

[DAPUNISHER]

Originally posted by: stevejst
And don't let anybody tels you that timings are not important. They are very important.
As said 2-3-2-6 at 262/209.6 gives me 5733/5703. If I relax it to 2-4-4-7 (CAS the same!) the loss is already obvious 5608/5629. To ofset this you have to increase the clock to 265.
Relaxing CAS will make this go significantly lower.

Sandra isn't a very good real world performance indicator ( although it does give a basis for comparison we can all relate to ) and the differences between the 2 bandwidth scores is inconsequential in terms of perceptable difference in the vast majority of tasks. If one does not look at the benchmark results then the difference between the timings tested becomes virtually undetectable otherwise.

 

[yak8998]

I agree with WobbleWobble:

And to draw a conclusion off of only two benchmarks is wrong. You need to run a wider variety of benchmarks to draw a conclusion. It's kind of like if you only look at two benchmarks where P4 beats an Athlon and conclude that a P4 is better, or vice versa.

But what you have here so far looks solid steve.

 

[Tom]

"Right, but you're using a 266 Mhz FSB rather than 200. You're not even considering the topic of this thread. Synchronous is important."

Frankly, I don't understand what it is you think should be tested ? You want to test at the same FSB and the same processor speed and the same memory speed, and then test at different memory ratios ??

If you think about it you would realize that is completely impossible. There is no way to change the memory ratio and keep the same cpu speed without changing either the FSB or the memory speed. Obviously if you change the memory speed to a slower memory speed the system will be slower. BUT THAT IS BECAUSE THE MEMORY IS SLOWER, it has nothing to do with synchronicity.

 

[Jeff7181]

It's not that complicated... think about it...

You have a 2.4C with a multiplier of 12 and a FSB of 200.

You overclock the FSB to 250 Mhz, and your clock speed is 3.0 Ghz.

If you only have PC3200 RAM, and run it async at it's intended speed, you're creating a bottleneck, because the FSB is running faster than the RAM and will have to wait periodically.

If you have PC4000 RAM and run it in sync, there's no bottleneck because the FSB will not have to wait for the RAM.

Steve said "syncronous is not important." Yes it is. Running async WILL hurt the performance as opposed to running in sync because running async means you have less memory bandwidth... well, unless you're weird and overclock the RAM but not the FSB.

I understand you're saying it has to do with memory speed and thus the amount of memory bandwidth available... I'm saying that's what happens when you run async, you have a lesser amount of memory bandwidth than what is needed.

Would you agree that running a 3.0 Ghz processor on a 250 Mhz bus but with the RAM only at 200 Mhz is slower than if you had a 3.0 Ghz processor on a 250 Mhz bus with RAM running at 250 Mhz? And to catch up to the performance of the 3.0/250/250 you might have to overclock the FSB even more to say... a 275 Mhz bus which would yield 3.3 Ghz with the slower 200 Mhz PC3200 RAM?

This is all I'm saying. A 3.0 Ghz processor on a 250 Mhz bus with RAM that can run in sync will most likely perform better than a 3.2 Ghz processor on a 266 Mhz bus with 200 Mhz RAM. If you can show me benchmarks that show otherwise, I'd love to see them. And I'm not talking about memory bandwidth... I'm talking about real benchmarks... UT2k3... Quake3... divx encoding... file compression.

 

[KillaBong]

I think that it's just a different look at what "important" is. Most anyone will be able to run their pentium at a higher clock if it is asynchronous so therefore they wouldn't run it synchronous at a lower speed. That would clearly have less performance and therefore running synchronous isn't important because you are almost certain to get a higher clock. On the other hand, If you can't get it any higher, of course it would be important to keep them synchronous because it will be slightly faster. Basically everyone is guaranteed to get a higher clock speed, and therefore the majority of the time it isn't important to purposfully keep them synchronous. Now it might not quite be right to say it isn't important under any circumstance, because it wouldn't be true. I think the main message he is trying to convey is that people will try too hard to keep it synchronous when they don't realize that they can easily make up for the performance by simply changing the ratio.

 

[Jeff7181]

Originally posted by: KillaBong
I think that it's just a different look at what "important" is. Most anyone will be able to run their pentium at a higher clock if it is asynchronous so therefore they wouldn't run it synchronous at a lower speed. That would clearly have less performance and therefore running synchronous isn't important because you are almost certain to get a higher clock. On the other hand, If you can't get it any higher, of course it would be important to keep them synchronous because it will be slightly faster. Basically everyone is guaranteed to get a higher clock speed, and therefore the majority of the time it isn't important to purposfully keep them synchronous.

That's not what this post is about though... the title says "INTEL - Syncrhonuous is not important"

Now if it said "INTEL MAXIMUM OVERCLOCKING - Synchronous is not important" then I wouldn't have argued it at all. But that's not what it says. When you run the FSB and RAM asynchronous, you introduce latency... whether it's Intel, AMD, Apple, Cyrix, or Playschool. Higher latency is never a good thing in respect to computers. You can overcome the negative effects of increased latency by raising the speed, but again, I'm not saying you can't... in fact, that's what I'm pointing out... you have to run the processor faster in order to compensate for the increased latency caused by running the RAM and FSB asyncronously.

*EDIT* As I pointed out before, when you start using RAM dividers, you're heading back into the 533 Mhz bus territory of the processor being starved for memory bandwidth. And when PC4000 and faster RAM becomes more available, the Intel guys are gonna start saying "look, P4's are good, we gain lots more speed by using faster RAM" ... well duh, that's because you handicapped it to begin with by using a memory divider.

 

[Tom]

"Would you agree that running a 3.0 Ghz processor on a 250 Mhz bus but with the RAM only at 200 Mhz is slower than if you had a 3.0 Ghz processor on a 250 Mhz bus with RAM running at 250 Mhz? "

Of course ! Ram running at 250mhz is faster than ram running at 200mhz. Obviously.

In real tests though, people have tested the following scenario, processor running at 200fsb and memory ratio of 1:1, 5:4, and 3:2, which gave results of about 2% slower for the 5:4 and 5% for the 3:2, across a wide range of benchmarks..

And if you consider that the memory speed was slower when using the ratios, the corresponding drop in performance is because of the slower memory speed, there isn't any penalty other than that. No additional burden because it's asynchronous, no additional slowdown because the processor is waiting for the memory.

If either of those things were happening there should be a bigger difference in performance. In the case of 5:4, memory speed is what 20% slower ? And there is only a 2% drop in performance ?

And the 3:2 ratio, memory is 30% slower but there is only a 5% drop in benchamrks.

What I think this means is in the real world what with processor cache, etc, and amount of data that is typically transferred, that most of the time there isn't any need for more bandwidth than you get with a ratio.

And this all ignores the fact that it's possible to build a faster systen for less money, as I described above, which I think is more "important".



Thread with some good info. Look for chart called memory ratio...

 

[Jeff7181]

You're totally missing the point, so I'm not even gonna try to explain it further.

 

[THUGSROOK]

Jeff ~ ill try to answer your question/bolster your point, im using a 2.4b

for me, 1:1 382ddr is faster then +4:5 438ddr even with both running @ c2622-1.
now if 1:1 is faster then +4:5, does it not make sense that 1:1 will be faster then a -5:4?
(disreguarding cpu OC of course)

1:1 is faster, anything else would be a comprimise for best performance.

HTH

 

[Jeff7181]

Originally posted by: THUGSROOK
Jeff ~ ill try to answer your question/bolster your point, im using a 2.4b

for me, 1:1 382ddr is faster then +4:5 438ddr even with both running @ c2622-1.
now if 1:1 is faster then +4:5, does it not make sense that 1:1 will be faster then a -5:4?
(disreguarding cpu OC of course)

1:1 is faster, anything else would be a comprimise for best performance.

HTH

That's what I'm saying... by using a divider you're handicapping the performance of the system as a whole because if you're running 5:4 you're starving it for memory bandwidth. Now I'm not saying don't use a RAM divider. Only an idiot wouldn't be able to see that you need to use a RAM divider with current RAM technology. There isn't RAM capable of more than 250 Mhz reliably.

It's like taking cylinders away from a car engine... sure you can rev the engine higher with a ligher rotating mass... but you don't see benefits right away, you have to rev the engine higher... and there will come a point when the V6 is more powerful than the origional V8. And then you still don't have the torque of the V8, so you have to raise the power band up higher in the RPM range to make the same power, and exceed the power of the V8. If the engine isn't capable of revving that high, then you switch back to the V8.

Same priciple for overclocking... If you can take a 2.4C to 3.0 Ghz keeping everything in sync, but can get it to 3.4 with it async, you have to test it and see if that extra 400 Mhz is pushing the RAM hard enough so that the total system performance is greater than that of it at 3.0 Ghz with everything in sync.

 

[THUGSROOK]

Jeff ~ and my 2.4b proves your point perfectly since not only am i able to run 1:1 with a 3.4+ghz OCed chip, but i am able to do it with every single memory tweak enabled full blast @ a nice cozy 191fsb 382ddr (and on cheap ram)

when running async you lose the abilty to enable most of the memory tweaks by default.

 

[eva2000]

Originally posted by: stevejst
There is a common misconception in overclocking community that sychronuous bus and memory clock is extremely important for memory bandwidth. That is true in AMD case but not in Intel case.
Here is a little test you can easy replicate yourself if you are running a potent Intel board. This is on IS7-E using 2x256 GEIL Ultra Platinum PC 3500 at 2-3-3-7 timings:

Syncronuous 1:1 using 210/210
Sandra: 5052/5091

Asyncronuous 5:4 using 236/189
Sandra: 5089/5063

As you can see the bandwidth is about the same when the CPU bus increases 26 and memory bus decreases 21.
So there is a certain loss due to waiting times (2-3 MHz) but that is negligible. In fact you can easily ofset that loss by using more aggressive timings that lower memory clock makes available.

Similar tests you can perform with all the clocks and timings you can come up with and it will all reach the same conclusion. This is all on 2.4 GHz HT Pentium 4.

If you want performance you need to find your own soft spot. I like mine at 262/209.6 using 5:4 that gives me 3.144 GHz and memory bandwidth of 5733/5703 in Sandra at 2-3-2-6 timings. I can get close to 6000/6000 but it is not worth the stress on the CPU in my view.

Again it depends on the memory use, size, at what particular speed, timings, and on what board and bios is used read my IC7 PC3700 Gold async vs sync benchmarks here

Now compare that to async 272FSB/218mhz scores on 4PCA3+ and my 2x 512MB XMS3500C2 here

Reason why low 4PCA3+ scores is due to the particular board and it's bios memory enhancements compared to my IC7 and PX865PE PRO2

 

[TerryMathews]

Also keep in mind that running a divider will introduce extra latency relative to the difference between FSB and RAM clock. Latency, unfortunately, is difficult to test yet easy to notice.

Layman's terms: the higher the divider, the more the latency.

 

[THUGSROOK]

Originally posted by: TerryMathews
Also keep in mind that running a divider will introduce extra latency relative to the difference between FSB and RAM clock. Latency, unfortunately, is difficult to test yet easy to notice.

Layman's terms: the higher the divider, the more the latency.

actually,
those latencies are user adjustable on springdale/canterwood motherboards.
but like i said, you lose that ability when running async.

thats right ~ you can actually OC the northbridge latencies! :evil:

HTH

 

[stevejst]

Here is another memory bench PCMark2002, running 2.4C with 2x256 Geil UP PC 3500 at 2-3-2-6 at

210:210 - 8155
240:192 - 8222

Alright, this is as much time I have to waste.
PCMark2002 say to me that synchronuous is not good for
6 MHz = (240+192)/2 - (210+210)/2
Sandra told me it is about 2-3 MHz, 3DMark says about 3.5 MHz. And notice that PCMark is not exactly the most reliable test bench there is, by a long shot. I won't run UT because I do not see a point testing processor and graphic card for this argument, beside I don't have time.
How much is all this important?
Lowering CAS from 3 to 2 will be worth anywhere between 6 and 12 MHz depending on your system. That easily erases any loss due to asynchronicity.
Actually I am the only one posting relevant numbers for those of you complaining about not enough tests.

And of course if you have memory that can run 500 MHz at aggressive timings (2-3-2-6, that would be enough only for synchronuous at 250), for Christ sake use it, but how much is it and where can you find it? Or if you can afford buy Pentium 3 GHz or 3.2 GHz.

 

[stevejst]

And again, I am not saying asynchronicity is completely the same as synchronicity. Not it is not, I'd say synchronicity is worth anywhere between 2-6 MHz, depending on the test bench. But that is a joke since running aggressive timings erases that easily. Not talking about gain on the CPU being able to run 3.2 GHz instead of 2.6-2.7 GHz.

THAT IS THE WHOLE POINT!

In case of AMD processor running asynchronuous is a complete loss, there is absolutely no gain running asynchronuous, since the processor clock lags behind the memory. Shortly AMD CPU cannot gain anything from a fast memory. Some of us that used the same fast memory in AMD PCs and Intel PCs already know that well.

That is why HT Pentium rules and will continue to rule for the foreseeable future.
The better memory comes out the bigger the difference between AMD and Intel will get. Opteron with its onchip controller is unable to come close, neither will Athlon 64, whether the 754 (single channel) or the 940 (dual channel) one. The coming Prescott will even increase that difference since the Grantsdale will be able use DDRII. As far as I am concerned, and the way it looks now, AMD CPU is not going to be in any PC I'll build in 2004.

Another thing: dual-channel on AMD platforms is a joke as single channel Soltek NV400-L64 and VIA KT600 proved that clearly. Even on the same board it is worth 5-8 MHz that can be easily erased by timings. Note that dual channel degrades the speed of the memory so overclocking is affected, even more so on AMD platforms.

 

[Jeff7181]

steve... Are you slow or just stupid?... If you want to turn this into an AMD vs. Intel battle again, we can do that...

Shortly AMD CPU cannot gain anything from a fast memory.

No sh!t, it's because even the 200 Mhz FSB processors aren't limited by memory bandwidth because the FSB can run in sync with PC3200 RAM!

Opteron with its onchip controller is unable to come close

Right, at 2.25 Ghz on pre-production hardware it passes the ability of your machine by a good 300 MB/s.

Link

Might also wanna take a look at this... since it's on THG which is normally Intel biased.
Link

This would be some good reading for you too since you're so ignorant about real world performance and all you can think about is Intel's marketing scheme that bigger numbers translate to better performance.
Link Specifically check out the last page which shows 64 bit performance.

 

[calfpheen]

well, i have a good example for you guys.... i had my 2.6c @ 3.11ghz, 239 fsb, 5:4 ratio, ram is 3500 was running at 191mhz. Got about 340-350 fps in quake 3 benchmarks. last night, i didnt want to push my vcore like i had been, so i tried some 1:1 ratio benchmarking. im running at 2.91ghz right now, 224fsb/224mhz ram (448mhz). I get 350-360 fps. Used same timings, gat settings, which are 2.5-8-4-4 and from top to bottom gat: auto/normal/auto/disable/disable. mem scores in sandra were the same, cause all i did was change the ratio, but my cpu arith was a lot lower, but i got more fps. so yeah, running both the fsb/ram NSYNC ( blahah! ) is faster then running things off. Cause of my unluckyness, to hit 3.11ghz fully stable i had to use a 1.65vcore, i felt unsafe doing that and decided to try a 1:1 ratio, ended up getting a lil more fps, off stock voltage. Thats a nice tradeoff in my opinion. Once i get money for faster ram, im going to do so, seeing how my ram is the bottleneck of my max oc. For now, im going to stick with the 1:1 ratio, i get a lilttle more fps and my cpu is at stock vcore.

 

[stevejst]

Yep, we'll see what comes out in reality. Opteron 2.2 GHz processor running at 2.25 GHz with 2 gig ECC memory on a "demo" motherboard is not available anywhere so AMDzone can post anything they want. Quad-Opteron 844 server is about $3,000 just for the four processors, not mentioning anything about the rest of the setup. Beside I cannot find Opteron 246 available anywhere.

Similar kinds of numbers AMD cheering squad has been posting for 3 years already, I remember it during Compudex last year, but nobody can get their hands on these imaginary computers.

And "your benchmarks" are known to me since you AMD parrots keep on repeating tham repeatedly in your fantasy setups you will be able to afford in 2010.

The reality is that your grandma can buy Dell Pentium and beat the hell out your overclocked Barton "3200+".

 

[calfpheen]

forgot to mention this... if i could run a higher fsb at stock vcore, or at least something under 1.6v, i would. but i cant, so i figured, why hurt my cpu when i can get same fps leaving it at stock vcore. but yeah, id still use a non-sync setup if i could, cause better performance is given... but in my case, my bottleneck is something i cant fix right now. so ill settle for 1:1

 

[stevejst]

Besides 6000 on Sandra is something I can do right now, both in Int and Float (imagine that!), on my real PC, that costed me about $500.

 

[calfpheen]

btw stevejst, im same guy as batmang. somehow i lost my info for that account..... but yeah. i think im gonna invest in some 4000ddr later on. i didnt get as lucky as you did, but i wish i had.