PDA

View Full Version : Memory question


Iblis
08-07-2002, 08:39 PM
Somebody's gonna roll in here and make me feel really stupid for asking this, BUT, :D If memory on video cards can hit 300 MHz (600 effective), why doesn't this just easily translate over to sytem memory? With DDR being relatively well-produced these days, and I expect yields are high as ever, shouldn't we see PC4800 modules being commonplace? Am I missing something? Is it that video cards can support higher voltages and the standard for DDR is 2.5V ? :confused:

Agent666
08-07-2002, 08:48 PM
I think the technology is there to do it but while not a lot of benifit is to be gained by having faster main memory esp if you consider the cost of the modules....... but an increase in vid memory speed = a real percievable increse in a computers speed ..... and since the vid card and HD are bottlenecks in most modern systems it makes sense to eliminate these

.... hopefully that made some sense?

dumass
08-07-2002, 09:06 PM
u also have to take into effect that there are no CPUs that have a bus of 300 mhz

Ass seen with the VIA KT333, having memory running faster than the CPU FSB = not much preformance increase.

KingJackal
08-07-2002, 09:20 PM
Originally posted by dumass
u also have to take into effect that there are no CPUs that have a bus of 300 mhz

Ass seen with the VIA KT333, having memory running faster than the CPU FSB = not much preformance increase.

That's the main reason. CPU's just can't use the extra speed that memory would provide. In order to use it, for example - Athlon's would need to be run at a 300MHz fsb. And given most people have problems reaching 200MHz..... :D

The other big reason ( as alluded to ) is cost.

Would you pay $500 for 256MB of RAM? I think not.... ;)

Volodkovich
08-07-2002, 09:31 PM
isnt it also due to the mobo not physically being able to run that fast becasue it is bigger than a vid card?

Iblis
08-07-2002, 09:36 PM
Originally posted by KingJackal


That's the main reason. CPU's just can't use the extra speed that memory would provide. In order to use it, for example - Athlon's would need to be run at a 300MHz fsb. And given most people have problems reaching 200MHz..... :D

The other big reason ( as alluded to ) is cost.

Would you pay $500 for 256MB of RAM? I think not.... ;)
Couldn't the P4's benefit from all that B/W? Isn't that what RDRAM is all about?

KingJackal
08-07-2002, 09:36 PM
Originally posted by Volodkovich
isnt it also due to the mobo not physically being able to run that fast becasue it is bigger than a vid card?

nope.... cos the RAM slots are normally pretty near the CPU, aren't they? :D

Trace length is a concern - but that's not the reason.

Iblis
08-07-2002, 09:37 PM
That was weird???

Volodkovich
08-07-2002, 09:39 PM
true...i was thinking about the whole clock..i.e. running the noard at 400mhz

KingJackal
08-07-2002, 09:41 PM
Originally posted by Iblis
Couldn't the P4's benefit from all that B/W? Isn't that what RDRAM is all about?

Yes - PC1066 RDRAM runs at 533MHz DDR ( 1066MHz effective ), and as such can fully fill the P4's bus with info. Which makes P4's and RAMBUS a very fast pair indeed.

However, 533MHz DDR is quite a step up - and would be VERY expensive in system memory size modules ( 256MB/512MB/1024MB ).

Deviant
08-07-2002, 09:45 PM
My $0.02c

It's the mobo's that cant do 600 MHz.. It was fine when we had 286's and 386's and the RAM was synchronous, but CPU manufacturing technology went faster than RAM. Didn't need faster mobos unless you had faster RAM, and RDRAM and DDR has suddenly changed that.

They can make RAM that fast, but the price for the quantities would be huge. Just think of getting RAM at 2-3 GHz, yes they could do, with the same manufacturing techniques as on XP's and P4's, but the cost would be huge.

Mobo's are the limitations, but they could even make them faster too. It really comes down to price vs performance.

For the sake of adding 64-512kB of on die cache you can dramatically increase performance, but even if you had 1-2MB of cache the extra performance would be small compared to the extra price.

If it were worth the cost, I think they would do it. After all we have already seen KT333 out when AMD CPU's can't even run 166 MHz FSB with out overclocking.

Iblis
08-07-2002, 09:47 PM
Originally posted by KingJackal


Yes - PC1066 RDRAM runs at 533MHz DDR ( 1066MHz effective ), and as such can fully fill the P4's bus with info. Which makes P4's and RAMBUS a very fast pair indeed.

However, 533MHz DDR is quite a step up - and would be VERY expensive in system memory size modules ( 256MB/512MB/1024MB ).
Actually I meant like 266 MHz (533 DDR), but now I see why the RDRAM is more favored. I thought that the P4's FSB was not a "true" 400/500 MHz, so it couldn't use all that bandwidth from the RDRAM, but I guess it can.:)

KingJackal
08-07-2002, 09:52 PM
Kind of. But memory speed hasn't kept up with processor speed.

They COULD NOT make 3GHz RAM if they tried.

...the gap between memory and processor speeds is growing. That's why they introduced processor caches ( a smaller, faster memory that came closer to processor speed ). Then the gap grew some more, and they split that into L1 and L2 caches ( normally L2 is slower and bigger than L1 ). L3 caches aren't far away - some SPARC architectures already utilise L1, L2 and L3 caches.

But yeah - Deviant is right in saying that the performance just isn't worth the cost. And at the end of the day, it's that bang/buck that they're looking at as well....

Deviant
09-07-2002, 07:41 AM
Originally posted by KingJackal

They COULD NOT make 3GHz RAM if they tried.



What I was meaning her KJ, was theorectically it would be possible to make Athlon XP's etc with 256 MB of L2 cache, or 512MB of L2 cache etc negating the requirement for any main memory at all cause it would all be on the chip itself. Possible but not practical and the expense would be huge. Would it work faster, maybe 10-20% and not worth the cost.

When we had 100MHz (10ns) processors, and 70ns RAM cache became a big issue cause RAM was 7x slower than the speed of the CPUs. So a small cache at 10ns gave an equavilant memory speed of around 15ns, which was good. If you doubled the cache you might get the equavilent memory speed down to 14ns, but not worth the expense at the time. That's what we see today (only the speed are a little different), and probably why we havent seen XP's with more cache.

KingJackal
09-07-2002, 08:48 AM
Originally posted by Deviant
What I was meaning her KJ, was theorectically it would be possible to make Athlon XP's etc with 256 MB of L2 cache, or 512MB of L2 cache etc negating the requirement for any main memory at all cause it would all be on the chip itself. Possible but not practical and the expense would be huge. Would it work faster, maybe 10-20% and not worth the cost.


Ah - gotcha. Yeah, that would be fast.....

( and lol @ the idea of $100,000 CPU's )

In fact - Athlons are a particularily strange example, in that their L2 cache is full speed along with their L1.

The same cost/performance principle continues out to the hard drive ( which interestingly now also has yet another cache between main memory and disk ). I wonder if we'll see L2 cache's on hard drives sometime - because hard drive speeds are growing even slower than memory speeds. ( there's a growing memory/disk speed gap, the same as the memory/processor gap )

Agent666
09-07-2002, 11:03 AM
what they should do is give the athlon a 166MHz FSB ..... to gain a bit of mem bandwidth back on those P4s

Deviant
09-07-2002, 05:16 PM
Originally posted by Agent666
what they should do is give the athlon a 166MHz FSB ..... to gain a bit of mem bandwidth back on those P4s

I know, and they are with the MP, but not the XP. But it doesn't matter to overclockers who unlock their chips anyway.

Agent666
09-07-2002, 07:21 PM
since when has the MPs had 166fsb?????????

I dont think so tim...... most dual boards have some 66MHz pci slots..... not getting confused with that are you??

Agent666
09-07-2002, 07:22 PM
also maybe you mean that the MPs multipliers are unlocked??

Volodkovich
09-07-2002, 07:40 PM
i think Deviant means the tbred MP's that arent out yet.

mird-OC
09-07-2002, 09:32 PM
yes, tbred MPs = 166MHz FSB & unlocked. mmmm.

Deviant
09-07-2002, 10:01 PM
Originally posted by mird-OC
yes, tbred MPs = 166MHz FSB & unlocked. mmmm.

Yep, Mird and Volodkovich are on to it. They have sampled and are out there, just not to the public yet. But suposidly AMD have confirmed they will ship unlocked, which make them attractive to the overclocker for a few dollars more, that way you haven't voided your warranty. I am guessing, but $40US = 80 NZ mre than current XP's for the same speed, but we may have to wait and see.

Agent666
09-07-2002, 10:10 PM
so I take it they will need a new MB to run them??....... but damn I didnt know that...... thanks.......

Volodkovich
09-07-2002, 10:13 PM
<deleted>

Iblis
09-07-2002, 10:14 PM
Hey Deviant, ya got any links? I kept postulating for a long time AMD would make the move to 166 FSB, I just have yet to see it in writing...

Volodkovich
09-07-2002, 10:14 PM
no, they should run in n e socket A mobo that a Xp tbred can run in. >>all Kt33 and kt266a's.

Deviant
10-07-2002, 07:46 AM
Originally posted by Iblis
Hey Deviant, ya got any links? I kept postulating for a long time AMD would make the move to 166 FSB, I just have yet to see it in writing...

Google is your friend, the links you are after are everywhere.
I will look for a specific one if you can't find one, but I have to got to work now cos I'm going to be late, again.

Iblis
11-07-2002, 04:22 PM
Originally posted by Deviant


Google is your friend, the links you are after are everywhere.
I will look for a specific one if you can't find one, but I have to got to work now cos I'm going to be late, again.
I REALLY did look. And I REALLY couldn't find anything. (in English)

Deviant
11-07-2002, 05:35 PM
Originally posted by Iblis

I REALLY did look. And I REALLY couldn't find anything. (in English)

It was really a conspiricy to trick you, but we couldn't get it past you could we.

Nah the real reason is I have spent 30 mins trying to find it myself and can't, so believe it or not, at least I'm not the only one who has read it.

That doesn't mean I believe it 100%. Until I see it, or it comes out, or is posted that way on WWW.AMD.COM it may not be true.