However the effect was much less dramatic with the A64 then the p4. ( I would however expect to see something simular happen to the k7) And with registered memory on the FX the problem did not exist. Score one for registered memory.
It's not what they tell you, its what they don't tell you!
Absolutely. That's why big-memory servers only use registered memory. The problem is that registered memory is ridiculously pricey for the average consumer... But if AMD will have us believe that we actually need or would greatly benefit from more than 1GB memory (very little users actually need more!), then maybe someone will have to deal with this price issue....
As demand increases price for registered memory will come down. And it is still my position that memory requirements will go up sooner than you would like to have us beleive.
Quote :
ridiculously pricey
Relative. is a $450.00 video card ridiculously pricey?
One of the selling points ( one of, not the only, or not even the most important) of the 64 bit architecture is to be able to address large amounts of memory ( comparitively speaking to 32 bit). What you are starting to see is exactly how pointless this would be without the use of registered memory.
The same people bashing the FX for its need to use registered memory would bashing it for performance penalities when it did if it used non-registered memory.
It's not what they tell you, its what they don't tell you!
As demand increases price for registered memory will come down. And it is still my position that memory requirements will go up sooner than you would like to have us beleive.
Yes, as long as the process of making the registered memory is not too expensive and the company is getting lots of income. If the process is expensive, though, then the company responsible for producing this memory will not keep up and will lose money... And I believe that the latter is the case, but I'm not sure...
Either that, or they change their manufacturing in such a way that their production costs are in agreement with their product prices, so they have income. Which is what I said might happen... But this is all just speculation, 'cause I don't quite know the details... In either case, we - the end users - are going to benefit from this.
Quote :
Relative. is a $450.00 video card ridiculously pricey?
Yessir, I would! at least for me...
Quote :
What you are starting to see is exactly how pointless this would be without the use of registered memory.
Yes, but the problem is that prices are high for it <i>right now</i>, and that also influences demand. So unless they change, I'm not going to buy registered money... I'm not that rich. At least not yet, I'm trying to work around that...
Exactly what is your point? If the registered memory is too expensive for you than the CPU must also be too expensive for you in which case <font color=red> Why do you even care</font color=red>?
Quote :
But this is all just speculation, 'cause I don't quite know the details.
Don't speculate....research. It will make your posts much more relevant. Do you know just exactly what is involved in adding a register to a SDRAM? Basically in its simplest terms its the addition of one very small IC, smaller than one of the Dimms themselves and a few additional traces.
It's not what they tell you, its what they don't tell you!
My anger and my wrath will be poured on this place on man and beast on the trees of the field and the fruit of the ground it will burn and not be quenched Jeremiah 7:20
I am ALPHA and OMEGA, burn!
Read this from Anandtech
<A HREF="http://www.anandtech.com/memory/showdoc.html?i=1839&p=4" target="_new">http://www.anandtech.com/memory/showdoc.html?i=1839&p=4</A>
It says to me that 4 X 256 is the best possible configuration for speeds higher than ddr 400.
They don't seem to mention 4 x 512 though.
In every real-world test, the performance drop with 4 DIMMs on the Pentium 4 is much more significant than the performance drop with 3 DIMMs on the Athlon64. Using tools such as the artificial SiSoftware Sandra 2004 to measure memory bandwidth can be very misleading as you can see above here. The value of this benchmark is seeing the differences in how AMD and Intel handle this increased number of memory banks, with AMD slowing down the bandwidth and Intel increasing latency.
(Taken from the firingsquad review)
Simply translated Sis Sandra does a lousy job of of measuring real world performance.
It's not what they tell you, its what they don't tell you!<P ID="edit"><FONT SIZE=-1><EM>Edited by ncogneto on 10/21/03 09:58 PM.</EM></FONT></P>
<<But if AMD will have us believe that we actually need or would greatly benefit from more than 1GB memory (very little users actually need more!), then maybe someone will have to deal with this price issue>>>
AMD needs to be able to produce a FX chipset that uses bog standard PC3200 ram in a dual channel format if they want the FX to be more popular.
<b>I am not a AMD fanboy.
I am not a Via fanboy.
I am not a ATI fanboy.
I AM a performance fanboy.
And a low price fanboy. Regards,
Mr no integrity coward.</b>
>But if AMD will have us believe that we actually need or
>would greatly benefit from more than 1GB memory (very
>little users actually need more!),
In 1999 I was running 128 Mb; upgraded it to 256 Mb in 2000, to 512 Mb in 2001, and recently to 1 GB. Now, I may not be your average consumer, but neither was I in 1999 when I had 128 Mb.
In 1999 the average new PC was running windows '98 with around 32 MB ram. Now, just a few years later, the average new PC has over 10x that ammount of memory. You really believe if you purchase a new PC now, you will keep your ammount of RAM constant for the next 3 or so years ? I'm pretty sure you will quadruple it.
= The views stated herein are my personal views, and not necessarily the views of my wife. =
Actually, I have several 'off the shelf' pc's from 99, they all have either 64mb (most) or 128mb (a couple-- may have been upgraded) in them. They were pretty middle line business pc's when they were purchased, not top of the line or out of the ordinary by any means (about 1800$ pc's counting monitor... but in 99 that was a pretty good price.)
I had 256 in 99-00 and 1g in 01-present. I do video editing, programming, gaming, etc... and I really don't yet feel a need to go beyond 1g, when I check the pf usage I'm generally under 384mb and have almost half a gig of pm free. I do expect ram usage ammounts to increase, but I don't expect even 1g to become the standard for a couple years yet... late 2005 to early 2006 would be my guess at the earliest. Hell just now pulling up hp, dell, and gateway's site, most of them come defaultly configed with 256mb ram. So that's what most people are going to get. Pretty much at this point the only thing pushing memory usage is brand new games and graphics editing programs... users generally have more problems with their choice of video card than they do with quantity of memory.
AMD needs to be able to produce a FX chipset that uses bog standard PC3200 ram in a dual channel format if they want the FX to be more popular.
Supposedly that's what Socket939 is for. My understanding is that it's supposed to get around this need for registered memory--supposedly enough to allow 4-layer PCB to be used for dual-channel motherboards. I'm betting it gets around the >2-DIMM performance problems as well, but I can't be sure.
<i>I can love my fellow man...but I'm damned if I'll love yours.</i>
they can get over this probleme but SIS via philosohie hit hard there.You dont take this measure unless you are sure to be the fastest any extra feature or restriction will lead to slower MCH 2 you need market segmentation like Optyeron 8xx series cost more that 2xx series even if they are the same core as you disable feature that are useful for server activate them at a price for server range.
they can get over this probleme but SIS via philosohie hit hard there.You dont take this measure unless you are sure to be the fastest any extra feature or restriction will lead to slower MCH 2 you need market segmentation like Opteron 8xx series cost more that 2xx series even if they are the same core as you disable feature that are useful for server activate them at a price for server range.
but I don't expect even 1g to become the standard for a couple years yet...
Nor do I expect the FX to be marketed as a "standard" Processor. That is what the A64 is for....get it?
Your just not going to see many (if any) HP, compaqs, emachine's, etc, etc use the Fx. That is what the A64 is for. This renders all these comments about "standard" rather pointless.
It's not what they tell you, its what they don't tell you!
I wasn't commenting on any of that kind of thing, i was simply commenting on memory ammounts... don't know where you got that from or what you are babbling about.
I wasn't commenting on any of that kind of thing, i was simply commenting on memory ammounts...
People have been commenting on what they think to be the standard amount of memory now and in the near future. Based on your comments to this argument I drew the conclusion that you were in agreement with the common misconception that the Fx's requirement of using registered memory was a bad thing. If I am wrong here forgive me. Following below is my rant to this misconception. Now let me get my crayons out and I will draw them a picture.
<font color=blue>The A64 fx processor is not a <font color=red> standard processor</font color=red> it is not intented for standard desktop machines. It is intended for high end machines. These same High end machines will most likely have larger than <font color=red> standard</font color=red> memory requirements. So, for the love of Pete, quite trying to tie a high end processor and what you consider standard memory amounts toghether.</font color=blue>
It's not what they tell you, its what they don't tell you!<P ID="edit"><FONT SIZE=-1><EM>Edited by ncogneto on 10/23/03 01:06 PM.</EM></FONT></P>
I'm betting it gets around the >2-DIMM performance problems as well, but I can't be sure.
How much you want to bet?
This realy is'nt a AMD vs Intel issue. It has more to do with signaling, trace length, signal integrity, and timing on the motherboard. As additional banks of memory are added to the motherboard they are located farther and farther away from the memory controler. Having this on the memory itselfs alievates this issue.
Registered/Buffered vs. Unregistered/Unbuffered Memory
<font color=green>Today’s motherboards can include one of two memory controller types : those that drive the memory control signals
via the controller and those that rely on the memory module themselves to drive the memory control signals.
Motherboards that employ the first controller, where the motherboard controller drives the memory control signals,
will utilize the non-registered/non-buffered memory modules.
Any motherboard that utilizes the second controller type, where the memory modules themselves drive the memory
control signals will utilize the buffered / registered modules (a register/buffer is actually an additional chip on each
memory module that acts to drive the memory control signal). One of the major advantages that this configuration has
is it’s ability to support large amounts of memory, because of the register/buffer chips on the modules. The major
disadvantage to these modules is a reduction in the performance of the memory module, as the act of using the
register/buffer requires one clock cycle of the module.
In the strict definitions of a buffer and register, a buffer is utilized on either EDO of Fast Page memory, while a register
is utilized with SDRAM memory modules. These lines have become blurred in the their usage to the point that they
basically mean the same thing.
Finally, registered/buffered memory may not be used in a machine that is also running unregistered/unbuffered memory.
This is because the slots in each type of module are created a little different, thus a registered/buffered module will not
physically fit the memory bank of a motherboard that only supports unregistered/unbuffered memory, and visa versa. \</font color=green>
It's not what they tell you, its what they don't tell you!
This realy is'nt a AMD vs Intel issue. It has more to do with signaling, trace length, signal integrity, and timing on the motherboard. As additional banks of memory are added to the motherboard they are located farther and farther away from the memory controler. Having this on the memory itselfs alievates this issue.
True, but one of the supposed benefits of Socket939 is the ability to safely use 4-layer PCB without an engineering miracle. Socket940 boards typically require at least 6-layer PCB.
Socket939 clearly does <i>something</i> WRT signaling stability, I just don't quite know what. Actually split the memory bus from 1x128bit to 2x64bit? Move the MCH bus pins to the periphery of the socket? I'm guessing all of the above.
That sort of rearrangement is probably a rather big deal when you don't have an off-die MCH sitting between the CPU and the memory.
That makes sense, however it re-inforces what I said earlier about it being a motherboard Issue, not a processor issue. While the transistion to the socket 939 allows for four layer PCB's, it still comes down to the motherboard and trace lenght to the Dimm slots. To make a long story short, better designed motherboards help alleviate the problem of added latency when using more than 2 sockets, but fundementally the problem still exists. Imagine having 8 dimm sockets? Ultimaely it will come to rest on the motherboard makers themselves as to how they implement the traces, and as the past has shown us they usually go the route of lower cost vs. better performance if they consider DIMM slots 3 and 4 to be of marginal importance.
It's not what they tell you, its what they don't tell you!
My reply was simply a comment on what BB wrote about how memory usage has grown and my belief that it doesn't grow quite as fast as he was stating.
Somehow from that you pulled an assumption out of yer ass that said I was talking about registered memory vs non-registered memory because someone else in the thread was when my post had NOTHING TO DO WITH THAT AND IT WAS NEVER MENTIONED IN MY POST.
Let me draw you a simple real picture with words (short ones so you dont get confused.) First, remove your head from your rectum and actually READ what people write instead of making assanine assumptions. Second, since you seem obsessed with getting a comment on the fx's use of registered memory-- I'll make one. Yes high end machines do generally have larger ammounts of memory, however, in this case it's a stopgap measure not the standard of what the fx line will be. Future fx chips will NOT require registered memory so this issue is highly illrelevent unless you absolutely have to have a fx right now and don't mind getting a socket that will go down the hole in about a year (since the socket remaining the same is a big amd selling point--upgradeability and all) and pay out the nose for memory that the next version of the chip probally won't require.
You are about to answer a thread that has been inactive for more than 6 months. If you still wish to proceed, please ensure that your posting is original and does not duplicate or overlap any prior responses to this thread.