Anyone planning any new hardware purchase...

[Joe DeFuria]

This whole conversation started because I chose a P4 over an Athlon, when asked why, the main reason was because plan on doing lots of "video editing / encoding", etc.

The majority of benchmarks show the P4 better than athlons at this task. Sometimes small, sometimes great.

The encoding benchmarks are much like game benchmarks. You can look at individual tests (and individual test files within those tests), but that only gives you a specific snapshot. In the end, for any given video encoding taks, the chances are it'll be better on a P4.

This is indicated by the overwhelming majority of tests from various sources show the P4 in a better light by varying degrees.

 

[incurable]

Why? Even considiring the move to Socket 939, the upgrade path is still better with Socket 754 than it is with Socket 478.

How so? IIRC, there two more speed grades for the Socket 478 coming (Prescott 3.4 and 3.6 GHz), but only a single one for Socket 754 ("3700+").

Mind you, I'm not saying that either got a good upgrade path, but I can't see how Socket 754 is superior in this respect.

cu

incurable

 

[beyondhelp]

just some ruminations and observations... At least as far as Tom's results go you only have to read as far as the memory Bandwidth tests to see why the P-IV is generally faster in encoding/streaming apps. An 800Mhz quad-pumped bus. Give the Athlon (any flavor) an equivalent Memory Bus speed and see what the results are. Yes, the Athlon does more work per clock than the P-IV but it can still only work on that Data at the speed the memory supplies said data to the CPU. Athlons architecture is very efficient compared to the P-IV, but the memory sub-system is quite inadequate.

Of course none of that takes into account the 64bit question. The P-IV and even Prescott is not going to do 64 bit. An A64 solution is.

RIGHT NOW, this instant the P-IV is faster in those apps generally, But if and when 64bit versions of those Encoding apps hit the shelves, It will probably be a different story. So It depends on How far down the road your looking.

For immediate gratification, the P-IV is probably the right choice for running those apps. But if you take a longer view and don't care to or can't afford to upgrade often, the A64/939MB may be a better choice in the long run. It's for sure you'll be looking at an Upgrade to 64bit sometime down the road if you opt for the P-IV/(Prescott) now/(soon). (along with a switch to Intel's Socket T MB's prob. which will have Dual 12v rails and require a new P.S. also)

For some really good insight into the future CPU/MB trends/ Pricing/Manufacturing issues, etc., browse through some of Ed Stroligo's Front page articles on the subject over at www.overclockers.com
Some interesting reading. Some of it reads more like editorial than article, but not biased.
He applauds/rants at both Manufacturers, as the shoe fits.

 

[incurable]

just some ruminations and observations... At least as far as Tom's results go you only have to read as far as the memory Bandwidth tests to see why the P-IV is generally faster in encoding/streaming apps. An 800Mhz quad-pumped bus. Give the Athlon (any flavor) an equivalent Memory Bus speed and see what the results are. Yes, the Athlon does more work per clock than the P-IV but it can still only work on that Data at the speed the memory supplies said data to the CPU. Athlons architecture is very efficient compared to the P-IV, but the memory sub-system is quite inadequate.

http://www.anandtech.com/cpu/showdoc.html?i=1946&p=5

Have a look at the last graph on this page, it shows that while having dual-channel RAM helps Hammer a bit (+7.5%), it only brings the top-of-the-line FX-51 on par with an extrapolated P4 2.6C.

Of course none of that takes into account the 64bit question. The P-IV and even Prescott is not going to do 64 bit. An A64 solution is.

RIGHT NOW, this instant the P-IV is faster in those apps generally, But if and when 64bit versions of those Encoding apps hit the shelves, It will probably be a different story. So It depends on How far down the road your looking.

How do you expect exposed 64-bitness to improve media encoding performance?

cu

incurable

 

[epicstruggle]

I didnt add this to my list, but im really considering having a water cooling system. Better performance, and less noise. Anyone else have it or considering it?

later,
epic

 

[Tim Murray]

I didnt add this to my list, but im really considering having a water cooling system. Better performance, and less noise. Anyone else have it or considering it?

later,
epic

I've been thinking about water cooling for a while, but it worries me. FOr one, next year, I'll have a roommate. I do not want my roommate to be able to accidentally destroy my computer, and water cooling sure seems like a good way to do that. Second, it's expensive. Third, is the cooling really that good compared to slapping a giant heatisnk and a fan on something, or is it more of a noise issue? And last.... if I get water cooling, what am I going to do when I get annoyed by that? Buy a Prometeia?

 

[John Reynolds]

You know, on the whole AMD vs. Intel issue all I can say--and this is only my own personal experience--is that in my years of system administration I have never seen a server built with a AMD CPU and chipset. In fact, I have never heard of it. Whether that's because of sound technical reasons or that Intel has some ethereal and yet ironclad lock on the server market, I won't say.

That said, I'm currently running a P4 3.06. My only real planned/expected upgrade for 2004 is either a R420 or NV40 and, knowing me, a refresh of one in the fall.

 

[Tim Murray]

John, that's because AMD had no server chips (except the Athlon MP, which frankly sucked) for years. Now with Opteron, that seems to be changing (since it's competing with the Itanic and the Xeon). Dunno, give it a few years. Server market seems hard to penetrate.

 

[Bouncing Zabaglione Bros.]

I didnt add this to my list, but im really considering having a water cooling system. Better performance, and less noise. Anyone else have it or considering it?

later,
epic

I've been considering one of these http://www.wetandchillychips.co.uk/ with a GPU cooler for a while, and they are not that expensive. However it would mean getting a new case if I wanted to fit the radiator the way I want it. With my next upgrade probably being a AMD64, maybe in BTX formfactor, a new case might be necessary, but the reported cooler temperatures on the XP64 might make watercooling unnessesary.

I'm not very interested in extreme overclocking, but the quietness would be of interest, maybe with some small overclocking.

 

[beyondhelp]

just some ruminations and observations... At least as far as Tom's results go you only have to read as far as the memory Bandwidth tests to see why the P-IV is generally faster in encoding/streaming apps. An 800Mhz quad-pumped bus. Give the Athlon (any flavor) an equivalent Memory Bus speed and see what the results are. Yes, the Athlon does more work per clock than the P-IV but it can still only work on that Data at the speed the memory supplies said data to the CPU. Athlons architecture is very efficient compared to the P-IV, but the memory sub-system is quite inadequate.

http://www.anandtech.com/cpu/showdoc.html?i=1946&p=5

Have a look at the last graph on this page, it shows that while having dual-channel RAM helps Hammer a bit (+7.5%), it only brings the top-of-the-line FX-51 on par with an extrapolated P4 2.6C.

Then I say consider the source. Look at the last chart on this page...
http://www.amdmb.com/article-display.php?ArticleID=260&PageID=11
The Tables are turned...

Of course none of that takes into account the 64bit question. The P-IV and even Prescott is not going to do 64 bit. An A64 solution is.

RIGHT NOW, this instant the P-IV is faster in those apps generally, But if and when 64bit versions of those Encoding apps hit the shelves, It will probably be a different story. So It depends on How far down the road your looking.

How do you expect exposed 64-bitness to improve media encoding performance?

cu

incurable

I'm not a Software Coder, so I can't say how they will benefit specifically. I suppose being able to work on larger Data sets or something like that. The larger Memory address space wouldn't really help unless your already hitting the 32bit limit. 4GB?
However it's accomplished, Most people have been led to believe that 64 "bitness" will speed up applications. To What degree is the real question.
imho, I believe a 64 bit version of say, Xmpeg would run faster on FX than the 32bit version on P-IV. No way to prove it yet. Only time will tell I guess.

l8r

 

[John Reynolds]

John, that's because AMD had no server chips (except the Athlon MP, which frankly sucked) for years. Now with Opteron, that seems to be changing (since it's competing with the Itanic and the Xeon). Dunno, give it a few years. Server market seems hard to penetrate.

Especially when you bring such a scenario up to your MIS staff and their eyebrows go up about an inch.

Right or wrong, that attitude is very prevalent.

 

[Mariner]

It's the old saying, isn't it: "Nobody ever got sacked for buying Intel".

Until companies such as Dell start selling AMD chips in both desktops and servers, large corporations are always going to be wary. Of course, Intel realises this and I'm sure they give Dell some fantastic discounts to stay "Intel Inside". 8)

I've a friend who sells servers and I recently asked him about the Opterons and what he knew about them. He didn't really know anything about the chip at all but just assumed that Itaniums and Xeons are better just because this is what everyone he deals with asks for!

 

[incurable]

http://www.anandtech.com/cpu/showdoc.html?i=1946&p=5

Have a look at the last graph on this page, it shows that while having dual-channel RAM helps Hammer a bit (+7.5%), it only brings the top-of-the-line FX-51 on par with an extrapolated P4 2.6C.

Then I say consider the source. Look at the last chart on this page...
http://www.amdmb.com/article-display.php?ArticleID=260&PageID=11
The Tables are turned...

Sorry, but your link doesn't have any data to support your theory that a lack of memory bandwidth is holding the A64 back in XMPEG/DivX encoding. (and is very unspecific as for what they actually testing, too)

I'm not a Software Coder, so I can't say how they will benefit specifically. I suppose being able to work on larger Data sets or something like that. The larger Memory address space wouldn't really help unless your already hitting the 32bit limit. 4GB?
However it's accomplished, Most people have been led to believe that 64 "bitness" will speed up applications. To What degree is the real question.
imho, I believe a 64 bit version of say, Xmpeg would run faster on FX than the 32bit version on P-IV. No way to prove it yet. Only time will tell I guess.

The part in italics sounds rather strange to me, and I'm anxious to see the reaction of today's most vivid 64-bit-on-the-desktop supporters when the first real-world benchmarks come in. (whenever that may be, Mircosoft doesn't seem to be in a hurry)

As far as XMPEG/DivX is concerned, IIRC fast/good quality encoders use the FPU/SSE(2) units, 64-bitness of the integer core wont buy you a thing there. Of course, AMD64 doubles the SSE2 registers in long mode, so that might speed things up, but frankly I'm doubtful that it'll gain AMD enough to go from 2.6C-eq to 3.2-eq in Anand's tests.

cu

incurable

 

[beyondhelp]

http://www.anandtech.com/cpu/showdoc.html?i=1946&p=5

Have a look at the last graph on this page, it shows that while having dual-channel RAM helps Hammer a bit (+7.5%), it only brings the top-of-the-line FX-51 on par with an extrapolated P4 2.6C.

Then I say consider the source. Look at the last chart on this page...
http://www.amdmb.com/article-display.php?ArticleID=260&PageID=11
The Tables are turned...

Sorry, but your link doesn't have any data to support your theory that a lack of memory bandwidth is holding the A64 back in XMPEG/DivX encoding. (and is very unspecific as for what they actually testing, too)

I didn't see any Data to support your counter argument in Anands article either, beyond the usual Lip service to the P-IV being better at Media encoding. Anand doesn't give any other reason. The FX seems to lag about the same percentage in Media encoding apps as It lags in memory Tests like Sandra. FX supports SSE2 so even if the P-IV implementation is better, FX shouldn't be giving up the 17% or so diff. in Anands Encoding results just because of better SSE2 support for P-IV. Computationally, FX and P-IV are close(Raw Clock speed notwithstanding as FX's higher IPC offsets that advantage.) and don't really account for that big a difference either, which leaves memory Bandwidth.

Digging up detailed Data on the FX/A64's imbedded memory controller is not easy.
(This next bit is an assumption, as I have not been able to nail down the actual info on How the memory Controller talks to the CPU. Several Review articles have referenced the Hypertransport channel as the mechanism used, But that may not be the case, as the Block diagrams do not describe the path or clock speeds from Memory Controller to CPU (internally). They show the memory going to the CPU(ie: memory Controller), but thats it. Nothing on how the Data actually gets from the Memory Controller to CPU proper.)
...So assuming the A64 and AFX both use thier single channel Hypertransport channel for memory transfers to the CPU working units I'd say that the A64 and FX still don't have the raw Bandwidth of the P-IV 800Mhz system bus as per my understanding after reading through the AMD Tech Docs of the FX and A64. The P-IV and FX both have Dual Memory Controllers feeding Data to the memory Controller. One(P-IV) is on the Motherboard and one (FX) is on die. They both support 6.4GB/s Bandwidth to the Controller. The P-IV's 800Mhz X 64bit wide interface can then send data from the Controller to the CPU @ the same 6.4GB/s. (If)The Athlon FX uses the Hypertransport link, which has an 800Mhz X 16bit pipe and Provides 3.2GB/s each way simultaneously, Then that doesn't seem to be the equal of the P-IV's system Bus to me.

So my observation that given the same Quad pumped Bus an XP or A64 would keep up with the P-IV hasn't been met by FX w/ Dual Memory Controllers yet because it's not the Dual memory Controllers that count really, It's the Transport mechanism used after the data reaches the memory controller that is the bottleneck for the FX. If FX does use Hypertransport to move Data from mem. Controller to CPU, It does not equal the throughput(one way) of the P-IV.

Feel Free to disabuse me of my Notions. I try to learn something new each day.

I'm not a Software Coder, so I can't say how they will benefit specifically. I suppose being able to work on larger Data sets or something like that. The larger Memory address space wouldn't really help unless your already hitting the 32bit limit. 4GB?
However it's accomplished, Most people have been led to believe that 64 "bitness" will speed up applications. To What degree is the real question.
imho, I believe a 64 bit version of say, Xmpeg would run faster on FX than the 32bit version on P-IV. No way to prove it yet. Only time will tell I guess.

The part in italics sounds rather strange to me, and I'm anxious to see the reaction of today's most vivid 64-bit-on-the-desktop supporters when the first real-world benchmarks come in. (whenever that may be, Mircosoft doesn't seem to be in a hurry)

As far as XMPEG/DivX is concerned, IIRC fast/good quality encoders use the FPU/SSE(2) units, 64-bitness of the integer core wont buy you a thing there. Of course, AMD64 doubles the SSE2 registers in long mode, so that might speed things up, but frankly I'm doubtful that it'll gain AMD enough to go from 2.6C-eq to 3.2-eq in Anand's tests.

cu

incurable

Why does that comment sound strange? I am willing to believe that, and I do stipulate that it is a matter of How much will it speed it up. I'm sure there will be some apps that just do not respond to being 64bit by being faster. I believe that comes down to the Talent writing the code to some extent and, of course, the nature of the app. As I said Time will tell, and I too am looking forward to those First Benchmarks to see what's what.

Just so you know, I am not an "avid" 64Bit on the Desktop supporter. (not vivid either, just the standard flesh tone thankyou! ) I am quite happy with my XP as is. I prob. won't go 64bit until Games are written to 64 bit and then only if it actually gains me something substantial, or if its becomes a matter of switching to 64bit or not being able to do anything at all. That will prob. be a while. Hopefully anyway, as I swore not to spend any more on my 'puter for the next 18 mos. (wish me Luck! )

Happy Computing!

 

[incurable]

I didn't see any Data to support your counter argument in Anands article either, beyond the usual Lip service to the P-IV being better at Media encoding.

The article I linked to shows that doubling the memory bandwidth available to the core (i.e. going from A64 2.2 GHz to AFX 2.2 GHz) does help the XMPEG/DivX performance somewhat (+7.5%) but nowhere near the amount it would (> +50%) were memory bandwidth really the limiting factor.

Anand doesn't give any other reason. The FX seems to lag about the same percentage in Media encoding apps as It lags in memory Tests like Sandra.

I can't remember ever seeing the FX-51 lacking in memory bandwidth in Sandra. Reference:
http://www.tech-report.com/reviews/2004q1/athlon64-3400/index.x?pg=3

FX supports SSE2 so even if the P-IV implementation is better, FX shouldn't be giving up the 17% or so diff. in Anands Encoding results just because of better SSE2 support for P-IV. Computationally, FX and P-IV are close(Raw Clock speed notwithstanding as FX's higher IPC offsets that advantage.) and don't really account for that big a difference either, which leaves memory Bandwidth.

Thing is, you can't really compare the Hammer and Netburst cores and say that either's advantages are nullified by the other's advantages under other circumstances, therefore there shouldn't be any performance differences. If an application relies heavily on a single part of the core, and that part is faster (through whichever means) in one companies product than this product is bound to offer higher performance for said app.

Digging up detailed Data on the FX/A64's imbedded memory controller is not easy.
(This next bit is an assumption, as I have not been able to nail down the actual info on How the memory Controller talks to the CPU. Several Review articles have referenced the Hypertransport channel as the mechanism used, But that may not be the case, as the Block diagrams do not describe the path or clock speeds from Memory Controller to CPU (internally). They show the memory going to the CPU(ie: memory Controller), but thats it. Nothing on how the Data actually gets from the Memory Controller to CPU proper.)

That's not exactly how it works. Internally, there's a switching fabric (often referred to as X-BAR) that is connected to the core, the 3 HT controllers and the memory controller. I don't think the exact configuration of this fabrics has been disclosed, but it's surely designed to accomodate the throughput of 3 16-bit HT controllers and a 128-bit memory interface simutaniously. In other words, it's not as throughput limiting as you assumed.

Why does that comment sound strange?

It sounds strange to me cause the people you're referring to weren't led to believe in the supposed advantages of 64-bit computing on the desktop through benchmarks that show such but rather by a simple assumption marketing people like to propagate: Bigger numbers = better!

I am willing to believe that, and I do stipulate that it is a matter of How much will it speed it up. I'm sure there will be some apps that just do not respond to being 64bit by being faster. I believe that comes down to the Talent writing the code to some extent and, of course, the nature of the app. As I said Time will tell, and I too am looking forward to those First Benchmarks to see what's what.

As I said before, I'm looking forward to them, too. Until such benchmarks appear, however, I won't assume that 64-bitness is gonna buy AMD anything. (which IMHO is the sane thing to do, as there's very little factual support for hopes)

Just so you know, I am not an "avid" 64Bit on the Desktop supporter. (not vivid either, just the standard flesh tone thankyou! ) I am quite happy with my XP as is. I prob. won't go 64bit until Games are written to 64 bit and then only if it actually gains me something substantial, or if its becomes a matter of switching to 64bit or not being able to do anything at all. That will prob. be a while. Hopefully anyway, as I swore not to spend any more on my 'puter for the next 18 mos. (wish me Luck! )

Happy Computing!

Good Luck.

cu

incurable

 

[Sxotty]

You know, on the whole AMD vs. Intel issue all I can say--and this is only my own personal experience--is that in my years of system administration I have never seen a server built with a AMD CPU and chipset.

There are servers with AMD opterons in them, so now you have at least heard of it .

edit:
As for me, I will wait until ... as long as possible (to hold off the bug [August]), and then buy the fastest chip for under $300 ( I will get AMD, if only to support a 2nd party on ethical reasons so that we have better competition like the current ATI/NV scenario), and MB with PCIexpress, and a new video card, also I want to get SATA raptor, and...so many things. Someday I especially want an LCD that doesn't ghost horribly, but until I see it myself i will stick with my 21" trinitron.

 

[epicstruggle]

I will get AMD, if only to support a 2nd party on ethical reasons so that we have better competition like the current ATI/NV scenario

This has been the tie breaker between many of my computer purchases. Id rather buy amd just so intel has someone to compete with. Plus I just sold all my amd stock on friday. So i have to give them some of the profits.(30 some percent in about 1 month)

later,
epic

 

[Althornin]

I didnt add this to my list, but im really considering having a water cooling system. Better performance, and less noise. Anyone else have it or considering it?

later,
epic

I've been considering one of these http://www.wetandchillychips.co.uk/ with a GPU cooler for a while, and they are not that expensive. However it would mean getting a new case if I wanted to fit the radiator the way I want it. With my next upgrade probably being a AMD64, maybe in BTX formfactor, a new case might be necessary, but the reported cooler temperatures on the XP64 might make watercooling unnessesary.

I'm not very interested in extreme overclocking, but the quietness would be of interest, maybe with some small overclocking.

Gods no, dont buy some crappy kit.
I watercool, and it doesnt have to be that expensive.
Its also been very reliable, cools better than aircooling, and is quieter.
My cooling solution has been running for over a year continuously.

Buy yourself a ViaAqua 1300 pump (~$25), some tubing (less than $10), and dont use one of those crappy multi-pass style radiators. Just go tot he junkyard and pick up a heatercore for cheap - ~$30 or less.
Then all you need is a waterblock. If all you are looking for is decent performance, than just about any waterblock from say, danger den, will do fine.

If you want ultimate performance,t han of course you need to buy a bigger heatcore, a better pump (say, Iwaki or the swiftech MCP600) and get yourself a Cascade waterblock - currently the best.
Anyways, you should put yourself in the know before doing this. Any questions, just ask - I'll answer or direct you to someone who knows the answer

 

[Bouncing Zabaglione Bros.]

Gods no, dont buy some crappy kit.
I watercool, and it doesnt have to be that expensive.
Its also been very reliable, cools better than aircooling, and is quieter.
My cooling solution has been running for over a year continuously.

Buy yourself a ViaAqua 1300 pump (~$25), some tubing (less than $10), and dont use one of those crappy multi-pass style radiators. Just go tot he junkyard and pick up a heatercore for cheap - ~$30 or less.
Then all you need is a waterblock. If all you are looking for is decent performance, than just about any waterblock from say, danger den, will do fine.

Well these "crappy kits" have been getting very good reviews, they are not that expensive, and don't involve trolling around junkyards

 

[incurable]

Well these "crappy kits" have been getting very good reviews, they are not that expensive, and don't involve trolling around junkyards

Well, I think water cooling kits are a lot like OEM computers, they're good if you're just starting with it or want a convenient solution that 'just works'. (as far as that can be said about water cooling systems, that is) However, as with computers, as you learn more, you can grow a desire to build a system that fits your needs better than a kit could ever do.

Personally I thought about watercooling for about a year (or more?) before making my choices but the time was needed to get up to speed and learn from other's mistakes. It enabled me to go out and buy the right stuff to reach my goals, and I can't say I've been disappointed with my setup. (had it running for a year (-downtime for PC upgrades), and it's still going strong)

cu

incurable