athlon 64: why is it's poor multitasking ignored/downplayed?

[AnnoyedGrunt]

OK, I am not exactly sure where the problem lies, but here are my thoughts:

First, a summary of the hardware side that has been discussed a bit earlier in the thread:
I can definitely understand how the HT on the P4 would help it's performance in this situation. If the P4 normally has a hard time keeping its stages full due to various factors, then it is logical that adding additional tasks would help keep the pipes full and would not significantly affect performance up to a certain point. So, it makes sense that the P4 might not see much of a performance hit when running two DAOC clients because the extra client is using stages that were going to waste when only one client was running.

Now, for the A64, since it runs closer to peak efficiency with a single client, it makes sense that adding additional tasks would start affecting that peak performance (whether it's noticeable or important to you would depend on many factors like the actual CPU requirements of the respective programs as well as your usage patterns). So, if you were running one program that was operating near peak efficiency, then adding a second program would start slowing the first even if the second was not very resource intensive.

So, I think the big thing is to try and find out more precisely how resource intensive DAOC is.

To do this, I suggest the following:
Underclock the A64 to approximately 1/2 the standard speed, and try running a single client. If the single client on a 1/2 speed computer runs simlarly to two clients on a full speed computer, then the multitasking problem would seem to be due to the A64's hardware design. However, if the single client on the 1/2 speed computer still runs smoothly, then it would seem there are some potential software issues @ hand and perhaps some ways to fix the problem by modifying settings, updating the game code, or maybe by an OS update.

So, Cainam (or anyone else) perhaps you could try out a test like this, and see what the results are.

Good luck!

-D'oh!

 

[Acanthus]

Originally posted by: Vee

Originally posted by: CaiNaM
OK, i finally found an online review that helps explain/support what i'm trying to say here.

running word and web browsing while copying files is not the what i mean by "multitasking". i'm talking about several apps that all try go hog large amounts of cpu resources. finally ran across a reveiw that actually stresses a system with multiple tasks: this one compressing a 330MB AVI file to an 84MB WMV file while FS2004 was running its test.

while the a64 3400+ easily held the lead running FS2004 solo (60-90 fps vs 50-70fps), it also stood out in last place while running FS2004 and encoding the media file; the 3400+ averaged less than 4fps while the 3.2ghz averaged 17fps. that's a huge difference, and shows exactly the point i've been trying to make here. you can see the graph here.

"When WME9 was running, the Athlon 64 averaged less than 4 frames per second. We did see one large spike in frame rate, but the curve pretty much remained under 4 fps for the majority of the run. All three Pentium 4 processors performed more poorly when running Flight Sim 2004 solo, but managed to average around 17 frames per second while WME9 was chugging along in the background. The other interesting data point is that Prescott's average frame rate of 17.2 fps when multitasking was essentially the same as the 3.2GHz P4EE's 17.5 fps. Of course, the frame rate dipped on occasion, but the point here is that Hyper-Threading clearly has a major impact."

i'm not sure why results like these aren't given much consideration in reviews comparing the two architectues; hell, it's almost like it's taboo to talk about these things, but honeslty i'm not play fanboy here; this is a legitimate concern (for some, such as myself more than others who don't normally run multiple cpu intensive apps/games) raised after i purchased an athlon64, and i'm hardly talking out my ass...

If you're referring to the 'benchmarking' done by Extremetech, it doesn't show what it seem to show. In many ways this is a fake or lying test. IIRC, the testers, originally when they made the first, and *developed* these "benchmarks", confessed that they had to set Windows to give priority to the background task, in order for the test to show what they wanted it to show.
(It also seems Extremetech have now stopped these "tests".)

I've tried to explain this to you several times in this thread, (but if you now detect a certain element of impatience, it might have more to do with your change of title for this thread):
- Like all CPUs, other than Intel's P4HT, Athlon64 only handles one thread. So Windows sheduler only sends it one thread. And if Windows sheduler is then told to give priority to WME9, it results in 4 Fps. What you see is not Athlon64's multitasking properties, but Windows', and a manipulated Windows at that.
And yes, the P4's manage 17 Fps, because Windows sends it a second thread, so 'turning off' FS2004 will not be as successful, as it is in the case of the A64.

This so called "test" only shows something that is already perfectly clear, from the CPU's technical specs: The HT-P4s have two "virtual" cpus, the A64 is just one as usual. - It shows nothing else!
It certainly doesn't show "multitasking properties". Good multitasking properties comes from a large and well working cache, and lots of RAM.
A true multitasking benchmark is for instance 'Veritest Business Winstone Multitasking'. This tests multitasking performance, when Windows is allowed to run threads as Microsoft intended.

Is it reasonable for anyone to use FS2004 at 17 Fps? While giving priority to encoding in the background?Of course not! The proper way to use any preemptive multitasking OS, is to give the foreground or interactive task priority. If you work that way, you will not notice effects from a background task running much, or at all. Similarily, two concurrent workloads running simultaneously at the same priority, will both run at half speed, and so on.

In order to screw this up, priorities need to be screwed up. That happens, so yes, having two CPUs, virtual or real, is sometimes advantageous. I've never denied that. But still, apart from that you can get higher collective performance from the setup, (just like on dual CPUs or dualcore CPUs) it's not a big deal, outside some special situations. (In fact, Intel's HT is primarily motivated by getting more work out of the execution unit. Not improving multitasking.)
And on a well working OS, you would have a very hard time seeing the difference between one or two threads released by the sheduler.

HT will only be advantageous when one undesired thread is hindering a desired thread from running. That can be avoided for normal multitasking uses. Besides, most other OSes don't seem to have the same problems at this, as Windows messagequeue driven program model.

The A64 does not have "poor multitasking", that's an absolute rubbish concept. If you experience "poor multitasking" it's either intentionally contrieved (as in extremetech's FS2004/WME9 benchmark) or an example of Windows' poor multitasking (that can probably be improved by manually tinkering 'basic process priority').

I've used P4s at daily professional work, for what seem to be ages. Lately P4C too. And yes, HT is nice but it still doesn't compare.
I'm totally in love with my A64 3400+. After sogging knee and thigh deep through swamps, for years with laggy Pentium 4s, that were management's favorites, but poorly suited for the technical work, A64 feels like a ride in a Bentley. It performs an incredible 91% (I've done my own application benchmarking) better than a 3.2 P4C. It's an even easier A64 choice, than for Flight Simulator 2004. And even while running computing in the background (at low priority) it's more snappy and comfortable to work with. (I'm not running "cool'n quiet", no point, but I don't think that will have any impact.)

As for your needs or "legitimate concern", why didn't you consider a dual CPU setup? (I/we certainly did.)
What are you complaining about? Did you somehow thought that when you bought the A64, it was a dual core CPU? I should think not. Did you not know, when you bought the A64, that Intel's hyperthreading feature meant that it acted as two virtual CPUs? If really not, what did you then imagine it was?

Time for a credibility attack.

Youre implying that the A64/P4 systems are running the same memory, video cards, and comparative age chipsets. They cant be, i have NEVER seen an A64 approach 100% over a comparative P4C in any benchmark, even in 64 bit.

 

[BDSM]

OK.. I haven't read the entire thread. Skimmed through maybe 50%.

But to try to answer the original question: It has been "overlooked" because the majority of A64 users don't encode video or have seti clients running while they play CS or UT!

In fact I know NONE who do that out of maybe 10-12 nerds I call friends. They don't do it because they just don't need to encode video or do folding or seti etc at all.

I bet that 99% of computer users (home and office) don't multitask in a way that proves HT the king some claim it to be.

Simple as that.

I should point out that I own and use a P4C 3.2@3.6Ghz and it runs great on a raptor 74g.

I've had both athlon and intel systems before this. I can't say either had any huge, real performance advantage in any way over the others.

The number one and only reason I chose Intel this time was 1. Compatibility and 2. stability.

No fuss with drivers. everything just works thank you.

I'm sure others had different experiences. This is mine.

Ciao

 

[Brian23]

I'd like to hear what Anand has to say about all this...

 

[glugglug]

Originally posted by: BDSM
OK.. I haven't read the entire thread. Skimmed through maybe 50%.

But to try to answer the original question: It has been "overlooked" because the majority of A64 users don't encode video or have seti clients running while they play CS or UT!

Distributed computing stuff like SETI clients all run at low priority and don't noticably effect games on A64. (unless you are swapping from too little RAM). Just the SETI client won't do much while the game is running.

However, with hyperthreading on on a P4, they will slow the games down substantially, because the distributed task ends up with a whole virtual CPU to itself, which ends up stealing time on various CPU execution units from the game.

 

[Lithan]

BDSM. I agree with you on compatibility (...with ram, although I hear 939 goes a long way to fixing how well you can overclock with a gig+ of ram). But it's been known for a long long time that there is no stability problem with any current cpu's. AMD or Intel.

And actually. Intel gave me a lot more problems with drivers than AMD did. With AMD I can get chipset drivers for the nf2/nf3 from either the mobo manufacturer or Nvidia themselves. With intel I had to download the drivers for the Intel865PERL (A board which I do not own) to get chipset drivers for my p4p800. Asus (for whatever reason) doesn't have a driver download for my board, and Intel doesn't have a driver download for the 865 chipset (at least not marked as such). To get 865 drivers I had to download the drivers specifically for Intels 865 board and install them (No workaround was needed, they installed fine).

 

[Accord99]

Originally posted by: Lithan
1. HT can degrade system performance. I seem to remember one of the pi calculaters takes (or took at one time) a significant hit when HT was enabled

Two instances of SuperPI runs at 145% of the throughput of one instance, two instances of PiFast runs at 130% of one instance.

2. Hypertransport > Pci bus

Shaving even 1000 ns on the bus is not going to matter when a disk acess is in the order of tens of milliseconds and network travels times are in the hundreds of milliseconds.

3. Note the words "most useful". What you said, while technically accurate, fails to tell the full story.

Note midnightwriter is a noted Intel hater on the Aces forum who has shown very little technical knowledge. And what I said is essentially the full story, run two CPU intensive apps and it will in the vast majority of cases increase throughput by 15-30%. You don't have to worry about foreground, background, priority or I/O.

 

[CaiNaM]

strange.. i never had to download anything for my 865 chipset... only for the built in intel gb ethernet.

regardless, since i started using nf chipsets with athlon xp and later, i havent had any compatibity/driver issues or driver hassles with my amd platform pc's either (previous via chipsets gave me headaches, however so i stuck with boards using amd's chipsets on early athlons).

imo, there is no stability advantage for either platform. they're both solid in my experience.

if i favor either in this area, it's actually the amd w/nf3, as it allows me to run both sata raid and ide raid simultanously, although (at least with the dfi ut nf3 250gb) serial 1 & 2 are useless to me as the ext phy limits overclocking, therefore i only have 2 sata devices avail for my use, even tho the board provides 4 sata connections.

my chaintech vnf3 250 doesn't have this issue, however it only has 2 sata devices avail. anyways...

 

[Lithan]

It might be included in Xp Cainam, remember I run Win2k. In win2k, I was recieving yellow spikes on cod:uo lagometer (distinctions between the server's response and the calculated response.) I was told this was likely an agp issue and installing chipset drivers would fix it. They did.


And Accord, I was referring to one instance; Hypertransport has shown the performance boost it gives over pci, and even without a boost, it would still amount to how much HT helps (which was the statement he made); I don't frequent Aces forums, but his reputation is irregardless. He was right. If what you say is true, all the benchmarks used to prove HT's case wouldn't follow the same pattern.

Via is garbage. Nvidia is Amd's Intel.

 

[CaiNaM]

Originally posted by: Lithan
It might be included in Xp Cainam, remember I run Win2k.

ahh.. that might explain it..

regarding superPI, i'm a bit confused on on how that's a "negative" for intel. as a matter of fact, in this review, the performance of the a64 3200+ and p4e 3.2ghz are identical in the "single task" superPI benchmark.

 

[Lithan]

"Ext phy limits"?

Oh yeah... and just incidentally, you have a really poor network if local access/response times are in the hundreds of milliseconds. All my buddies LAN's (on cheap cabling and linksys routers) have response of ~ 3-7ms max.

Superpi is a big intel-leaning benchmark. I'm refering to testing that found intel minus HT outperformed Intel + HT when running a single instance. It's only relavance is defending the statement that HT can cause a performance hit in instances where only one heavy-load process is running.

 

[CaiNaM]

Originally posted by: Lithan
"Ext phy limits"?

yes. this is a post on extremesystems forum by Oscar Wu, the man behind the great abit overclocking boards, who was stolen by DFI and helped design the LanParty UT NF3 250GB:

Currently there are two overclocking issue in NF3 250/250GB/Ultra Chipset ...

1 . External PHY SATA is useless once HTT clock > 240 MHz

Solution : No solution yet ...

2 . HTT clock PLL register is powered by stanby voltage , which means after you power down the system , the HTT clock register will remain the same before power down ... If you use low cpu ratio and high FSB combination , the board may fail to boot in next power up ... Ex : 3400+ cpu OC to 300 * 8 , in the next power up , CPU frequency will be 300 * 11 = 3300 mhz immediately

Solution :

1 . Always clear stanby power good signal to make HTT clock back to 200 mhz in every power up , but some stanby power good keeped funciton can not be used

2 . Always turn off stanby power after you powered down the system

3 . Use Mobile CPU instead of Desktop version

4 . Use Spread Spectrum Register to program HTT clock , which is not stanby powered keeped ... but it will cause other problem ..."

Oh yeah... and just incidentally, you have a really poor network if local access/response times are in the hundreds of milliseconds. All my buddies LAN's (on cheap cabling and linksys routers) have response of ~ 3-7ms max.

regarding lan latency, i've never mentioned, complained, or even stated i have a lan, so your point is lost on me. posting in too many threads, perhaps?

Superpi is a big intel-leaning benchmark. I'm refering to testing that found intel minus HT outperformed Intel + HT when running a single instance. It's only relavance is defending the statement that HT can cause a performance hit in instances where only one heavy-load process is running.

hmm.. okay, but even giving you the benefit of the doubt here, performance is on par with the a64, so i don't see its significance.

at any rate, for many tasks, the a64 is just flat faster, plain and simple. the on-die memory controller, the higher efficiency of the shorter pipelines compared to intel, etc. gives the amd a slight performance edge in single tasks, despite its much lower clock rate (and this gives it a power consumption and thermal output advantage as well), but that's not really the point of this thread. we all know the virtues of the a64 architecture, which is talked/written about everywhere; my concern was the relatevely ignored weakness i've encountered with this setup.

 

[Lithan]

Nah I was just replying to the guy above you. Accord said lans operate in hundreds of millisecs.

 

[Accord99]

Originally posted by: Lithan
And Accord, I was referring to one instance; Hypertransport has shown the performance boost it gives over pci,

HyperTransport doesn't replace PCI. It only connects PCI devices to the southbridge. On a typical desktop, it doesn't make the HD or network card any more responsive.

and even without a boost, it would still amount to how much HT helps (which was the statement he made); I don't frequent intel forums, but his reputation is irregardless. He was right. If what you say is true, all the benchmarks used to prove HT's case wouldn't follow the same pattern.

Just because benchmark sites lack imagination doesn't mean he's right. What I say is true because it is my experience. HT doesn't need a foreground operation that frequently stalls the pipeline, foreground and background operations are irrelevant and even two heavy I/O apps will see benefits.

Oh yeah... and just incidentally, you have a really poor network if local access/response times are in the hundreds of milliseconds. All my buddies LAN's (on cheap cabling and linksys routers) have response of ~ 3-7ms max

I was refering to Internet packets, but even with local networks, 3ms is still 3 million nanoseconds.

 

[DrMrLordX]

Monster thread in the making. I'd like to hear Anand weigh in on this one as well, if he has the time. He may not.

I have to agree a little with Lithan here, in that slagging the Athlon 64 as a "poor multitasker" on account of DAoC results is, perhaps, a bit narrow-minded. I have played DAoC before, and I never found myself wanting two clients running, much less on the same machine. Most people don't play DAoC(if you don't believe me, check out their subscriber base, and then compare it to that of Lineage, Lineage 2, and FFXI). Among those that DO play DAoC, not all want multiple accounts(which would be necessary, as I recall, to make multiple instances of the client worth using).

Also, gururu said . . .

i think it is an interesting observation considering that pre-Athlon64 cores have 'less' of an issue running two resource-intensive tasks.

I don't think this has been established. Has anyone here "proven" that the Barton or Thoroughbred-B cores handle two instances of DAoC better than Clawhammer or Newcastle?

On the contrary, Lady Vee ran some CPU-intensive benchmarks of her own and . . .

Ok, I've done some experiments with multiple instances of an old heavy computing app, that allows you to see the progress in realtime.
I also think Glugglug has hit the nail and come up with the source for the problem. It's quite visable how long quanta WindowsXP are providing. Windows98 is something completely different. So we now have a possible explanation for why it worked with Axp and P4B, but not A64. If you (CaiNam) were running W98 in that case? (I ran it on a HT-disabled P4 too. No contest, so I don't think anything is wrong with A64 performance under multitasking, but the timeslices seem as long for the P4 too, even if not much is done in them.)

Well there you have it, at least from Lady Vee. I have yet to see anything in this thread, or elsewhere, that shows the Athlon 64 to be a weak multitasker as compared to a Barton or a non-HT P4 in a WinXP environment(or any environment, for that matter).

As an aside, could we drop the attempts to assess the value, or lack thereof, of HT? I hope everyone here realizes that nobody will ever come to any real agreement about the real-world value of HT. All that can be said about it with any degree of certainty is that

1). it exists
2). it's easy for Intel to put it on their long-piped P4s without increasing transistor counts or CPU costs considerably
3). in most cases, it doesn't hurt performance
4). it can be turned off

It's a perk of owning a P4, and nothing to write home about, really. Any benefit it provides is nothing compared to what you'll get from a multi-CPU setup. It doesn't really hurt anyone to use it, in most cases, and if it does, it can be deactivated. Few, if any, people seem to be able to agree on whether or not it is really useful, and that includes those who have actually used P4s and have multitasked extensively while doing so.

As far as speculation regarding dual-core CPUs from AMD is concerned, I would recommend checking this little snippet here

http://www.tomshardware.com/ha...s/20041005_205914.html

It's short on information, but AMD seems to think they can get a net performance boost out of their dual-core Opterons, even with clock speeds 600-1000 mhz lower than those found on their single-core Opteron predecessors. It should be interesting to see if they can back up those claims.

 

[CaiNaM]

Originally posted by: DrMrLordX
Monster thread in the making. I'd like to hear Anand weigh in on this one as well, if he has the time. He may not.

as we all would (from any review site, for that matter), i'm sure.. including myslef.

I have to agree a little with Lithan here, in that slagging the Athlon 64 as a "poor multitasker" on account of DAoC results is, perhaps, a bit narrow-minded.

again, while that's certainly important to me, that's was only what started me on this subject, not the only case in which this can be noticed.

I have played DAoC before, and I never found myself wanting two clients running, much less on the same machine. Most people don't play DAoC(if you don't believe me, check out their subscriber base, and then compare it to that of Lineage, Lineage 2, and FFXI). Among those that DO play DAoC, not all want multiple accounts(which would be necessary, as I recall, to make multiple instances of the client worth using).

well, there's 2-2.5k on at any given time on the server i play (3k right now, and 33,500 worldwide as i write this), and that's only the segment of the subscriber base as they obivoulsy don't all play at the same time). but again, this behavior spans more than just daoc.

I don't think this has been established. Has anyone here "proven" that the Barton or Thoroughbred-B cores handle two instances of DAoC better than Clawhammer or Newcastle?

yes, while nowhere near as smooth as my p4c, both my barton and p2.0b can run a second client well enough that you can "stick" the chars and they will follow (it is laggy, and framerate is inconsistent in the second client). the a64 cannot do this, and the background client goes "linkdead" after a few minutes of activity.

On the contrary, Lady Vee ran some CPU-intensive benchmarks of her own and . . .

Ok, I've done some experiments with multiple instances of an old heavy computing app, that allows you to see the progress in realtime.
I also think Glugglug has hit the nail and come up with the source for the problem. It's quite visable how long quanta WindowsXP are providing. Windows98 is something completely different. So we now have a possible explanation for why it worked with Axp and P4B, but not A64. If you (CaiNam) were running W98 in that case? (I ran it on a HT-disabled P4 too. No contest, so I don't think anything is wrong with A64 performance under multitasking, but the timeslices seem as long for the P4 too, even if not much is done in them.)

Well there you have it, at least from Lady Vee. I have yet to see anything in this thread, or elsewhere, that shows the Athlon 64 to be a weak multitasker as compared to a Barton or a non-HT P4 in a WinXP environment(or any environment, for that matter).

there are multiple examples of this behavior on other applications other than daoc

As an aside, could we drop the attempts to assess the value, or lack thereof, of HT? I hope everyone here realizes that nobody will ever come to any real agreement about the real-world value of HT. All that can be said about it with any degree of certainty is that

1). it exists
2). it's easy for Intel to put it on their long-piped P4s without increasing transistor counts or CPU costs considerably
3). in most cases, it doesn't hurt performance
4). it can be turned off

It's a perk of owning a P4, and nothing to write home about, really. Any benefit it provides is nothing compared to what you'll get from a multi-CPU setup. It doesn't really hurt anyone to use it, in most cases, and if it does, it can be deactivated. Few, if any, people seem to be able to agree on whether or not it is really useful, and that includes those who have actually used P4s and have multitasked extensively while doing so.

again, there are many benchmarks, articles regarding the usefulness of HT. i do however agree that it's not really the point of this thread (although some interesting posts have been resulted from its discussion)

As far as speculation regarding dual-core CPUs from AMD is concerned, I would recommend checking this little snippet here

http://www.tomshardware.com/ha...s/20041005_205914.html

It's short on information, but AMD seems to think they can get a net performance boost out of their dual-core Opterons, even with clock speeds 600-1000 mhz lower than those found on their single-core Opteron predecessors. It should be interesting to see if they can back up those claims.

little early to tell, but i certainly think it's the correct direction to pursue.

 

[AnnoyedGrunt]

Cainam,

Were you able to try running a single DAOC instance with your A64 underclocked to 1/2 speed as I mentioned above?

I'd be interested in seeing how it performed.

Thanks,
D'oh!

 

[CaiNaM]

Originally posted by: AnnoyedGrunt
Cainam,

Were you able to try running a single DAOC instance with your A64 underclocked to 1/2 speed as I mentioned above?

I'd be interested in seeing how it performed.

Thanks,
D'oh!

sure. from the "for what it's worth department":

2ghz (10x200) Avg FPS: 67.258 (Min: 37 - Max: 96)
1ghz (5x200) Avg FPS: 43.900 (Min: 3 - Max: 104)

 

[Shenkoa]

"I've used Athlon, Tbird, Xp, Barton, A64, p4c and p4b. They all multitask the same for me. In my opinion you are full of the dookie. "

Yeah my experients is full of dookie, I cant say what I want to say because of the sensorship in Amerika these days. But I will say that you are very closed minded. Stick with your AMD if it seems better for you as for me, well I allready stated my experients with both processors and there is no right or wrong to it.

 

[iwantanewcomputer]

please not that i am not a fanboy for either side.

i've got the following systems:
p4b 2.8 GHz 533fsb, single channel ram
p4c 2.6 800 fsb @ 3.4 GHz with 261 fsb, 1:1 dual channel ram
amd 64 3000 at 2.4 GHz with 240 fsb and single channel 1:1 ram

while doing background divx encoding the p4b 2.8 is noticably slower than either of the others at normal windows operation/internet surfing, etc. on the athlon 64 and p4c the athlon 64 definatly seems faster at completing random tasks in windows, etc, but unless careful attention is paid to the time it takes to do a task on both, they are practically the same.

that said, i think the only way to settle this is for anandtech to throw together some benches, but for me the multitasking performance is a non-issue and i'd go with whatever is faster for single tasks

 

[CaiNaM]

Originally posted by: iwantanewcomputer
please not that i am not a fanboy for either side.

i've got the following systems:
p4b 2.8 GHz 533fsb, single channel ram
p4c 2.6 800 fsb @ 3.4 GHz with 261 fsb, 1:1 dual channel ram
amd 64 3000 at 2.4 GHz with 240 fsb and single channel 1:1 ram

while doing background divx encoding the p4b 2.8 is noticably slower than either of the others at normal windows operation/internet surfing, etc. on the athlon 64 and p4c the athlon 64 definatly seems faster at completing random tasks in windows, etc, but unless careful attention is paid to the time it takes to do a task on both, they are practically the same.

that said, i think the only way to settle this is for anandtech to throw together some benches, but for me the multitasking performance is a non-issue and i'd go with whatever is faster for single tasks

if it's a non-issue to you as you'd "go with whatever is faster for single tasks", i can appreciate that. as i and others have stated, not everyone uses their pc's in the same way or for the same purpose, nor the same standards. that's cool.

would be interesting since you have simailar systems to me if you'd try the rthdribl sequence as mentioned earlier in this thread and post the differences you find, if any.

as for the 2.8b, it's normal it's noticeably slower due to differences in clockspeeds & architecture. one thing tho; regarding the daoc thing that started me down this path, i always attributed the perf of the 2.0b to clockspeed, which is why the barton handles daoc better, and while that certainly would have something to do with it (it is certainly "choppier"), and the second daoc client doesn't go linkdead as it does with the a64, it does lose "stick" and is more troublesome than even the barton. i'm not sure clockspeed is the only reason however, as even the barton handles 2 clients better than the a64. this got me thinking tho... while the barton doesn't have the "horsepower" of the a64, nor the HT architecture, what is does have is 2x the memory bandwidth as it's running dual channel ddr400.

does anyone think it's possible that memory bandwidth limitations is contributing to what's causing this to be on the a64? the fact that glugglug had the same results when multitasking the rthdribl, but requiring 1 more instance to see the "hit" i see (he's running s939), seems to at least indicate this could part of the issue....

one of the downsides of the on-die memory controller - can't just throw in a dual channel mb and check this; have to replace cup as well

 

[BDSM]

CaiNaM .. hey there!.. The Athlon xp (thoroughbred or barton cores) don't have more bandwidth than the a64!. Yes the nforce 2 is dual channel, but the athlon xp only has a single channel itself.. So most of that bandwidth is wasted if not used with on board gfx.

 

[CaiNaM]

Originally posted by: BDSM
CaiNaM .. hey there!.. The Athlon xp (thoroughbred or barton cores) don't have more bandwidth than the a64!. Yes the nforce 2 is dual channel, but the athlon xp only has a single channel itself.. So most of that bandwidth is wasted if not used with on board gfx.

meh.. you're right.. ok, scratch that thought

more theoretical bandwidth, but xp limited by its fsb.... course it still might hold water regarding the diff between mine and glug's?

 

[Lithan]

I doubt mem bandwidth is the issue. In most instances the limits at which the cpu can address/recieve the data the memory sends hits a wall before or at the same time as the memory interfaces bandwidth. And since dcDDR does so little for A64's I'd guess that it's safe to say that it's in the same boat as Axp's and in short doesn't really need the extra memory bandwidth the way P4's do.

But Iwant, I agree that the reason your p4b trails is that it is clocked slower, along with it's much lower FSB, and no dcDDR. My p4b @ 3800mhz 224fsb dcDDR multitasks and singletasks just fine.

 

[AnnoyedGrunt]

Originally posted by: CaiNaM

Originally posted by: AnnoyedGrunt
Cainam,

Were you able to try running a single DAOC instance with your A64 underclocked to 1/2 speed as I mentioned above?

I'd be interested in seeing how it performed.

Thanks,
D'oh!

sure. from the "for what it's worth department":

2ghz (10x200) Avg FPS: 67.258 (Min: 37 - Max: 96)
1ghz (5x200) Avg FPS: 43.900 (Min: 3 - Max: 104)

Cool, thanks!

I thought the point of the test was readily apparent but perhaps that wasn't the case. Just to clarify, there was a significant amount of discussion regarding the efficiency of the A64 vs. the P4, and the thought was that perhaps the P4 maintained a higher overall speed with 2 intances because it was better able to keep its stages filled. There was a question regarding how the A64 would handle that since it perhaps was more maxed out with a single instance. I therefore thought that you could simulate the resources one of your two instances would have available by running a single instance @ a lower clock speed. This would eliminate any multitasking questions from the OS and show how that instance would perform in an "equal" multi-tasking environment.

I was thinking that if the speed on slower CPU was still quite fast, then you could eliminate the hardware aspect from your investigation and focus more on the software side. However, since your minimum framerate dropped down to 3 FPS, it looks like maybe the problem could still be hardware related. Without knowing more details about the playability on the slower CPU during the test, and whether or not you could maintain a link to the server, it's hard to draw any accurate conclusions, but perhaps you can go into more detail regarding the performance and compare the "slow CPU" performance to the "2-instance" performance.

Finally, perhaps my theory is flawed and this test doesn't really show anything, but I thought the results might shed some light on the situation.

Thanks for indulging my curiosity, even if it wasn't worth anything.

-D'oh!