TR Forums

"AMD's X2 Faulty Multitasking, Over At Toms."

Tom's Hardware guide is doing a live multitasking test of the new AMD X2 4800++ and the Intel 840EE CPU's. What is being revealed in the test is that the AMD X2 is failing to equally multitask the running applications. This AMD failure is a major concern for future customers.

In the live ongoing results, the frame rate in the looped game Farcry is neck and neck but whilst they run the game, the both processors are expected to also do dvd encoding to divx format and also loop compress files. The AMD X2 is failing to prioritise the difficult task of divx encoding and is bottlenecking that thread in the CPU's machine code prioritising stage.

It will be interesting if AMD can redesign the fault out of the coming faulty product or have window patch the kernel to somehow balance the expectations of threads to be normal.

It's more shocking news for AMD.

(please only reply if you are commenting on the topic.) (AMD have some good products but this conversation is regarding the faults in the coming X2 CPU from AMD.)

The whole project : http://www.tomshardware.com/cpu/20050603/index.html
The test : http://www.tomshardware.com/stresstest/charts.html

.

We've got an AMD forum. Please use it. Thread moved.

Kevin

Kevin wrote:

We've got an AMD forum. Please use it. Thread moved. Kevin

No problem Kevin. I thoguht this was general news. I thought also the AMD area maybe be rude are to post as the sub section has alot of fans that are so pro product/brand.

.

Hey Porkster, read this article yet? Just curious . . .

Ma10n3! wrote:

Hey Porkster, read this article yet? Just curious . . .

I've read some of it, are you referring to the PowerPC's out of order fetching of instructions from the L1 cache?

.

Ma10n3! wrote:
Hey Porkster, read this article yet? Just curious . . .


I've read some of it, are you referring to the PowerPC's out of order fetching of instructions from the L1 cache?

Ma10n3, if you're trying to refute Porkster's post, just remember that all high end x86 processors out there are out-of-order.

Porkster, were you even looking at the same article I was? Look at the Numbers on the Stress Test Page:

Lame Encoder
Encoded CDs on Intel System: 101 CDs
Encoded CDs on AMD System: 128 CDs

= AMD TEH WINNAR!

WinRAR File Compression
Packed RAR archives on Intel System: 536 Archives
Packed RAR archives on AMD System: 1132 Archives

= AMD TEH WINNAR!

Farcry Timedemo
Current Frame Rate on Intel System: 36 FPS at 480 runs
Current Frame Rate on AMD System: 38 FPS at 720 runs

= AMD TEH WINNAR!

DivX
Encoded Time on Intel System: 470 Minutes
Encoded Time on AMD System: 30 minutes

= AMD TEH WINNAR!

You really need to stop posting things you cannot back up with facts.

As we all(bar the cpu devs) know it's too out of our league to know what is exactly happening with the AMD X2's failure to task properly. It maybe that it's putting priority on the Farcry task since it maybe compiled to suit multitasking threads, or something.

Like any tests usign so called real world applciations and as utilities, you have to consider the results of the test on the day. Obviously there can always be patching or rebuilds to suit a CPU or their issues.

.

newtrip wrote:

Porkster, were you even looking at the same article I was?

Yes Newtrip, but you're reading them incorrect. examine them a bit more deeply and you will see what is being said.

The AMD is beating the Intel test system at LAME encoding and the WinRAR, but it's ignoring the Divx encoding. When you factor in the divx encoding you see that there is a major problem with the AMD X2.

.

Porkster wrote:

Yes Newtrip, but you're reading them incorrect. examine them a bit more deeply and you will see what is being said.

The AMD is beating the Intel test system at LAME encoding and the WinRAR, but it's ignoring the Divx encoding. When you factor in the divx encoding you see that there is a major problem with the AMD X2.
.

Sorry Porkster, but when I read reviews I normally want to see smaller encoding times with my processor. Not bigger ones. With video encoding Smaller Numbers = Less time encoding = AMD Winning!

P4 is better at Media Encoding which is pretty darn obiviously. I suspect there's a software glitch for the K8 in the Divx encoding program they used. Dual-Core stuff just was ironed-out, and Divx encoding is known to be a netburst-whore.

Anway, long-term stress tests at stock speeds and voltages are pretty dumb if you ask me. Unless, you're looking for failing memory, cooling and PSUs as curpits for recent system instablity.

newtrip wrote:

but when I read reviews I normally want to see smaller encoding times with my processor.

DivX
Encoded Time on Intel System: 470 Minutes
Encoded Time on AMD System: 30 minutes

Means that during the test, the Intel has been able to encode 470 mintues of a DVD data, when the AMD X2 could only do 30 mintues.

The AMD X2 is bottlenecking the Divx encode, so if it was force to equally do tasks then the other results would probably be vastly different. The signs are that the X2 maybe a great single task CPU but a bad multitasking CP Unit.

Crayon Shin Chan, actually I was was just curious as to what Porkster's perspective was on the results of that article. It appears Porkster is homing in on the details of how the processors operate instead of what the benchmark results are.

Krogoth225, you are reading that wrong. You want your video to take a shorter time to encode, not a longer one. Bigger graphs do not always equal better performance. In this instance, the smaller graph is the winner. (Less time is better)

Take a look at the ExtremeTech review with Divx encoding performance, this shows that the AMD X2 beats out Intel's processors.
http://www.extremetech.com/article2/0,1 ... 754,00.asp

Look at the graph on the link I just posted. Even if what Porkster was saying was true, why would Tom's come up with completely different results than another hardware site that compared the same exact two processors? (Well that may be a question for another thread).

Krogoth255 wrote:

P4 is better at Media Encoding which is pretty darn obiviously. I suspect there's a software glitch for the K8 in the Divx encoding program they used. Dual-Core stuff just was ironed-out, and Divx encoding is known to be a netburst-whore.

Lets hope so as alot of customers maybe buying into something with a problem if they expect the processor to be a work horse.

It would be good if Tom could change the priority of the divx thread and see if he can force the AMD X2 to do the job as it's clearly pushing the task to the side.

.

newtrip wrote:

Sorry Porkster, but when I read reviews I normally want to see smaller encoding times with my processor. Not bigger ones. With video encoding Smaller Numbers = Less time encoding = AMD Winning!

Read more closely. The chart isn't measuring encoding time, but how many minutes of video have been encoded....And AMD is gettting the crap beat out of them.

I don't really see how this can be a cpu issue...it's gotta be the already f'ed up Windows task scheduler.

Right now it looks like the only benchmark Intel is winning at is DivX encoding. Could just be a goof with their program though.

Ma10n3! wrote:

Crayon Shin Chan, actually I was was just curious as to what Porkster's perspective was on the results of that article. It appears Porkster is homing in on the details of how the processors operate instead of what the benchmark results are.

Single benchmark results aren't always a clear sign of what's going on, especially on a multitasking cpu being tested with single strain bench tests.

Also I'm more interest in general test programs, usually, that are programmed in a fair way.

Even if the Intel is better at Divx encoding I'm looking for an average good performance on all the tasks expected.

.

khjkghjk

You know Porkster, to make a claim as to what processor is a better multi-tasker for business applications, maybe the business applications should be used for the comparison; tasks the business'es IT folk may actually care about.

How many folks out there actually try to DIVX encode, compress MP3s, and play a DirectX 9.0+ game all at the same time on the same machine?

daveagn wrote:

Right now it looks like the only benchmark Intel is winning at is DivX encoding. Could just be a goof with their program though.

Yeah, this IS Toms afterall! Combine Toms+crappy task scheduling, and what do you get?...

TheDVDMan wrote:

...I don't really see how this can be a cpu issue...it's gotta be the already f'ed up Windows task scheduler.

I'm guessing it's an issue with the fetching prioritising. If so AMD should hold back on the X2 release. It maybe good for them to do so anyway, so as to wait for the new socket they plan.

.

Divx encoding numbers read the number of minutes worth of data that was encoded.

I just realize that this is the only time HT actually makes a difference in the 840XE. Divx is optimized for HT and loves the Netburst architech. That's why 840XE is handling the Divx encoding far better under heavy stress conditions then the X-2 4800. I suspect this advanage is only limited to apps that are optimized for HT.

In 90% of multi-tasking enviorments and multi-threaded apps the 4800 is still better then the 840XE.

Porkster wrote:

Single benchmark results aren't always a clear sign of what's going on, especially on a multitasking cpu being tested with single strain bench tests.

Also I'm more interest in general test programs, usually, that are programmed in a fair way.

What exactly do you mean by "single strain bench tests?"

So, general benchmark programs have more significance than benchmarks of the actual applications people may be running on the multi-core/cpu systems?

You know, the DIVX encoder is an actual application, not a generalized benchmark program.

Krogoth255 wrote:

I just realize that this is the only time HT actually makes a difference in the 840XE. Divx is optimized for HT and loves the Netburst architech. That's why 840XE is handling the Divx far better under heavy stress conditions then the X-2 4800. I suspect this advanage is only limited to apps that are optimized for HT.

The important thing to notice is that while playing Farcry on the Intel 840EE at the same frame rates as the AMD X2 system, you can encode a Divx, a music CD and a RAR at the same time, fairly well.

On the AMD X2, you may get faster Lame encoding and RAR but at the expense of the Divx encoding. The Divx encoding is nearly at a stand still on the AMD X2.

.

Porkster wrote:

On the AMD X2, you may get faster Lame encoding and RAR but at the expense of the Divx encoding. The Divx encoding is nearly at a stand still on the AMD X2.

So where is the issue with the X2's again?

Be hilarious if they update with "whoops, bob forgot to set divx's priority back to normal, sorry."

Anyway, I think one should hold off on conclusions until they start analyzing their data. It's certainly going to be interesting to see what's going on.

In other news, I thank jeebus for TR every time I see Tom's site design.

Ma10n3! wrote:

So, general benchmark programs have more significance than benchmarks of the actual applications people may be running on the multi-core/cpu systems?

Would you trust a brenchtest result when you own a ATI graphcis card but the game has Nvidia plasted all over the product?

It is a fact, some companies allow for pro-advertising. Pro-Advertising is where another product is used to bolster another. Much like TV station do when one tv show will promote another within its content. The game develeopers do the same and will tailor a game to suit a paying client's promotion, be it Nvidia or AMD, or Intel, etc.

Doom3 is a Nvidia and AMD paid game, so I don't take accoutn of it's results in speed tests.

.

Porkster wrote:

Krogoth255 wrote:

I just realize that this is the only time HT actually makes a difference in the 840XE. Divx is optimized for HT and loves the Netburst architech. That's why 840XE is handling the Divx far better under heavy stress conditions then the X-2 4800. I suspect this advanage is only limited to apps that are optimized for HT.

The important thing to notice is that while playing Farcry on the Intel 840EE at the same frame rates as the AMD X2 system, you can encode a Divx, a music CD and a RAR at the same time, fairly well.

On the AMD X2, you may get faster Lame encoding and RAR but at the expense of the Divx encoding. The Divx encoding is nearly at a stand still on the AMD X2.

.

HT advantage in multi-tasking and threading for the 840XE only works with apps that are optimized for it, otherwise it goes to waste.

Most of the dual-cores CPUs are current geared towards the non-gaming crowd. 840XE is t3h win in heavy-duty media encoding if $$$$ is a non-issue. 4800 is better at everything else. What I interested in is how the normal 840 fares against the 840XE in stress test, just to see the benefit of HT.

Porkster wrote:

DivX
Encoded Time on Intel System: 470 Minutes
Encoded Time on AMD System: 30 minutes

Means that during the test, the Intel has been able to encode 470 mintues of a DVD data, when the AMD X2 could only do 30 mintues.

The AMD X2 is bottlenecking the Divx encode, so if it was force to equally do tasks then the other results would probably be vastly different. The signs are that the X2 maybe a great single task CPU but a bad multitasking CP Unit.

Actually, the way I read this review, well, lemme quote from the review itself, on this page: http://www.tomshardware.com/cpu/20050603/stresstest-10.html

James Bond DVD Compression. We count the total minutes since commencement of the stress test.

It seems that they're counting the amount of time it takes to compress the entire movie "Die Another Day" into the Divx format. It takes the AMD processor 30 minutes to do this, and the Intel processor 470 minutes. It's illogical to assume it's the other way around on the sheer basis that there's not 470 minutes of stuff on a single DVD to encode, so this is likely the correct interpretation, which means that it is Intel that is having its butt handed to it on a plate.

I can even explain this discrepancy rather easily: They're using the dual-core P4 "Extreme Edition", with Hyperthreading enabled. This means there are two physical cores and two "virtual" cores, and the OS and multithreaded programs aren't really at the stage where they can efficiently discriminate between the two, so they just kind of throw things at the first and then second "core" they see, whether it's physical or virtual. I'm betting that this was a dual-threaded app, and that the first "core" being fed was the first physical core, and the second "core" being fed was the "virtual" HT core on that same first physical core. This would leave the second physical core unused while the first core continually conflicts with itself trying to process two threads simultaneously.

While HT offers some performance benefits, those benefits come from scenarios where data is not being fed to the CPU in a continuous stream. In things like video encoding, HT tends to be a hindrance, sometimes a major one. This is exactly the kind of horrific performance I'd expect to see on the P4 dual-core with HT enabled, and thanks to this page, we're sure it is: http://www.tomshardware.com/cpu/20050603/stresstest-09.html

So I think this is actually revealing a well-known flaw in the dual-core P4 EE, that being the inability for the OS to discriminate between physical and virtual cores. The only way to get around this "flaw" is to disable HT on the processor altogether, which means losing the benefits of HT in situations where it would actually benefit, or rebooting every time you switch applications on your computer so you can reset the HT enabled/disabled toggle in the BIOS.

Thanks, Porkster, for bringing this gaping flaw in a $1000 Intel processor to our attention!