AMD's Phenom Processor @ lostcircuits.com

[Corsa]

AMD's Phenom Processor - Beyond Erratum 298

What we did was to try to get an idea of where AMD’s Phenom CPU could be placed in the current CPU landscape. Initially, we didn’t have very high hopes, especially after reading some of the early reviews. Then we started digging and, to be sure, Phenom is not a diamond in the rough, but it turned out to be better than what we anticipated.

What it boils down to is that there is a tremendous potential embedded in Phenom that does not come into play with the existing software infrastructure. Particularly single threaded games are a lost cause and unfortunately, this is still the standard of the industry. UnrealTournament3 in this respect is a milestone also for the gaming industry and there are other games to follow in the same footsteps that employ more artificial intelligence and PhysX effects, which will cause most dual core processors to choke. But then, there is also the problem of the GPU limitations. Well… as long as nVidia still has high end cards ….

[Charles Wirth]

Pretty good writeup, lots of benches to measure up with. I didnt see mention of the timing on the 9900 system.

In all fairness, crysis numbers.

[Mathos]

From what I've seen they can get even better numbers if they up the nb/imc clock. Doing so outright raises the cpu scores in 3dmark. Almost a 400pt difference between my cpu at 2.6 by 2.0 and 2.6 by 2.4.

[Aussie FX]

I wouldn't take Crysis into account. You only have to look at the big intel logo at the start to guess whos compiled it.

One thing that gives me the s**ts about these review sites, is that they see Phenom flogging intels finest cpus like the ut3 bench above yet nobody has asked why.
It seems to me that highly multithreaded games win hands down with Phenom.
Bioshock also gives intels finest a kick in the ar**.

If intel won those benchmarks they would rave about it, yet when AMD wins no one wants to comment.
That's why I believe most review sites these days are intel shills.

Thanks for the thread Corsa.

/rant

[SQjay]

Pretty good writeup, lots of benches to measure up with. I didnt see mention of the timing on the 9900 system.

In all fairness, crysis numbers.

In Crysis, the situation is similar to FarCry and World In Conflict with just a bit more than 25% CPU utilization according to Windows Task Manager.

Trully a modern game... comapred to x-years old FarCry!

[YukonTrooper]

Rough. Takin' out the garbage.

[GenTarkin]

Pretty good writeup, lots of benches to measure up with. I didnt see mention of the timing on the 9900 system.

In all fairness, crysis numbers.

All I can say is WOW!!!! (even though I hate that game =P ) hahah anyways...
that crysis chart and all those other charts according to their measurements show games that dont use multithreading and if so, dont use it well at all.
They said that crysis was showing 25% on the CPU...which makes sense...and if thats true then we really have a poorly multithreaded enabled game or a single threaded game. just like world in conflict benchmarks they showed.
What really pumps me up...is those 2 or so games they showed in which they used multithreading...I think was it FEAR and UT3..The Phenom nearly blows everything away, especially in UT3. Fear its up there with the big dog intels.

That says a whole heck of a lot for the phenom we all thought was weak and slow and yeah....
If how they measured those things is true ... then the phenom is simply a multithreaded capable beast of a processor.
I dont know about you guys, but I actually never saw this article and was quite saddened about all the others I read where the phenom is sucky and bla bla bla....but I saw this one and actually seen the difference between its multithreaded capability vs its weak single threaded capability(mind you its still stronger than K8 architecture in single threaded)
and I got a surge of excitement seeing that this phenom is truely special when it comes to multhreaded games at least.
Thats the way of the future and I now....strongly believe phenom is here to stay and more improvements to come.
So, despite its seemingly weak performance, I believe the exact opposite is true about it, and will be revealed in the future.
So, heres to AMD to keep chuggin along cuz yes...I love AMD and I know they will come through.
I mean, despite the phenoms bad overclockability at this point in time...at stock speeds phenom vs intel I believe its quite competitive.
I also dont see intel as being as "innovative" by slapping 2 C2D together one package and calling it a quad...sure it works amazing and it highly clockable...but thats only as far as it can go.
There is really nothing too much innovative behind intels Quads.
AMD on the other hand has bit the bullet and instead of slapping 2 X2's together and getting their manufacturing process as good as can be and clocking them as high as possible...they set out to make something nearly from the ground up.
And right now it stands, its just a baby teething...once that happens and then when the tooth fairy comes to visit to take out the baby teeth and the adult teeth grow in....the phenom will be something superior.

[Charles Wirth]

sure it works amazing and it highly clockable...but thats only as far as it can go.

Right...

Look at the other multithreaded benches, Phenom get spanked over and over. UT3 appears to have code that favors the platforms memory latency/ODMC.

Like all new stuff, give it time to mature.

Innovation is not clipping a core to meet TDP spec, squeezing six cores under the lid is.

[KTE]

LOL guys this is very old, it's already been posted and covered 2 months back

[Corsa]

"Spider" Platform:

* ASUS M3A32-MVP (0603 BIOS)
* AMD Phenom 9900, 9600
* 2 x 1 GB OCZ PC2-7200 Dual Channel Platinum XTC modules
* OCZ PowerStream 520ADJ
* All benchmarks run at 1066 MHz DRAM frequency ++
* DRAM Auto Timing - seems to be 4-4-3-15
* DRAM Controllers set to "unganged mode"
* AOD "red"
The QX9770 was run at 8 x 400 MHz with DDR3 1600 at 8-8-8-27 latency settings

UnrealTournament3 heavily taxes the CPU with artificial intelligence of the bots and PhysX calculations and runs easily up to about 80% CPU utilization whenever the scene gets crowded with bots. Even on a quad core system, multiple bots in the same scene can cause the system to choke.

3.2Ghz Intel QX9770 cant match the AMD Phenom Processor Model 9600 2.3GHz

Arguably, there is that thing called legacy support and legacy applications are still prevailing in today’s computing environment. On the other hand, there are truly multithreaded applications and as it turned out, the more thread level parallelism is used in any application, the better fared the Phenom processor.

Like a coiled spring Phenom lies and waits for applications that will allow it to strike, meanwhile the QX is making the most of what we have today
edit: searched b4 i posted KTE and didn't find anything? ....my bad

[stm]

the ut3 benchmarks seem to be inconsistent compared to other ut3 benchmarks anways..
http://www.pcgameshardware.de/?menu=...mage_id=730203

[Boschwanza]

http://img134.imageshack.us/my.php?i...nebenchzp1.jpg

Phenom @ 3.2 GHz

11727 CB-CPU on x32 !

[techguy2k5]

I can't believe anyone could be well-informed and still be an AMD fan at this point. You have to completely ignore reality to pull that one off. Phenom wins one bench and all of a sudden it's "a coiled spring waiting to strike" :roll:

I used to be one of the biggest self-admitted AMD fanboys on these here intarwebs, btw. I just woke up one day when they utterly failed for the umpteenth time and had no choice but to go with Intel + Nvidia for my current build.

[grunge100]

Fugger,

do you have a link concerning UT3's code and how it is optimized????

[Charles Wirth]

Nope, clueless to the UT3 coding. But when the shoe is on the other foot, meaning Crysis, the claims of coding in favor of Intel seems to be the consensus. No fault of AMD's that it performs sub par?

Boshwanza, thanks link, some progress.

[JumpingJack]

AMD's Phenom Processor - Beyond Erratum 298

He mis-represented this benchmark ... he ran capped. Just look at his data ... does any of it make sense, if it such a great mulithreaded engine, does any thing make sense dual core to quad core. I mean, how does a 5000+ at 2.6 GHz beat an FX62 at 2.8 Ghz?

This entire review was botched from top to bottom. I am was extremely disappointed with LC on this run.

He claims A) FEAR is multithreaded, it is not. B) he also claims Crysis is not multithreaded and it is, he simply lowered the CPU heavy settings to their minimum and as a result did not produce a high enough load to demonstrate all 4 cores working... C) he ran UT3 using UTBENCH3 but installed the demo, such that UTBENCH3 did not configure the botmatch they way he thought it was intended.

Better yet, he comments on the erratum, and uses an Asus board (M3A) to demonstrate the 'effects', yet the BIOS verison (603) had not incorporated the TLB patch, that was not cut in until the leaked 701 and is not present until 801. He benched the erratum patch without actually using the friggin' patch.

[JumpingJack]

the ut3 benchmarks seem to be inconsistent compared to other ut3 benchmarks anways..
http://www.pcgameshardware.de/?menu=...mage_id=730203

It's inconsistent because he ran capped. The 1.1 patch of UT3 added this to the engine.ini file:

[Engine.GameEngine]
bSmoothFrameRate=TRUE
MinSmoothedFrameRate=22
MaxSmoothedFrameRate=62

Now, the utility he used makes a copy of this to the user profile directory in the My documents folder and updates that code to read:

[Engine.GameEngine]
bSmoothFrameRate=TRUE
MinSmoothedFrameRate=22
MaxSmoothedFrameRate=99

The problem is this... and this is where my argument has a hole because he does not provide all the details of how he ran the bench, but it was the only way I could reproduce anything close to his numbers... if you install the DEMO version of UT3, it defaults to Unreal Tournament 3 Demo as the folder name, to get UT3BENCH to run, you have to rename UT3DEMO.EXE to UT3.EXE, and it will run... the problem is that it does not config the INI files in the user profile correctly and Smooth stays true and frame rate stays 62.

So here is the experiment you want to run to verify ...
1. Install a retail copy of UT3
2. Download UT3BENCH from http://olrac.org/ut3bench and copy the executable to the UT3/Binaries directory.
3. Run UT3BENCH and select the DM-HEATRAY-BOT option, leave bots at 12, time at 60, and on the HW settings tab select 1 for low quality settings or 5 for high quality.

If you run it using a decent card (3870 or 8800 GT), you will get completely different numbers than he did....

Here is 4 runs, the first two were at HW settings of 5,5 the second set of two were at 1,1 this was done on a Phenom 9600 BE at stock conditions (TLB disabled), 8800 GTX, 2 gigs of ram ...

Dumping FPS chart at 2008.03.07-20.37.53 using build 3521 built from changelist 206281
Bucket:  0 -  5  Time:  3.73  Frame:  0.14
Bucket:  5 - 10  Time:  0.00  Frame:  0.00
Bucket: 10 - 15  Time:  0.89  Frame:  0.17
Bucket: 15 - 20  Time:  0.49  Frame:  0.12
Bucket: 20 - 25  Time:  0.36  Frame:  0.12
Bucket: 25 - 30  Time:  0.66  Frame:  0.26
Bucket: 30 - 35  Time:  0.82  Frame:  0.38
Bucket: 35 - 40  Time:  1.58  Frame:  0.83
Bucket: 40 - 45  Time:  2.58  Frame:  1.56
Bucket: 45 - 50  Time:  3.79  Frame:  2.56
Bucket: 50 - 55  Time:  4.98  Frame:  3.70
Bucket: 55 - 60  Time:  8.15  Frame:  6.64
Bucket: 60 - 99  Time: 71.96  Frame: 83.52
4218 frames collected over 59.55 seconds, disregarding 0.01 seconds for a 70.84 FPS average, 93.85 percent of time spent > 30 FPS
Average GPU frame time: 0.00 ms




Dumping FPS chart at 2008.03.07-20.39.52 using build 3521 built from changelist 206281
Bucket:  0 -  5  Time:  2.54  Frame:  0.10
Bucket:  5 - 10  Time:  0.00  Frame:  0.00
Bucket: 10 - 15  Time:  0.79  Frame:  0.13
Bucket: 15 - 20  Time:  0.72  Frame:  0.15
Bucket: 20 - 25  Time:  0.44  Frame:  0.13
Bucket: 25 - 30  Time:  0.36  Frame:  0.13
Bucket: 30 - 35  Time:  0.26  Frame:  0.10
Bucket: 35 - 40  Time:  0.31  Frame:  0.15
Bucket: 40 - 45  Time:  0.47  Frame:  0.25
Bucket: 45 - 50  Time:  1.24  Frame:  0.73
Bucket: 50 - 55  Time:  2.00  Frame:  1.32
Bucket: 55 - 60  Time:  3.96  Frame:  2.85
Bucket: 60 - 99  Time: 86.90  Frame: 93.97
4780 frames collected over 59.44 seconds, disregarding 0.00 seconds for a 80.40 FPS average, 95.14 percent of time spent > 30 FPS
Average GPU frame time: 0.00 ms




Dumping FPS chart at 2008.03.07-20.41.40 using build 3521 built from changelist 206281
Bucket:  0 -  5  Time:  2.81  Frame:  0.06
Bucket:  5 - 10  Time:  0.00  Frame:  0.00
Bucket: 10 - 15  Time:  0.13  Frame:  0.01
Bucket: 15 - 20  Time:  0.38  Frame:  0.04
Bucket: 20 - 25  Time:  0.30  Frame:  0.04
Bucket: 25 - 30  Time:  0.18  Frame:  0.03
Bucket: 30 - 35  Time:  0.21  Frame:  0.04
Bucket: 35 - 40  Time:  0.32  Frame:  0.08
Bucket: 40 - 45  Time:  0.32  Frame:  0.09
Bucket: 45 - 50  Time:  0.18  Frame:  0.05
Bucket: 50 - 55  Time:  0.32  Frame:  0.11
Bucket: 55 - 60  Time:  0.35  Frame:  0.13
Bucket: 60 - 99  Time: 94.53  Frame: 99.31
9272 frames collected over 59.81 seconds, disregarding 0.00 seconds for a 155.00 FPS average, 96.21 percent of time spent > 30 FPS
Average GPU frame time: 0.00 ms




Dumping FPS chart at 2008.03.07-20.43.23 using build 3521 built from changelist 206281
Bucket:  0 -  5  Time:  2.90  Frame:  0.07
Bucket:  5 - 10  Time:  0.24  Frame:  0.01
Bucket: 10 - 15  Time:  0.00  Frame:  0.00
Bucket: 15 - 20  Time:  0.21  Frame:  0.02
Bucket: 20 - 25  Time:  0.37  Frame:  0.06
Bucket: 25 - 30  Time:  0.18  Frame:  0.03
Bucket: 30 - 35  Time:  0.25  Frame:  0.06
Bucket: 35 - 40  Time:  0.14  Frame:  0.03
Bucket: 40 - 45  Time:  0.35  Frame:  0.10
Bucket: 45 - 50  Time:  0.11  Frame:  0.03
Bucket: 50 - 55  Time:  0.19  Frame:  0.07
Bucket: 55 - 60  Time:  0.35  Frame:  0.14
Bucket: 60 - 99  Time: 94.72  Frame: 99.37
8752 frames collected over 59.80 seconds, disregarding 0.00 seconds for a 146.31 FPS average, 96.11 percent of time spent > 30 FPS
Average GPU frame time: 0.00 ms

Jack

[theinternet]

at least 5000+BE is still competitive
at only $99 it'd be my gaming cpu of choice on the cheap

[JumpingJack]

at least 5000+BE is still competitive
at only $99 it'd be my gaming cpu of choice on the cheap

Yeah, I have a 5000+ BE and this is probably AMD's best dual core... OCs well, and so long as FPS stays above 60 ... who cares ?

EDIT: It is the G2 stepping, much improved over the G0.

[JumpingJack]

Fugger,

do you have a link concerning UT3's code and how it is optimized????

http://www.unrealtournament3.com/us/index.html

There is a big Intel logo at the bottom

[KTE]

Jack: what do you get if you run the same two benches as you outlined above with your QX, using the same GPU?

Also, while interested in throwing the bottleneck the CPU way rather than GFX numbers, what res. is that bench running at HW 1.1 and what was the CPU low, av, max usage?

I have all the versions of the full and the demos, so I can check them out later. Thanks for the info and some test results.

[grunge100]

Jack,

The link you provided is nice but has nothing to do with the game engine coding.

Regards

[kretfr]

http://img134.imageshack.us/my.php?i...nebenchzp1.jpg

Phenom @ 3.2 GHz

11727 CB-CPU on x32 !

Can you give me CPU-Z validation and some information about other hardware (mobo, RAM)? Results of Super Pi 1M/32M?

[Thesavage]

It's inconsistent because he ran capped. The 1.1 patch of UT3 added this to the engine.ini file:

[Engine.GameEngine]
bSmoothFrameRate=TRUE
MinSmoothedFrameRate=22
MaxSmoothedFrameRate=62

Now, the utility he used makes a copy of this to the user profile directory in the My documents folder and updates that code to read:

[Engine.GameEngine]
bSmoothFrameRate=TRUE
MinSmoothedFrameRate=22
MaxSmoothedFrameRate=99

The problem is this... and this is where my argument has a hole because he does not provide all the details of how he ran the bench, but it was the only way I could reproduce anything close to his numbers... if you install the DEMO version of UT3, it defaults to Unreal Tournament 3 Demo as the folder name, to get UT3BENCH to run, you have to rename UT3DEMO.EXE to UT3.EXE, and it will run... the problem is that it does not config the INI files in the user profile correctly and Smooth stays true and frame rate stays 62.

So here is the experiment you want to run to verify ...
1. Install a retail copy of UT3
2. Download UT3BENCH from http://olrac.org/ut3bench and copy the executable to the UT3/Binaries directory.
3. Run UT3BENCH and select the DM-HEATRAY-BOT option, leave bots at 12, time at 60, and on the HW settings tab select 1 for low quality settings or 5 for high quality.

If you run it using a decent card (3870 or 8800 GT), you will get completely different numbers than he did....

Here is 4 runs, the first two were at HW settings of 5,5 the second set of two were at 1,1 this was done on a Phenom 9600 BE at stock conditions (TLB disabled), 8800 GTX, 2 gigs of ram ...

Dumping FPS chart at 2008.03.07-20.37.53 using build 3521 built from changelist 206281
Bucket:  0 -  5  Time:  3.73  Frame:  0.14
Bucket:  5 - 10  Time:  0.00  Frame:  0.00
Bucket: 10 - 15  Time:  0.89  Frame:  0.17
Bucket: 15 - 20  Time:  0.49  Frame:  0.12
Bucket: 20 - 25  Time:  0.36  Frame:  0.12
Bucket: 25 - 30  Time:  0.66  Frame:  0.26
Bucket: 30 - 35  Time:  0.82  Frame:  0.38
Bucket: 35 - 40  Time:  1.58  Frame:  0.83
Bucket: 40 - 45  Time:  2.58  Frame:  1.56
Bucket: 45 - 50  Time:  3.79  Frame:  2.56
Bucket: 50 - 55  Time:  4.98  Frame:  3.70
Bucket: 55 - 60  Time:  8.15  Frame:  6.64
Bucket: 60 - 99  Time: 71.96  Frame: 83.52
4218 frames collected over 59.55 seconds, disregarding 0.01 seconds for a 70.84 FPS average, 93.85 percent of time spent > 30 FPS
Average GPU frame time: 0.00 ms




Dumping FPS chart at 2008.03.07-20.39.52 using build 3521 built from changelist 206281
Bucket:  0 -  5  Time:  2.54  Frame:  0.10
Bucket:  5 - 10  Time:  0.00  Frame:  0.00
Bucket: 10 - 15  Time:  0.79  Frame:  0.13
Bucket: 15 - 20  Time:  0.72  Frame:  0.15
Bucket: 20 - 25  Time:  0.44  Frame:  0.13
Bucket: 25 - 30  Time:  0.36  Frame:  0.13
Bucket: 30 - 35  Time:  0.26  Frame:  0.10
Bucket: 35 - 40  Time:  0.31  Frame:  0.15
Bucket: 40 - 45  Time:  0.47  Frame:  0.25
Bucket: 45 - 50  Time:  1.24  Frame:  0.73
Bucket: 50 - 55  Time:  2.00  Frame:  1.32
Bucket: 55 - 60  Time:  3.96  Frame:  2.85
Bucket: 60 - 99  Time: 86.90  Frame: 93.97
4780 frames collected over 59.44 seconds, disregarding 0.00 seconds for a 80.40 FPS average, 95.14 percent of time spent > 30 FPS
Average GPU frame time: 0.00 ms




Dumping FPS chart at 2008.03.07-20.41.40 using build 3521 built from changelist 206281
Bucket:  0 -  5  Time:  2.81  Frame:  0.06
Bucket:  5 - 10  Time:  0.00  Frame:  0.00
Bucket: 10 - 15  Time:  0.13  Frame:  0.01
Bucket: 15 - 20  Time:  0.38  Frame:  0.04
Bucket: 20 - 25  Time:  0.30  Frame:  0.04
Bucket: 25 - 30  Time:  0.18  Frame:  0.03
Bucket: 30 - 35  Time:  0.21  Frame:  0.04
Bucket: 35 - 40  Time:  0.32  Frame:  0.08
Bucket: 40 - 45  Time:  0.32  Frame:  0.09
Bucket: 45 - 50  Time:  0.18  Frame:  0.05
Bucket: 50 - 55  Time:  0.32  Frame:  0.11
Bucket: 55 - 60  Time:  0.35  Frame:  0.13
Bucket: 60 - 99  Time: 94.53  Frame: 99.31
9272 frames collected over 59.81 seconds, disregarding 0.00 seconds for a 155.00 FPS average, 96.21 percent of time spent > 30 FPS
Average GPU frame time: 0.00 ms




Dumping FPS chart at 2008.03.07-20.43.23 using build 3521 built from changelist 206281
Bucket:  0 -  5  Time:  2.90  Frame:  0.07
Bucket:  5 - 10  Time:  0.24  Frame:  0.01
Bucket: 10 - 15  Time:  0.00  Frame:  0.00
Bucket: 15 - 20  Time:  0.21  Frame:  0.02
Bucket: 20 - 25  Time:  0.37  Frame:  0.06
Bucket: 25 - 30  Time:  0.18  Frame:  0.03
Bucket: 30 - 35  Time:  0.25  Frame:  0.06
Bucket: 35 - 40  Time:  0.14  Frame:  0.03
Bucket: 40 - 45  Time:  0.35  Frame:  0.10
Bucket: 45 - 50  Time:  0.11  Frame:  0.03
Bucket: 50 - 55  Time:  0.19  Frame:  0.07
Bucket: 55 - 60  Time:  0.35  Frame:  0.14
Bucket: 60 - 99  Time: 94.72  Frame: 99.37
8752 frames collected over 59.80 seconds, disregarding 0.00 seconds for a 146.31 FPS average, 96.11 percent of time spent > 30 FPS
Average GPU frame time: 0.00 ms

Jack

Not to sure i agree with your verdict Jack if the game runs capped, and thats how you play it online then it must be tested like that to show how a relevant platform performs.

[JumpingJack]

Not to sure i agree with your verdict Jack if the game runs capped, and thats how you play it online then it must be tested like that to show how a relevant platform performs.

http://www.tweakguides.com/UT3_7.html

The game engine is capped to a maximum framerate, as determined by the value of the MaxSmoothedFrameRate variable (see below), which is 62FPS by default. This is done by the developers to prevent FPS spikes and thus provide smoother performance. However, by setting this option to False, you can completely remove this FPS cap.

It is a very simple experiment ... which I outlined above. He ran capped there is no other explanation. No the data I showed above ran on a Phenom 9600 BE, 8800 GTX, and the exact same board he uses (Asus M3A32-MVP). Now if you don't believe my interpretation, simply run the experiment.

It actually get's better, download the demo, load up the map in spectator, go to the ledge where he grabs his frame, look the same direction and capture on high settings and low settings.... his screen dump all things high, but he says all things low. Ok... no big deal, he flubbed the way he ran the bench.

UT3 is a hard game to use as a bench but also (as is with the other UT games) a great one to use as well ... for a couple of reasons. First, you can choose a flyby (scripted) or you can intiate an all bot match, which is essentially real game play. The problem with the latter is that you get huge variability, so you need to run it long (> a few minutes) and many times to get good enoughs statistics to compare.