ExtremeTech review of new Opteron 146 - Closest thing to the A64 this side of 9/23

[PrinceXizor]

ExtremeTech has the brand new Opteron 146 pitted against a P4-3.2Ghz and the Athlon XP 3200+

The Opteron 146 is clocked at 2.0Ghz so no O/C'ing is necessary to estimate the performance of the soon-to-be-released Athlon64. Pretty impressive gaming numbers.

Linkified for the Lazy

P-X

 

[jfall]

Looks very promising. I for one am waiting for a few months, I want to see how the 64 does. Glad to see that they put it up against a 3.2 instead of a 3.0 P4. Hope this is the beginning of a CPU war

 

[jfall]

dbl post

 

[NFS4]

I'm definitely getting an Athlon 64. Screw toasty P4's & Prescotts. My 2.4C @ 3.06GHz is already burning...even with a big-ass Zalman CNPS-7000 ALCu

 

[PrinceXizor]

Originally posted by: NFS4
I'm definitely getting an Athlon 64. Screw toasty P4's & Prescotts. My 2.4C @ 3.06GHz is already burning...even with a big-ass Zalman CNPS-7000 ALCu

Ah..c'mon...don't you want a computer that multitasks? It can render 3D scenes, encode mpegs and heat your house...all at the same time

Funny how heat dissipation is no longer such a "huge" problem when its Intel chips clocking in at 103 Watts.

P-X

 

[TMTCC]

Color me unimpressed, but that review didn't show the Opteron dominated the P4 by much. Hopefully AMD will price these things more reasonably; otherwise, I don't see much reason to be hyped up over this.

 

[rjain]

The athlon 64 will be a bit faster because it doesn't need registered ECC RAM, as any server board should require.

 

[PrinceXizor]

Originally posted by: TMTCC
Color me unimpressed, but that review didn't show the Opteron dominated the P4 by much. Hopefully AMD will price these things more reasonably; otherwise, I don't see much reason to be hyped up over this.

Breaks out the crayons. The Athlon 64 won some and lost some in the "normal" application based usages. It looked very 50-50. So you are correct there. BUT in the gaming benchmarks it cleaned the P4's clock. And the P4-3.2Ghz are still might pricey in their own rights so I wouldn't expect a huge premium (though I figure the standard Athlon 64 to be priced at or slightly below the P4-3.2Ghz. Also, don't forget about the Athlon 64 FX @ 2.2Ghz.

Also, remember, this is the A64's 32 bit performance. So you get 32 bit performance that is either slightly better or slightly worse depending on you business application, and kicks major butt in games....AND you get x86-64 goodness that you can use when 64bit Windows hits and games recompiled and/or optimized for 64 bit coding. What these "opteron" previews do is whet our appetite for the real deal when its finally released.

P-X

 

[Accord99]

If the A64 "cleaned the P4's clock" in games, then the P4 annihilated the Opteron in video encoding and 3D rendering. And there is no evidence yet that the x86-64 will do anything than provide minor changes in performance as compared to 32 bit for applications that have no need for native 64 bit integer support or flat addressing of large amounts of memory.

 

[Lyfer]

Looks promising.

 

[TMTCC]

Cleaned house eh? I looked back at the review again, and although the Opteron won a lot of the Direct3D benchmarks, it came up 2nd in the OpenGL benchmarks. Not what I call cleaning house...

 

[PrinceXizor]

Originally posted by: Accord99
If the A64 "cleaned the P4's clock" in games, then the P4 annihilated the Opteron in video encoding and 3D rendering. And there is no evidence yet that the x86-64 will do anything than provide minor changes in performance as compared to 32 bit for applications that have no need for native 64 bit integer support or flat addressing of large amounts of memory.

Fair enough. However, as far as 64-bit adressing. Its no panacea of humongous speed gains...but UT2003 optimized for 64 bit code is 20-30% faster according to the game developers. That is definitely real and definitely tangible. And the simple ability to address larger ammounts of memory WILL affect performance in certain applications that are currently affected by accessing the I/O subsystem to complete tasks, as opposed to being able to access it out of memory. Of course it won't affect current 32-bit applications nor will most current applications benefit from being jury-rigged to support x86-64. But if you can get a chip that performs 32 bit operations competively and STILL has 64 bit capabilites...why WOULDN'T you?

P-X

 

[NFS4]

Originally posted by: Accord99
If the A64 "cleaned the P4's clock" in games, then the P4 annihilated the Opteron in video encoding and 3D rendering. And there is no evidence yet that the x86-64 will do anything than provide minor changes in performance as compared to 32 bit for applications that have no need for native 64 bit integer support or flat addressing of large amounts of memory.

Didn't the P4 also get its "clock cleaned" in video encoding and multimedia apps when it first came out? Once applications start taking advantage of its optimizations, its performance took off.

 

[cow123]

general OS responsiveness got about as good as it needed to be with a 1ghz cpu anyway, i see no point in increasing the speed in that respect, only the drive subsystems need improving to see any difference. unless you want to encode divx movies and such, there should be no point in a p4 imo when there is the a64 around - gaming all the way!

 

[dullard]

Originally posted by: PrinceXizor
Fair enough. However, as far as 64-bit adressing. Its no panacea of humongous speed gains...but UT2003 optimized for 64 bit code is 20-30% faster according to the game developers. That is definitely real and definitely tangible. And the simple ability to address larger ammounts of memory WILL affect performance in certain applications that are currently affected by accessing the I/O subsystem to complete tasks, as opposed to being able to access it out of memory. Of course it won't affect current 32-bit applications nor will most current applications benefit from being jury-rigged to support x86-64. But if you can get a chip that performs 32 bit operations competively and STILL has 64 bit capabilites...why WOULDN'T you?

P-X

There is a bigger question to be asked - and as far as I know hasn't been answered yet. The makers claim that UT2003 compiled specifically for the Athlon 64 in 64 bit mode is 20%-30% faster compared to the normal UT2003 without any optimizations for AMD chips. So there are multiple changes: (1) 32 bit to 64 bit and (2) finally code optimized for an AMD chip instead of the larger market share Intel chips. How much of that 20%-30% is due to #1, and how much is due to #2? I feel that most is due to #2. Do you have any evidence otherwise?

 

[OldSpooky]

I think AMD has a winner of a CPU here, but I also think that the Empire will quickly strike back with Prescott (and Yamhill if x86-64 takes off).

Opteron utterly destroys Athlon XP and Pentium 4 in games, but it is also pretty clear that the Pentium 4 cleans up in multitasking.

 

[PrinceXizor]

Originally posted by: cow123
general OS responsiveness got about as good as it needed to be with a 1ghz cpu anyway, i see no point in increasing the speed in that respect, only the drive subsystems need improving to see any difference. unless you want to encode divx movies and such, there should be no point in a p4 imo when there is the a64 around - gaming all the way!

you bring up quite an interesting point cow...we've long since gone past the time where the label of "fastest processor" is wholly accurate. Even with Intel's "edge" over Athlon's of late. It STILL couldn't beat it out in some benchmarks.

In some applications Instructions Per Cycle is a huge factor and it doesn't matter how high your clockrate is, IPC just is too important.

In some applications clockspeed is a huge factor and it doesn't matter how high your IPC is, clockrate is just too important.

So you actually have to *gasp*, use your brain to decide which chip is right for you. If you do general business applications and gaming (like I do) then AMD is you choice. If you do media encoding and 3D rendering, then Intel is your choice. If you do a smattering of everything, you have to factor in the fact that AMD's chip can handle x86-64 instructions and memor adressing. Those that enjoy silent PC's will also have to consider that some signifcant cooling (and thus increased noise) will be require to run newer Intel chips.

It all boils down to the end user deciding which chip best suits their particular usage.

P-X

 

[PrinceXizor]

Originally posted by: dullard

Originally posted by: PrinceXizor
Fair enough. However, as far as 64-bit adressing. Its no panacea of humongous speed gains...but UT2003 optimized for 64 bit code is 20-30% faster according to the game developers. That is definitely real and definitely tangible. And the simple ability to address larger ammounts of memory WILL affect performance in certain applications that are currently affected by accessing the I/O subsystem to complete tasks, as opposed to being able to access it out of memory. Of course it won't affect current 32-bit applications nor will most current applications benefit from being jury-rigged to support x86-64. But if you can get a chip that performs 32 bit operations competively and STILL has 64 bit capabilites...why WOULDN'T you?

P-X

There is a bigger question to be asked - and as far as I know hasn't been answered yet. The makers claim that UT2003 compiled specifically for the Athlon 64 in 64 bit mode is 20%-30% faster compared to the normal UT2003 without any optimizations for AMD chips. So there are multiple changes: (1) 32 bit to 64 bit and (2) finally code optimized for an AMD chip instead of the larger market share Intel chips. How much of that 20%-30% is due to #1, and how much is due to #2? I feel that most is due to #2. Do you have any evidence otherwise?

Nope, and its a very interesting point. The 20-30% answer was to a question specifically ABOUT the benefits of 64 bit applications. So I would assume that it has an impact or else it wouldn't have been mention. That being said...it IS an assumption on my part.

P-X

 

[splice]

I'm willing to bet that most games, when recompiled for x86-64, will gain in performance.... don't forget, that when going the 64bit mode, you get double the available registers which are 64bits wide or 2x32bits. You can do 32bit int. caching tech. in a 64bit register, reducing the need to access RAM/system-cache. Register space has always been lacking on the X86 platform.

 

[Accord99]

But with 64 bit, you also have to deal with inferior compilers and increased memory requirements, memory traffic and reduced cache effectiveness.

 

[dullard]

Originally posted by: PrinceXizor
Nope, and its a very interesting point. The 20-30% answer was to a question specifically ABOUT the benefits of 64 bit applications. So I would assume that it has an impact or else it wouldn't have been mention. That being said...it IS an assumption on my part.

P-X

Give me one single reason to believe that. They take code that was optimized for Intel chips, and then optimize it for a specific AMD chip. Of course it goes faster. What we need is a comparison of (1) UT2003, (2) UT2003 optimized for the Athlon 64 in 32 bits, and(3) UT2003 optimized for the Athlon 64 in 64 bits. No one has shown me a link or any proof yet showing that 20%-30% was going from (2) to (3). And I'm pretty sure that is since they jumped right from (1) to (3). I'm not saying you are wrong. I'm just saying it is a cloudy area that hasn't been fully explored yet.

 

[rjain]

Originally posted by: Accord99
But with 64 bit, you also have to deal with inferior compilers and increased memory requirements, memory traffic and reduced cache effectiveness.

Er... you'd probably use the same compiler, and the ISA is fundamentally similar, so the amount of tweaking needed to go from athlon to x86-64 is not that huge.

You only need increased memory, etc. if you use larger data structures. The only time you need to increase the data structure size is when you're storing pointers (unless there's a way for an app to request 32 bit addressing with all other x86-64 features available).