Slashdot Mirror
AMD's Athlon-64 Benchmarked With UT2003
Sander Sassen writes "Wondering about the performance of AMD's Athlon-64? Want to how well it runs in 64-bit mode? Hardware Analysis managed to run a few benchmarks on a AMD Athlon-64 demo system using the 64-bit version of Epic' Unreal Tournament 2003. There's also an update with the latest about Athlon-64, Opteron and mobile Athlon-64 including streaming video and pictures of a quad Opteron server."
I saw that and determined that they were more Windows type people and plain flat didn't know that the OpenGL renderer is much weaker (not due to the API, but due to this being much the first cut of the thing...) than the D3D. What they measured was pretty good considering that detail.
I am not merely a "consumer" or a "taxpayer". I am a Citizen of the State of Texas
If one compares the claimed 42fps with other cpu:s, it seems it is at a level of a Celeron 500 MHz...
There is something fishy here as the UT2k3-makers themselves claimed there is a 15% increase in 64-bit mode (on Windows). Normally Quake3Arena for Linux is on par with the Windows version, so it should not be the OS' fault either.
The Mac port? Rumor has it that there was such a thing, but for some reason it disappeared into the ether and was never seen again. ;)
:p
Anyhow, if the game ever comes out for the Mac I'll buy it, but for some reason I think I might die of old age first.
the real benchmarks I'm interested in:
How many *C does the CPU run at?
What size PSU does it need?
I've worked inside a handful of medium-sized machines, including a couple Microway dual Alpha "rugged racks" and a Compaq ES40 Model II. The 8u (or 9?) ES40 is nicely laid out, but removing the motherboard or messing with drive cables is a pain. The 4u "rugged racks" are a disaster of fans and wires. We've had 4 or 5 fans go out on those, and it takes approximately 45 screws (I counted) and about 60 minutes of fast work to remove and replace a midboard fan.
That opteron case, on the other hand, appears to have plenty of cooling that is easy to reach. I don't see any wires permanently attached to the case. It looks very clean and easy to service, except possibly getting the motherboard out.
-Paul Komarek
I recently got into video editing. Until now, I've never needed anything faster than a single CPU system.
Now I understand, completely, what those who do rendering gripe about when it comes to CPU speed never being fast enough.
2:57 of video takes my 1GHz w/ 1GB RAM machine nearly 2 HOURS to render. Just for 3 stinking minutes of video!
The fastest current single CPU would only decrease that number to about 40 minutes, which is still too slow.
A dual CPU solution would bring it down to 20 minutes, but again, if I ever wanted to render even 15 minutes of video, that would be 1 hr 40 mins of CPU time.
And forget doing anything else with the computer while it's rendering. It will start dropping frames like mad, and you have to start over.
Now a 4-way workstation is something that would work. With a 4-way 3GHz Opteron system, I could render in near real-time, and a regular sized MB, if not slightly oversized, could handle 4 procs.
SuperMicro is the only MB mfg. I know of that makes a 4-way board, but it's for Xeons and is insanely expensive ($1800 +/- $100) and that's before you add the overpriced CPUs.
If AMD came out with a moderately prices 4-way workstation, they could get the CAD/CAM, video editing, 3D modeling, rendering and compiling crowd all at once, in addition to the freak gamers and Gotta Have The Best Even Though I'll Never Use It crowds.
The 4-way system is a neglected niche. AMD should fill it.
Talisman
"Study your math, kids. Key to the universe." -The Archangel Gabriel
Well, not directly answering your question, but... x86-64 (the 64 bit architecture that the Athlon64s and Opterons use) is more than just more bits. There are also a lot more registers which will help out code that is recompiled, because programs won't need to do do quite as much moving of values into and out of memory (or cache, I guess). There are other improvements too, but I think the register count is one of the most important ones - with respect to playing games at least. =)
A wider memory buss can help, but it is not really connected to the instructionset or the ability to crunch larger numbers.
One thing that the Opteron has going for it though is the fact that x86-64 have more registers. This makes a real difference. I wonder if the mmx registers are shared with the registers, and if not, why not?
"Considering the fact that this is a 2GHz Athlon-64 processor teamed up with a GeForce Ti 4600 we honestly expected a whole lot better. A 1.6GHz Pentium 4 with that very same GeForce Ti 4600 videocard would have no problems clocking in a similar score while running under Windows XP."
...Which you would expect if you were under the false impression that internal bus bandwidth, addressing mode and clock frequency have considerable impact on a 3D game-quality rendering system.
The graphics hardware does most of the work (ie. the computationally intensive rendering), the CPU is used for game logic, culling and feeding data to the graphics card.
I would say the bottleneck is AGP bandwidth and limited on-board high-speed memory on the graphics card.
Good work on the port(s).
Before the Intel/AMD fanboys go crazy. I wanted to get a few questions in:
1. Can you tell us what specific optimizations you have done/are planning to do for the 64 bit architecture?
2. What optimization benefit do get from a straight "re-compile" of the UT codebase in 64 bit mode?
cheers,
j.
"My cat's breath smells like cat food." - The Tao of Ralph Wiggum.
Anyone that has played the pathetic UT2003 port to Linux will know that it is many times slower than the Windows version. The game was coded for DirectX. It uses a wrapper in Linux to convert Direct3D calls to OpenGL in real-time, and it *REALLY* puts a tax on the host CPU.
In my experience, properly ported OpenGL games on Linux (like RTCW) were faster than under Windows, but UT2003 is definately not the case. For that reason, these comparisons are way too early. I can't speak for the port to the Athlon64 architecture, but when it comes to the 32 bit version of UT2003 for Linux, it's very slow in comparison to the Windows release. This is what happens when you code a game for one platform, one API, and then try to port it to other operating systems.
Did anyone notice that the location of the video, keyboard, mouse, floppy drive, cd-rom,, and power cable are opposite from the pci slots? I have worked with quite a few different rack servers myself in the past, but this sure is one strange looking monster. Not to mention that it really does have some massive heat sinks and the RAM slots appear to be staggered around the motherboard, somewhat like the Sun's motherboards.
Perhaps someone has some insight into other types of rack mounted systems and motherboard configuratioins they can share?