Wednesday, July 23, 2008

AMD Phenom 9850 Quad-Core

(The previous benchmark report is at NVIDIA GeForce 8800 GT.)

Overall with the old CPU (dual-core Athlon64 X2 5200+) was 10283 3DMarks. With the new 2.5GHz quad-core Phenom, I'm seeing:

Dual 8800 GT in NON-SLI mode:

11458 3DMarks
(4723 SM2 4835 HDR 3719 CPU)

Dual 8800 GT in SLI mode:

12844 DMarks
(4731 SM2 6286 HDR 3718 CPU)

The big gain here is the CPU performance, which boosted from ~2000 up to ~3700. The SM2 number got a boost, even without switching to SLI mode while the HDR test took a hit because I wasn't running in SLI mode.

In SLI mode, numbers for SM2 are higher (4700 vs 4300) and HDR is 6300 vs 5800. So between the Phenom CPU and the faster DDR2 800MHz memory (instead of 667MHz), I'm seeing a fairly decent boost without paying an arm and leg. About what I thought I'd see, so I'm pleased.

Labels: 2008, Benchmarks, NVIDIA

posted by Wuphon's at 10:45 PM (0 comments)

Saturday, March 29, 2008

NVIDIA GeForce 8800 GT

After running with a pair of GeForce 7900 GT PCIe 256MB PCIe cards, I've decided to make the jump to a pair of GeForce 8800 GT 512MB cards. The 7900s just aren't able to run Oblivion well at all, not even at 1280x720 with moderate detail settings. Everything I've heard about the various 8800 series cards is that a single 8800 will beat out a pair of 7900 cards.

Well, we'll see if that's true. According to 3DMark06, I should manage to see about a 2x improvement in FPS from the pair of 8800 GT cards. I've been waiting on this upgrade for 6 months, not sure which 8800 card to purchase. But the key benefits of the 8800 GT series cards are:

- 65nm tech instead of 90nm tech on the GTS/GTX cards
- 110W peak draw vs ~150W peak draw for the 8800 GTS and 8800 GTX
- performs as well as the 9600 GT
- price is very reasonable
- 512MB vs my old 256MB card
- scales well in SLI mode until you run past the video memory limit

So I'll have a card that is roughly 2x powerful, draws only a modest amount of power, and is DX 10 capable.

Note: In terms of performance, it goes GTS (lower) -> GT -> GTX (higher). The GTX cards are very expensive, and the GTS cards are usually less expensive. So the GT cards usually end up as the mid-range card that offers a lot of power for the price.

...

Updates (Apr 2nd): Installed the new 8800 GTs and am working on running 3DMark06. The old numbers were in the 7500-7600 range.

Dual 8800 GT in SLI mode:

10283 3DMarks
(4301 SM2 5801 HDR 2004 CPU)

Oblivion is a bit smoother, but seems to now be CPU-bottlenecked, so I may need to upgrade my Athlon64 X2 5200+ to something faster down the road. Unfortunately, the higher speed X2s are a lot more demanding on the power supply and put out a good bit more heat. The Phenom X4s (quad-core) may be a good choice in about another 3-6 months.

Labels: 2008, Benchmarks, NVIDIA

posted by Wuphon's at 1:09 PM (0 comments)

Wednesday, February 14, 2007

New PC

So I finally upgraded my game box. I was using an Opteron 148, 2GB RAM, NVIDIA GeForce 7800GS AGP. The new system is an Athlon64 X2 5200+, 2GB RAM, and (2) NVIDIA GeForce 7900GT 256MB PCIe cards (running in SLI mode). I spent very little on the upgrade, staying below the "knee" in the price curve where things get expensive quick.

I figured it would give me about a 30% bump in CPU speed (just for the one core, not counting the 2nd core). But according to 3DMark06, the new box is a screamer.

Old: 2918 3DMarks
(1263 SM2 1247 HDR 835 CPU)

New: 7576 3DMarks
(3401 SM2 3280 HDR 1985 CPU)

So I'm very pleasantly surprised for being a mid-range upgrade.

Labels: 2007, Benchmarks, NVIDIA

posted by Wuphon's at 9:32 PM (0 comments)

Wednesday, December 22, 2004

AquaMark3

A few more AquaMark3 scores

GeForce 6800
NVIDIA 66.93 (stock, quality, 128MB aperture)
Asus SK8N Opteron 148 2.2Ghz PC2700 2GB ECC DDR
AquaMark Score: 53166 (CPU: 9047, GFX: 7527)

GeForce FX 5200
NVIDIA 66.93 (stock, quality, 128MB aperture)
Asus SK8V Opteron 144 1.8Ghz PC2100 1GB ECC DDR
AquaMark Score: 7923 (CPU: 7582, GFX: 834)

GeForce FX 5900 XT
NVIDIA 66.93
Pentium 4 1.6Ghz, 1GB PC2100
AquaMark Score: 25992 (CPU: 4044, GFX: 3831)

Labels: Benchmarks

posted by Wuphon's at 10:45 AM (0 comments)

Sunday, November 14, 2004

AquaMark3 NVIDIA GeForce 6800

Toasted my old Ti4600 card this week, the fan finally went out on it and I can't manage to get the old cooler off without destroying the card. So I picked up a new GeForce 6800 for $300 (which was the next reasonable step up from the FX 5900 XT).

NVIDIA GeForce Ti 4600
NVIDIA 61.77 (stock, high performance, 128MB aperture)
A7N8X-Deluxe motherboard AthlonXP 2600+ CPU PC2700 DDR
AquaMark3 Score: 19480 (CPU: 6286, GFX: 2304)

GeForce FX 5900 XT
NVIDIA 61.77 (stock, high performance, 128MB aperture)
Asus SK8V Opteron 144 PC2100 1GB
AquaMark3 Score: 36520 (CPU: 7260, GFX: 4878)

GeForce 6800
NVIDIA 61.77 (stock, high performance, 128MB aperture)
Asus SK8V Opteron 144 PC2100 1GB
AquaMark3 Score: 46982 (CPU: 7008, GFX: 7122)

So, it's about 28%-45% faster then my old FX 5900 XT card. Not a bad bump in speed, and a good deal faster then the old Ti 4600 card. Top scores for NVIDIA cards are in the 55000-60000 range, so it's not that bad of a score.

Labels: Benchmarks, NVIDIA

posted by Wuphon's at 2:05 PM (0 comments)

Thursday, September 09, 2004

QuickPar benchmarks

Created a QuickPar parity set today on the new box. My AthlonXP 2600+ with PC2700 DDR usually manages around 450MB/s and the new Opteron 144 with PC2100 ECC REG DDR is running at 665MB/s. Around 48% faster, even though it has slower memory. Dunno if getting PC3200 RAM for the system would speed that up at all or not.

If I had gone with the Opteron 148 chip, that would probably be around 800MB/s (or 78% faster).

Labels: Benchmarks

posted by Wuphon's at 6:13 PM (0 comments)

Tuesday, August 31, 2004

AquaMark3 Scores (GeForce4 Ti 4600)

AquaMark3

The following are with an AMD AthlonXP 2600+, DDR PC2700 (2x512MB), and a NVIDIA GeForce4 Ti4600 128MB AGP.

NVIDIA 43.45 (stock, best quality, 128MB aperture, A7N8X-Deluxe)
AquaMark Score: 22253 (CPU: 6270, GFX: 2706)

NVIDIA 56.72 (stock, best quality, 128MB aperture, A7N8X-Deluxe)
AquaMark Score: 19090 (CPU: 6253, GFX: 2253)

NVIDIA 61.77 (stock, high performance, 128MB aperture, A7N8X-Deluxe)
AquaMark Score: 19480 (CPU: 6286, GFX: 2304)

I bought a GigaByte FX 5900 XT this week, so I'll be posting new scores at some point. Hopefully, I'll see scores in the 40k range, which will be roughly twice as fast as what I have now. I looked around for a regular FX 5900 card, but couldn't find one in stock. The price in the FX 5900 XT's is only $180-$200, which was right in my target range.

Labels: Benchmarks, NVIDIA

posted by Wuphon's at 11:41 AM (0 comments)

Wednesday, January 07, 2004

PCMark2002 Benchmarking

Older test results:

List of benchmarking software
Old game machine scores
Scores for my RAID5 server
Asus A7N8X w/ Athlon XP 2600+ CPU

So these are the scores from last February... I should see similar scores since I'm re-running the same tests as before. (Asus A7N8X Deluxe motherboard, 1Gb PC2700 DDR, 128Mb nVidia Ti4600 AGP, Athlon XP 2600+)

CPU: 6300
MEM: 4116
HDD: 442

My old Athlon XP 1800+ game machine:

CPU: 4500
MEM: 3000
HDD: 550

Scores from this test:

CPU: 6125 (Athlon XP 2600+)
Memory: 4353 (single-mode)
HDD: 998 (using a single 160Gb ATA/100 5400rpm 2Mb cache)

If I switch the memory from single-mode to dual-channel mode (moved the sticks to slot 1 and 3), I get the following values:

CPU: 6160
Mem: 4351
HDD: 910

Odd that... pretty much identical scores. MemTest86 shows that dual-channel mode is active (1182MB/s vs 734MB/s in the single-channel configuration). But I have seem rumors that the nForce2 chipset doesn't get much of a boost from DDR mode.

Labels: Benchmarks

posted by Wuphon's at 12:07 AM

Sunday, January 04, 2004

Asus A7N8X DDR

Well, that's interesting... the memory stick (B) that I thought was bad survived 9 hours and 34 passes of the default MemTest86 tests when it was the only stick in the computer (in slot 1). The test that failed was configured as 1/B, 3/A (slot 1 had stick B, slot 3 had stick A). I'm now testing stick A (in slot 1) to see if that memory stick is as good as I thought (it worked fine through 2 passes).

According to the manual, to do DDR, you put sticks in slots 1/3 or 2/3 if you want DDR with 2 sticks of memory. I don't know if it means anything, but on my motherboard, slot 1 is made of black plastic while slots 2/3 are made of blue plastic. That might be a clue that it's better to fill slots 2/3 if you have two sticks of RAM.

So I put memory in configuration, 2/B+3/A, and ran MemTest86. It went through two passes with zero errors, so I gave it a boot-up into WinXP. Well, it only took an hour for SC4 to start crashing every 15-30 minutes, so apparently that memory configuration is no good. (The second crash reset the PC.) So now the question is... is one of my memory modules bad and I just haven't found the problem yet? Or is there simply a problem with trying to run this particular brand of memory in DDR mode? (FYI, a link to the Asus approved memory list for the A7N8X motherboard.)

Ran configuration 1/B through 9 hours of MemTest86 without any problems. Going to configure as 1/B+2/A (both sticks on the first memory channel, non-DDR configuration) to see if possibly it's an issue that the memory simply won't work in a two-channel setup. If this works, I'll be happy with the extra memory, but sad that I won't see the 60% performance boost (estimated) from running dual-channel.

Well, setting up both DIMMs in DDR channel 1 didn't work either... so now I'm going to patch the BIOS to the latest (I have the v2 motherboard, but the 1001G revision of the BIOS) which is version 1007. Then I'll fire up SC4 on a large city and see if it runs a few hours without crashing. If that works then I'll put them back into a dual-channel setup and test again.

Update: Well, the box has been up and running a SimCity4 city for almost 24 hour straight without a crash. So the memory is good, but the BIOS needed to be updated. Gonna benchmark the single-channel setup, switch to dual-channel DDR mode and re-benchmark later.

Labels: Benchmarks

posted by Wuphon's at 8:42 AM

Saturday, January 03, 2004

MemTest86 Benchmarks

Just some benchmark values from the machines I've been running MemTest86 on.

First box is a Pentium III 550Mhz with (I believe) PC100 memory (MemTest86 reports it as a 568Mhz CPU). L1 cache (32Kb) speed is 5567 MB/s, L2 cache (128K) speed is 2523 Mb/s, and main memory clocks in at 198 MB/s.

Second box is my Athlon XP 2600+ with GeIL DDR-333 (PC2700) ram in a 2x512Mb configuration. CPU speed is reported as 2088Mhz, L1 cache (128Kb) speed is 12808 Mb/s and main memory speed is 1182 MB/s. This is the box with what I believe was the bad memory stick, so I've got both sticks in and I'm going to run tests overnight. (Found errors after 1 hour, pulled the "good" module and left the suspected "bad" module in. Memory speed is now reported as 734 MB/s since the DDR isn't active. A single test-pass of 512Mb, using the default set of tests, takes around 30-35 minutes on this machine.)

BTW, MemTest86 is a sweet little program. If your machine can boot from CD-ROM (most can if they were made in the last few years), there's an ISO that you can download and create a bootable CD. Pop the CD in, power-up and MemTest86 will automatically start running the default tests. I ended up using an 8cm mini-CD since the ISO is only 16Mb or so. Only disadvantage of the tiny size is that I can't use it in a side-ways mounted CD-ROM drive (no clips to hold it in place, tray has to be horizontal). Also, once MemTest86 starts, you can remove the CD-ROM and go boot it in another PC.

Labels: Benchmarks

posted by Wuphon's at 9:52 PM

Thursday, December 18, 2003

Disk Test Benchmark Results

Benchmarking a server before I upgrade the RAID controller and drop in larger hard drives.

The current RAID controller is built into the motherboard; Promise FastTrak100 Lite (ATA/100) and I have (2) 160Gb 5400rpm 2Mb cache 11ms drives hooked up. While testing with 4Gb worth of data, and averaging the transfer rate over 300 seconds, I get the following results:

18.500 Mb/sec sequential read
5.586 Mb/sec sequential write
4.212 Mb/sec random read
4.892 Mb/sec random write
(no data for mixture tests)

Sequential read scores are likely affected by the 1Gb of RAM installed in the system (which is why I used a 4Gb data set to try and counteract the effect of memory caching). I'm going to switch back to larger 7200rpm drives with 8Mb cache memory next week and see if I can't get better scores.

Sequential access is comprised of rolling through the data set in sequential order, with a 5% chance of randomly seeking to another area of the data set after each 128Kb chunk is read. Random testing is done with an 80% chance of randomly seeking to another part of the data set after each 128Kb chunk. As a result, sequential scores are not quite top-shelf throughput and random scores are not quite worst-case throughput values. Mixture tests are done by setting the system to have an 80% chance of reads vs writes after each chunk (combined with sequential or random settings). Goal of the mixture tests is to give a rough idea of how the system would perform in real-world conditions. Sequential mixture would be similar to a file server while random mixture would be more typical of DBMS access.

Some comparison values:

A 9 month old SCSI-based server (RAID1 drives) is as follows (odd that random write is larger then sequential write, but it does happen on multiple tests). A test of the (3) discs that were RAID5'd resulted in scores that are around 25-33% lower:

56.418 Mb/sec sequential read
9.590 Mb/sec sequential write
18.815 Mb/sec random read
11.345 Mb/sec random write
18.488 Mb/sec sequential mixture
15.045 Mb/sec random mixture

System with a Promise FastTrak66 (ATA/66) 512Mb of RAM and 7200 rpm IBM DeskStar drives 8Mb cache:

18.990 Mb/sec sequential read
9.547 Mb/sec sequential write
4.722 Mb/sec random read
8.013 Mb/sec random write
9.213 Mb/sec sequential mixture
5.913 Mb/sec random mixture

Labels: Benchmarks

posted by Wuphon's at 12:56 AM

Thursday, November 06, 2003

Preliminary USB 2.0 Benchmarks

Write speed to the external USB 2.0 drive seems to be around 1.5 to 1.75 Mb/sec (5.27 to 6.15 Gb per hour), which means roughly 24 hours to fill the 75Gb drive that I have in there. That's a bit slow (should be up around 3.0Mb/sec).

The RAID1 set of 5400rpm drives has a bandwidth of around 6Mb/sec (copying 6Gb of data from one part of the disk to the other gave me 3Mb/sec read, 3Mb/sec write).

I'm going to go dust-off my little performance app that I wrote and see how those numbers compare to earlier baselines.

Labels: Benchmarks

posted by Wuphon's at 5:01 PM

Friday, March 14, 2003

PCMark2002 Benchmarks

Finally got the new server up and running using the Promise SuperTrak SX6000 PCI card. This sweet card lets you chain together up to (6) ATA/100 drives into a RAID5 (or RAID0) configuration, plus uses ECC PC133 memory for cache (I dropped a 128Mb stick in). I took (6) 75Gb drives and put them in a RAID5 configuration with (1) hot-spare drive and now have a server with about 275Gb of user storage. (Don't think I'll be filling that up any time soon, although I have managed to fill 50 or 60 Gb.)

I re-ran the benchmark scores because I was curious as to what kind of performance boost I'd get from the RAID5 array. The CPU/Memory numbers may be a bit skewed as a result of installing a slower 8Mb AGP video card instead of the GeForce2 MX 400 64Mb AGP that was in there. I also had to get rid of a bad stick of memory.

old score
Athlon 1Ghz
768Mb PC133
mirrored 7200rpm IBM DeskStar ATA/100 HDs
CPU: 2500 MEM: 2000 HDD: 575

new score
Athlon 1Ghz
640Mb PC133
raid-5 7200 IBM DeskStar ATA/100 HDs
CPU: 2550 MEM: 1875 HDD: 690

Mmmmm, not bad, the RAID5 system gives me a decent boost over the old RAID1 setup (20%).

Labels: Benchmarks

posted by Wuphon's at 3:40 PM

Friday, February 21, 2003

New Toys

Hooking up my new MB / CPU / RAM in my 2nd game machine... this motherboard has everything (Firewire, USB v2, two LAN ports)... (way too many connectors to get hooked up). Took me about 2 hours to rip out the old motherboard and install the new one (this upgrade cost me $850, but is replacing a 1Ghz AMD which was using 768Mb of PC133 memory). The only thing I was not able to do was switch over to SATA/150 drives (the motherboard comes with a built-in RAID1 SATA/100 controller. I did look for the drives, but all of the SATAs that I could find were backordered roughly a month or more. So I'm going to stick with my tried and true Promise FastTrak66 for the moment and upgrade the drives this fall (since that'll force me to rebuild the O/S again... which I tend to do every 6-9 months).

AthlonXP 2600+
2 x 512Mb PC2700 CL2
Asus A7N8X Deluxe
NVidia Ti4600 GeForce AGP 4x 128Mb
Promise FastTrak66 PCI RAID Card
(2) 40Gb Maxtor HDs

Currently rebuilding the mirror set, then I get to install the O/S, all of the drivers, the O/S updates... then run the burn-in program on it for a day or two. At which point I'll have to rearrange some things on my desk because I'm going to want to switch which PC is where. If I skip the burn-in test, I could be up and logging into EQ (shock!) sometime tomorrow night. The old system's PCMark2002 scores were CPU: 2500 MEM: 2000 HDD: 575 and I'm expecting new scores of roughly CPU: 6000 MEM: 4500 HDD: 600.

So far so good on the burn-in (let it run for 8 hours while I slept). CPU only got as high as 42/43C which is perfect. (I'm using the stock fan that AMD sends in the retail version of the CPU). Currently doing the Windows Update shuffle... then I get to shuffle the office around this morning.

Labels: Benchmarks, NVIDIA

posted by Wuphon's at 8:16 PM

Saturday, February 08, 2003

Benchmarking Software

Currently looking for computer benchmarking software... I want to verify some transfer rates on my various systems, CPU performance, memory performance (don't care about graphical performance since I'm mostly checking servers).

PCMark2002 - Produced by FutureMark, they also make a benchmark program to test laptop performance and battery life, 3D video performance, and a benchmark designed to test simulated performance of business software. There is a limited edition that you can download, a registered version that is US$10, and the Pro version is US$40.

Dr. Hardware 2003 - More of a system information program, diagnostics, plus benchmarking. Haven't really looked at this yet.

SysOpt.com Benchmarking Database - not explored yet.

Performance Test - Has a benchmark utility, plus a burn-in test (seperate) available. Not really explored yet.

Labels: Benchmarks

posted by Wuphon's at 11:44 AM