PDA

View Full Version : Curious Benchmark Results with Geforce 7800GTX vs 2 Geforce 7800GTX in SLI


skimike
07-17-05, 10:15 PM
Hello,

I encountered some curious results in benchmarking that I would not have expected when running FFXIBench3.exe. This is one of the "benchmarking" programs released by Square-Enix for FFXI.

First, my rig's specifications:
AMD Athlon 64 FX-57
ASUS A8N-SLI Premium (1005 BIOS)
2GB OCZ Platinum DDR400 2-5-3-2 (2 x 1GB w/ 1T command rate)
2 WD Raptor 74GB 10K RPM Raptor (RAID 0)
Creative Audigy 2ZS Gamer
Enermax Noisetaker 600W SLI-ready PS
WinXP Pro SP2
nVidia 77.72 drivers

When running the benchmark at "high-res" with 1 BFG Geforce 7800GTX OC with the *.25 BIOS, FFXIBench3.exe scores the system at 8400. All video driver settings are at defaults.

When running the benchmark at "high-res" with 2 BFG Geforce 7800GTX OC's with the *.25 BIOS, FFXIBench3.exe scores the system at 7600. All video drivers were set to defaults except that SLI multi-GPU was enabled.

Why would the score decrease? The only thing I can think of is that the benchmark is far more CPU-intensive than GPU-intensive and the extra time the driver is spending on the CPU figuring out which GPU should do what is dragging the score down. Is that a fair conclusion?

I have an Athlon 64 X2 4800+ on the way as well so I guess I will know more then but I'd appreciate the insight of people that are much brighter than I.

If anyone cares to try and duplicate, the benchmark can be found here: http://dl.square-enix.co.jp/taru/FFXiB3Setup.exe

Thanks.

-=DVS=-
07-17-05, 10:36 PM
Up AA and AF and see if you can get it GPU dependant.

sniggle
07-18-05, 12:28 AM
i got 6214 with high-res on my 6800gt at 370/1000 with athlon64 winchester at 2.4ghz. seems like it could be very cpu-dependent, since i would think that i would get about half of what your single 7800 did if it was entirely gpu dependent. we'll see though as other people try this benchmark...

i must say that this benchmark was really drawn out and got pretty boring after a couple minutes, but maybe that's because i'm not a final fantasy fan. :)

edit: oh and, my drivers are 76.44, quality settings on on "quality", lod clamped.

jAkUp
07-18-05, 12:58 AM
It has to be CPU intensive. Crank up the AA/AF and you should see the SLI setup spank the single card.

SLI adds a little CPU overhead, so that would explain the FPS decrease when enabling SLI.

Zurble
07-18-05, 08:09 AM
did you made a profile for the .exe ? or did you set AFR by default for unknow applications?

try AFR2, instead, and check the load balancing to see what happens.

scott123
07-18-05, 09:28 AM
Reviews of that benchmark indicate it is primarly a CPU limited benchmark.

skimike
07-18-05, 11:22 AM
If it is primarily a CPU-based benchmark (and thus it seems to be as both the benchmark and game seem to experience framerate issues where most any modern-day graphics card would scoff at the required rendering), then it's safe to say that on an FX-57, the nVidia driver will consume approximately 10% of the CPU in determining which portion of the frame to render? Given that the difference in scores was approximately 10% (7600 vs. 8400).

Zurble:

This is my first encounter with SLI so I wound up choosing whatever the default method is. I'm assuming the default method is where each card renders a portion of the frame determined dynamically by the driver. I'm led to this conclusion because when I check the option in the driver to show what GPU is doing what (forgive me but I forget what it's called - I'm at work and not in front of the PC at the moment), I see two green bars, one vertical and one horizontal. The horizontal bar rises and falls, presumably based upon how much of the frame each GPU is responsible for. It certainly sounds reasonable that alternate frame rendering would be faster since the driver no longer has to figure out which GPU needs to render what - it just round robins amongst the cards, correct? As I understand it, a select few games have a problem with this mode due to how transparencies are calculated - something about needing to store previous and successive frames in the frame buffer at once? Anyway, I will mess around with it this evening. Even if it is a CPU benchmark, it'll be interesting to see what the difference is in the driver's impact on the CPU between the two modes, assuming I can figure out how to switch it to the other mode. :)

I may be making a bad assumption here but it seems reasonable that AFR would be faster on a single-core system. In a single-threaded game which would allow the nVidia driver to execute freely on the other core, is there likely to be much of a difference between the two modes of rendering? I will know for sure once that 4800+ arrives but just wondering in advance.

Thanks for all the replies! I originally tried bumping AA and Anisotropic filtering up to their maximum levels on my other system with a Geforce 6800 Ultra and the game refused to run. I had to tune them down to about half their maximum settings to get the game to run at all and, when it did run, it was clear based upon what I was seeing on the screen and from screenshots taken with FRAPS that they were not having any effect.

Thanks again!

Zurble
07-18-05, 11:41 AM
if you see the green thing, then u are in SLI mode, but i think by default everything is set to be run on Single GPU except .exe with profiles. And i didn't saw a profile for the Final Fantasy exe. (do u see a profile for your exe in the list? what is the default/global setting for SLI: single/multi GPU?)

not to mention that if u use coolbits (set it to 8, or B if u also want overclocking) u can select wich type of SLI u want: SFR, AFR, AFR2

AFR2 is used when u have trouble with AFR, using some compatibility bits, it seems to work better in many unprofiled .exe then AFR

skimike
07-18-05, 12:38 PM
I set the global settings to multi-GPU and the green bar does seem to fluctuate with what's being rendered so I think it's definitely working in SFR. I haven't messed with coolbits yet but I did stumble across nHancer which I am going to play around with this evening, as well. Thanks!

OWA
07-18-05, 03:58 PM
Hello,

I encountered some curious results in benchmarking that I would not have expected when running FFXIBench3.exe. This is one of the "benchmarking" programs released by Square-Enix for FFXI.

First, my rig's specifications:
AMD Athlon 64 FX-57
ASUS A8N-SLI Premium (1005 BIOS)
2GB OCZ Platinum DDR400 2-5-3-2 (2 x 1GB w/ 1T command rate)
2 WD Raptor 74GB 10K RPM Raptor (RAID 0)
Creative Audigy 2ZS Gamer
Enermax Noisetaker 600W SLI-ready PS
WinXP Pro SP2
nVidia 77.72 drivers

When running the benchmark at "high-res" with 1 BFG Geforce 7800GTX OC with the *.25 BIOS, FFXIBench3.exe scores the system at 8400. All video driver settings are at defaults.

When running the benchmark at "high-res" with 2 BFG Geforce 7800GTX OC's with the *.25 BIOS, FFXIBench3.exe scores the system at 7600. All video drivers were set to defaults except that SLI multi-GPU was enabled.

Why would the score decrease? The only thing I can think of is that the benchmark is far more CPU-intensive than GPU-intensive and the extra time the driver is spending on the CPU figuring out which GPU should do what is dragging the score down. Is that a fair conclusion?

I have an Athlon 64 X2 4800+ on the way as well so I guess I will know more then but I'd appreciate the insight of people that are much brighter than I.

If anyone cares to try and duplicate, the benchmark can be found here: http://dl.square-enix.co.jp/taru/FFXiB3Setup.exe

Thanks.
Any tips on getting that setup stable or did everything just work for you? I'm close to dumping my setup b/c I've been trying for a couple of weeks and can't get things stable. Our systems are pretty close so I'm hoping you'll know something I don't in terms of getting things working.

Absolution
07-18-05, 05:13 PM
ffxi is almost very cpu dependent and it doesnt support AF/AA (its a pretty bad engine imo)

this problem is in various games with sli

Ronin
07-18-05, 08:03 PM
After spending extensive time with AM3 this weekend, I can absolutely verify that with standard settings, SLi with these cards is CPU intensive, and running the cards in SLi at the same speeds as a single card, while showing a benefit, will not yield the same results as you'd expect.

skimike
07-18-05, 09:12 PM
Any tips on getting that setup stable or did everything just work for you? I'm close to dumping my setup b/c I've been trying for a couple of weeks and can't get things stable. Our systems are pretty close so I'm hoping you'll know something I don't in terms of getting things working.

I don't have the 4800+ in my hands yet - I guess I'll be part of the second wave of people to get theirs but with the FX-57, everything went pretty smooth. I can give you the rundown of my settings/slots.

IRQ's:
ISA 0 - System timer
ISA 1 - Keyboard
ISA 6 - Floppy
ISA 8 - CMOS/RTC
ISA 9 - ACPI
ISA 12 - PS/2 Mouse
ISA 13 - Numeric data processor
PCI 11 - NVIDIA nForce PCI System Management
PCI 14 - NVIDIA nForce4 PATA Controller
PCI 18 - Creative SB Audigy 2 ZS (WDM)
PCI 18 - NVIDIA GeForce 7800 GTX
PCI 19 - IEEE 1394 Controller
PCI 21 - NVIDIA Network Bus Enumerator
PCI 23 - NVIDIA nForce4 ADMA Controller

(only have 1 of the 7800GTX's in my system at the moment - I'll install the second card after I get some more benchmarking done and update the IRQ list with the second card installed later).

System temps according to nVidia Monitor are (at idle):
CPU: 38C
System: 36C
GPU: 48C

I am using the stock HSF cooling solution (at 5-6K RPMs when the system is really loaded, the FX-57 fan is deafening but it keeps it cool). I'd like to look at a water cooling solution for the X2 4800+ but that will be my first attempt at doing so, so if anyone has any suggestions for a water-cooling n00b as to an easy and efficient system to install, I'm all ears. I think I need a cooling kit that won't interfere with the heatpipe on the ASUS A8N-SLI Premium motherboard so the only thing I really want to cool at this point would be the CPU - I'll experiment with the GPUs once I've gained some confidence.

I, like most everyone else, stuck the Audigy 2ZS Gamer in the 3rd PCI slot. When only using 1 video card, I populated the top PCI-express slot. With both cards in, both slots are populated, obviously. :)

I am using the 1005 BIOS revision and here are my settings:

Main:
Legacy Diskette A: - 1.44M, 3.5 in.
Primary Master IDE - Sony DVD-ROM DDU
All others are "None"
HDD SMART Monitoring - Disabled
Installed Memory - 2048MB

Advanced - CPU Configuration:
CPU Type - AMD Athlon(tm) 64 FX-57 Processor
CPU SPeed - 2800MHz
Cache RAM - 1024K
Hyper Transport Frequency - Auto
AMD K8 Cool'n'Quiet control - Enabled

Advanced - CPU Configuration - DRAM Configuration:
Timing Mode - Manual
Memclock - 400MHz
CAS# latency (Tcl) - 2
Min RAS# active time (Tras) - 5T
RAS# to CAS# delay (Trcd) - 3T
Row precharge Time (Trp) - 2T
(I have not tried tuning Trc, Trfc, Trwt, or Twr so any suggestions for performance are welcome)
Row cycle time (Trc) - 11T
Row refresh cyc time (Trfc) - 24T
Read-to-Write time (Trwt) - 4T
Write Recovery time (Twr) - 3T
1T/2T Memory Timing - 1T
S/W DRAM OVer 4G Remapping - Disabled
H/W DRAM OVere 4G Remapping - Disabled

Advanced - PCIPnP:
Plug&Play OS - No
Init Display First - PCIEx
Resources Controlled By - Auto

Advanced - Onboard Device Configuration:
Onboard NV Lan - Enabled
Onboard LAN Boot ROM - Disabled
Onboard Marvell LAN - Disabled
AC97 Audio - Disabled
PCI IEEE 1394a - Disabled
Silicon SATA Controller - Disabled
Serial Port1 Address - Disabled
Parallel Port Address - Disabled
Game Port Address - Disabled
Midi Port Address - Disabled

Advanced - Onboard Device Configuration - IDE Function Setup:
OnChip IDE Channel0 - Enabled
OnChip IDE Channel1 - Disabled
IDE DMA transfer access - Enabled
SATA Port 1, 2 - Enabled
SATA DMA transfer - Enabled
SATA Port 3,4 - Disabled
IDE Prefetch Mode - Enabled

Advanced - Onboard Device Configuration - NVRAID Configuration:
RAID Enable - Enabled
IDE Primary Master RAID - Disabled
IDE Primary Slave RAID - Disabled
Dirst SATA Master RAID - Enabled
Second SATA Master RAID - Enabled

Advanced - Onboard Device Configuration - USB Configuration:
USB Controller - Disabled

Advanced - SLI Configuration:
EZ-Plug Warning - Enabled
SLI Broadcast Aperature - Auto
AI-Selector - Auto

Advanced - JumperFree Configuration:
Overclock Profile - Auto

Advanced - LAN Cable Status:
POST Check LAN Cable - Disabled

Advanced - PEG Link Mode:
PEG Link Mode - Auto
PEG Root Control - Auto
PEG Buffer Length - Auto

Advanced - Speech Configuration:
Speech IC Reporter - Disabled

Advanced - Instant Music:
Instant Music - Disabled

Power:
ACPI Suspend Type - S1&S3
ACPI APIC Support - Enabled

Power - APM Configuration:
Restore on AC Power Loss - Disabled
PWR Button < 4 secs - Instant-Off
Power on by PCI Devices - Disabled
Power on by External Modems - Disabled
Power on by RTC Alarm - Disabled
Power on by PS/2 Mouse - Disabled
Power on by PS/2 Keyboard - Disabled

Power - Hardware Monitor:
Q-Fan Controller - Disabled (I enabled this once and when I cold booted, the system complained that the CPU fan speed was too slow)
Vcore Voltage - 1.39V
3.3V Voltage - 3.32V
5V Voltage - 5.02V
12V Voltage - 11.90V
No fluctuation in any of the voltages - they're all rock steady. I have an Enermax Noisetaker 600W SLI-ready PS, model EG701AX-VE(W) SFMA(24P) connected to a CyberPower 1500AVR-HO UPS connected to its own dedicated 20amp circuit.

CPU Fan Speed warning - 1200 RPM
CHA1 Fan Speed warning - Disabled

I find it odd that the hardware monitor section of the BIOS doesn't report the power supply fan speed - it doesn't even have a spot for it. I also have a 3-pin chassis fan connected to the CHA1 connector and it doesn't report, oddly enough. Might be the fan itself - it came with the Lian-Li case I am using.

Boot - Boot Device Priority:
1st Boot Device - CDROM
2nd Boot Device - Hard Disk
3rd Boot Device - Removable
4th Boot Device - Disabled

Boot - Removable Drives:
1 - Floppy Disks

Boot - Hard Disk Drives:
1. SCSI-0 : NVIDIA STRIPE 138.49G
2. Bootable Add-in Cards

Boot - CDROM Drives:
1. 1st Master: Sony DVD-ROM DDU1615

Boot - Boot Settings Configuration:
Case Open Warning - Disabled
Quick Boot - Disabled
Boot Up Floppy Seek - Disabled
Bootup Num-Lock - On
Typematic Rate Setting - Disabled
OS Select for DRAM > 64MB - Non-OS2
Full Screen LOGO - Disabled
Halt On - All, But Keyboard

I leave the ASUS EZ-Plug connected regardless of whether I have both video cards installed or not.

I have memory slots DIMM_A1 and DIMM_B1 populated with OCZ Dual 2G (1Gx2) DDR400 RT 2-3-2-5 Platinum memory (Part #20-227-210 from newegg.com).

I haven't really done anything special. It mostly just worked out of the box. I did tighten the memory timings down too tight once which required me to pop the CMOS battery and move the jumper for 15 seconds but beyond that, I have not experienced a problem. Unless you count that the Cool&Quiet functionality doesn't seem to function. According to the ASUS Cool&Quiet utility, the CPU never seems to clock down at all despite having followed the instructions in the manual about setting power management to minimal. I'll bet if I slap Linux on it and cat /proc/cpuinfo, it'll show me variable speeds depending upon load, though.

Once monarch ships my X2, I'll have 2 identical systems, one with the FX-57 and one with the 4800+ to test on.

Hope that helps. Good luck!

And I agree with you, Absolution. The engine does suck. :) And it will probably always continue to suck because SE doesn't want the PC experience to outshine the PS2 or Xbox 360 experience. About the only thing that can be done to dress the game up is to modify the values for background rendering in the Windows registry. I had been running with a background resolution of 2048x2048 and an overlay resolution of 1600x1200 (I guess this is the poor man's super-sampling method :p ) which made the Geforce 6800 Ultra work a bit. Since I got the 7800GTXs, I've been running in 2048x1536 on a Sony Trinitron 21" monitor. I can't make one card work hard, let alone 2, which is why I started looking at the FFXIBench3 benchmark in the first place and that's when I noticed the odd "drop" in performance in the benchmark score. The "drop" isn't discernable in the game, however, and I suspect it's all theoretical anyway.

Again, thank you for all the help!

vläd
07-18-05, 09:31 PM
I dunno if this will help any or not, but ASUS web site (http://support.asus.com/cpusupport/cpusupport.aspx?SLanguage=en-us&model=A8N-SLI%20Deluxe) indicates that you shoyld be running the 1011 BIOS w/ that CPU.

Opps! My bad! Misread the MB as an SLI Deluxe. :/

skimike
07-18-05, 10:36 PM
The 1011 BIOS is the latest non-beta BIOS for the A8N-SLI Deluxe. The A8N-SLI Premium is a slightly different board with the latest non-beta BIOS being 1005. :)

OWA
07-19-05, 02:03 PM
I don't have the 4800+ in my hands yet - I guess I'll be part of the second wave of people to get theirs but with the FX-57, everything went pretty smooth. I can give you the rundown of my settings/slots.


Thanks for taking the time to list all your settings.

oC|-TiTaN
07-19-05, 10:35 PM
First, my rig's specifications:
AMD Athlon 64 FX-57
ASUS A8N-SLI Premium (1005 BIOS)
2GB OCZ Platinum DDR400 2-5-3-2 (2 x 1GB w/ 1T command rate)
2 WD Raptor 74GB 10K RPM Raptor (RAID 0)
Creative Audigy 2ZS Gamer
Enermax Noisetaker 600W SLI-ready PS
WinXP Pro SP2
nVidia 77.72 drivers

Holy crap! You have my system nearly to a T...

Mine:
FX-57
2x BFG 7800GTXOC
2GB OCZ 2-3-2-5-1T
A8N-SLI Premium
2 x 74GB Raptor
Audigy 2 ZS Gamer
PC Power and Cooling 510W SLI
WinXP Home SP2/ WinXP x64 Dual Boot
nVidia 77.72 drivers