World of Warcraft Performance Guide
by Anand Lal Shimpi on March 23, 2005 12:17 PM EST- Posted in
- GPUs
In PC gaming, first person shooters get all of the attention. The releases of Doom 3 and Half Life 2 were accompanied by a swarm of hardware upgrade guides all over the net, including those published on AnandTech. If you were waiting for Doom 3 or Half Life 2 before upgrading your machine, you had every ounce of information at your disposal upon their release.
The same type of attention is rarely cast on other genres of games in the PC world for a handful of reasons. For starters, FPSes are the most likely to have built-in benchmarking tools, making our ability to present you with performance data infinitely easier. There's also a good deal of emotional attachment to anything that comes out of id Software, Epic Games or Valve, given their history with PC games - in their own way, they are the developers who brought the PC its Super Mario Brothers or Legend of Zelda. But reasoning aside, there's much more to PC gaming than just FPSes; the best, most recent, example of an extremely successful non-fps is none other than Blizzard's foray into the Massively Multiplayer Online Role Playing realm - World of Warcraft.
MMORPGs, like World of Warcraft, rely on hordes of subscribed users (hence the massively multiplayer aspect); otherwise, they lose a big chunk of their appeal. World of Warcraft has been particularly successful in this respect. Earlier this month, Blizzard announced that WoW had reached over 1.5 million subscribers worldwide, with over 800,000 subscribers in North America alone. At any given time, there are over 500,000 users logged into one of Blizzard's many WoW servers at a speed of adoption never before seen in the MMORPG market. But as we saw with the introduction of Warcraft III, anything Warcraft from Blizzard is not only well done, but turns to gold upon release.
Given the tremendous market penetration of WoW, we felt that it was time to take a look at its performance demands. But unlike Doom 3 and Half Life 2, there is no static element of WoW gameplay - everything takes place in an ever-changing online world. The result is that finding a repeatable benchmark to run is fairly difficult...but not impossible. As a MMORPG, World of Warcraft doesn't depend on the razor-sharp reaction time of a fast-paced first person shooter; instead, you spend most of your time walking around performing quests and battling at a much slower pace, in a much larger, more interactive world. As such, there are two scenarios when performance in WoW becomes an issue: when a lot of characters are present on the screen, and simply rotating the camera in the world. The former is a virtually impossible scenario to use as a benchmark, as you can't reliably get a bunch of people to do the exact same thing at the exact same time in a repeatable fashion, but the same can't be said about scenario 2. The world of Warcraft is truly enormous and in order to prevent overcrowding, there is a large number of servers for you to choose on which your character may play - each server has a complete copy of the Warcraft world. Even on the highest populated servers (one of which we conducted our test), there are many areas where you can go that are devoid of any player controlled characters - making them ideal for benchmarking. We chose one such spot in the Night Elf city of Darnassus.
The other aspect of WoW that played into our favor was the fact that exiting and returning to the game using the same character always put you in the world at the exact same place. Equipped with FRAPS, we were able to get an extremely repeatable average frame rate out of a sequence of five full camera rotations in our chosen spot in Darnassus. The number of rotations ensured that minor variations such as a squirrel walking across the screen would have minimal impact on the benchmark, and the results backed that up. We found a maximum of 3% variation between runs as long as there was no disk swapping that occurred during the benchmark (more on that later). The only downside is that our benchmark is not externally reproducible, but similar techniques can be used to achieve a similar end result. We did have one user search us out in the world and interrupt our benchmarking by asking us if we had any linen cloth, but after a quick "no", we were back to benchmarking without being disturbed.
Graphically, WoW isn't another Doom 3 or Half Life 2; it's not pushing the limits of DirectX 9 nor is it going to knock your socks off visually. For a MMORPG, it looks pretty good, similar to Warcraft III, but obviously much more detailed, since you are taking the perspective of a single character in the world of Warcraft, rather than controlling an army from up above.
Blizzard states the minimum GPU requirement as a GeForce 2; while a GeForce 2 will run the game just fine at lower resolutions and detail settings, it will do so at the sacrifice of a good amount of image quality. Below, we have a comparison between a playable WoW system on a GeForce4 MX (GeForce 2 class DirectX 7 hardware) and on a Radeon X800 (DX9 hardware). You'll see a DX7 screenshot, and mouse over to see the same scene running on DX9 hardware:
As you can see, the improvement from older hardware to newer DX9 cards is fairly significant.
The same isn't necessarily true for DirectX 8 hardware, as the GeForce4 looks just as good as the X800, just slower, in WoW.
Below, you'll see a DX8 screenshot, and mouse over the image to see the DX9 version:
How much slower is a GeForce4 than a Radeon X800? That comparison and many more is the subject of our first comparison...
The Test
Our hardware configurations are similar to what we've used in previous comparisons.
AMD Athlon 64 Configuration
Socket-939 Athlon 64 CPUs
2 x 512MB OCZ PC3200 EL Dual Channel DIMMs 2-2-2-10
NVIDIA nForce4 Reference Motherboard and MSI K8N-Neo2 (nForce3) for AGP tests
Intel Pentium 4 Configuration
LGA-775 Intel Pentium 4 and Extreme Edition CPUs
2 x 512MB Crucial DDR-II 533 Dual Channel DIMMs 3-3-3-12
Intel 925XE Motherboard
AMD Athon XP Configuration
Athlon XP 3200+ Barton CPU
2 x 512MB OCZ PC3200 EL Dual Channel DIMMs 2-2-2-10
ASUS nForce2 Motherboard
ATI's Catalyst 5.3 and NVIDIA's ForceWare 71.84 Drivers were used.
Facebook
Twitter
RSS
59 Comments
View All Comments
digit - Thursday, March 24, 2005 - link
#23did you check the motherboard specs on that computer?
when i was looking at them to get one for my girlfreind the motherboard didnt have an agp or pci-x slot for an addon video card.
i could be wrong though...
g33k - Thursday, March 24, 2005 - link
Typo on pg 4. "Thankfully from a performance perspective, the Radeon 9800 Pro behaves very similarly to the X700 Pro (a big slower, but nothing huge)"ppi - Thursday, March 24, 2005 - link
Thanks for the article, I would however have one comment.The test itself does not show it, but in practice the game is quite CPU dependant, especially in areas where it counts - when the game gets crowded.
My resoning behind this is that when I'm increasing or decreasing resolution by one step, performance difference is always minimal in any realistic scenario.
I'm not sure how to test this, though. Maybe if you could stuff a full raid (40 ppl) in some corner of the game. I'm quite sure CPU dependancy would then be MUCH more pronounced.
Hans Maulwurf - Thursday, March 24, 2005 - link
"I'm all about getting a lot of bang for the buck and here's what kind of system I chose to get"-Well, you shouldn´t buy a Dell.
Anand, great review. I had to build a PC for EQ2 some time ago and couldn´t find any benchmarks. This WoW benches would have helped. Thank you!
What I don´t understand is why Blizzard offers no benchmark for WoW. It would be very usefull. I thinks your benches are like the flyby UT-bench (though it´s not your fault - anything else is impossible for you), but a botmatch-like bench would be more interisting. I guess Blizzard could build a good benmark level with ease.
Oh, one more question. Why don´t you write anythink about the command rate (1T or 2T) you used - it´s quite important! And if I remember correctly a tRas of 10 is not optimal for NF4 boards. I think its 7 or so. Possibly there are similar problems with the P4-setup.
segagenesis - Thursday, March 24, 2005 - link
I almost thought that was a downgrade until I read you plan on adding memory/video. Not too bad a deal but I almost feel compelled to ask "why?" when its not significantly different than your previous system. Im still using the same XP 2400+ for ages yet get a consistent 30+ fps just about everywhere (discounting large towns where it will drop to 15, but still... what kind of computer does good in Ironforge when there are 50 chars on screen?). No jittering at all.Mizuchi - Wednesday, March 23, 2005 - link
Great article. This was exactly what I had been wanting to know, but it came 1 day too late, as I've already ordered a system without any recommendations.I am on a two year old 2500+ XP Barton with 512 MB PC2700, 80GB 7200RPM 8MB Cache, Radeon 9700 Pro system. General gaming away from many players is smooth, but going through the Crossroads, Origrammar, Ironforge, and flying on a gryphon/wyvern would be terribly choppy. I would guesstimate where I would run by holding down the movement keys during lag and then wait a second for an update and continue until I did what I had to.
It is also a pain to alt+tab out to use Thottbot, which is why I so rarely use it... (I end up being the one of the newbs to beg in /1 (general chat) for hints).
Along with the new system I've bought, I ordered extra RAM (2x512 Cosair Value Select $93 shipped from Chiefvalue.com) in order to multi-task better. And plan on looking up a good priced 6600 GT card.
I'm all about getting a lot of bang for the buck and here's what kind of system I chose to get:
Dimension 4700
Pentium® 4 Processor 530 with HT Technology (3.00GHz, 800 FSB), Microsoft® Windows® XP Home Edition Qty: 1
Unit Price: $739.00
Processor Pentium® 4 Processor 530 with HT Technology (3.00GHz, 800 FSB)
CP308B
[221-5224]
Memory 256MB DDR2 SDRAM at 400MHz (1x256M)
256M4
[311-3619]
Keyboard Dell Quietkey® Keyboard
QK
[310-1582]
Monitor FREE! 17 inch Ultrasharp™ 1704FPT Digital Flat Panel
1704FP1
[463-8573]
Video Cards Integrated Intel® Graphics Media Accelerator 900
IV
[320-3872]
Internal Hard Drives 80GB Serial ATA Hard Drive (7200RPM)
80S
[341-0883]
Floppy Drive and Memory Keys No Floppy Drive Included
NFD
[340-8688]
Operating System Microsoft® Windows® XP Home Edition
WHXP
[412-0721]
[412-0688]
[463-2282]
[420-4834]
[420-4927]
[420-5099]
Mouse Dell 2-button scroll mouse
SM
[310-1871]
Network Interface Integrated Intel® PRO 10/100 Ethernet
IN
[430-0412]
Modem No Modem Requested
N
[313-3607]
Document Management Adobe® Acrobat® Reader 6.0
AAREAD
[412-0705]
CD or DVD Drives -- Read, Write and Store Data Single Drive: 48x CD-RW Drive 48CDRW
[313-2434]
Sound Card Integrated 5.1 Channel Audio
IS
[313-2758]
Speakers No Speaker Option
N
[313-4514]
Productivity Software Pre-Installed WordPerfect®, Powerful Word Processing
COREL
[412-0701]
Security Software Pre-Installed No Security Subscription
NS2
[412-0754]
Digital Music Dell Jukebox - easy-to-use music player and CD burning software
MMBASE
[412-0741]
Digital Photography Paint Shop™ Pro® Trial plus Photo Album™ Starter Edition
DPS
[412-0521]
Limited Warranty, Services and Support Options 1 Year On-site Basic Plan
B111YOS
[950-1230]
[950-9797]
[412-0360]
[960-2800]
Onsite System Setup No Onsite System Setup
NOINSTL
[900-9987]
Internet Access Services No ISP requested
NISP
[412-0148]
[412-0670]
Mail- In Rebate $150 Mail In Rebate
MIR150
[463-4507]
Miscellaneous Dimension 4700
02DIM47
[463-8574]
Financial Software QuickBooks® Simple Start Special Edition (limited to 50 contacts)
QBSSP
[420-5136]
Purchase Intent Purchase is not intended for resale.
NOT4SEL
[462-4506]
Save $50 on select Dimension™ 4700 desktops through Dell Small Business.
Expires on 2005-03-24 06:30:00 - $50.00
Belkin SurgeMaster 7-Outlet Qty: 1
Unit Price: $17.00
You save 50% on Belkin surge protectors from Dell Small Business!
Expires on 2005-03-24 23:59:59 - $8.50
Free 720 Color Printer (Value $49.00)
Free Dell Color Printer 720 Qty: 1
Unit Price: $0.00
Dell Color Printer 720 Free Dell Color Printer 720
FREE720
[255-0071]
Hardware Support Services 1Yr Ltd. Warranty- Advance Exchange
AE1YR
[902-7040]
[960-4638]
[960-2840]
ADDITIONAL DISCOUNTS AND COUPONS
Small Business customers receive FREE 3-5 day Shipping on select all Software and Peripheral items priced $50 or higher!
Expires on 2007-12-31 06:30:00 - $5.50
Sub-Total $697.50
Shipping Discount -$5.50
Shipping $0.00
Estimated Tax $41.85
Total $739.35
Dell just had a good deal going, considering the $150 MIR ($600 cost). It's like getting the monitor for $300, processor $200, HD $50, Mobo $50 and everything else free (Windows XP Home, Printer, Memory, CD-RW drive, KB/M, software, and Dell's support).
JarredWalton - Wednesday, March 23, 2005 - link
Just a quick FYI: Level of Detail (LOD) has nothing to do with the textures. LOD is a ploygon scaling algorithm, so as you get closer to an object more polygons will be added to make it look more realistic. Done properly, it should be hard to spot. Unfortunately, doing LOD properly is very difficult.With the high-end GPUs, polygon performance generally isn't enough of a problem to make enabling LOD necessary. Lower end CPUs and GPUs can benefit, of course. In the past, I've seen LOD have less than a 10% performance impact, so I'm happier leaving it enabled in most games. (Not that most games actually expose LOD as a tweakable setting....)
ViRGE - Wednesday, March 23, 2005 - link
I wholeheartedly agree with the Mac statements after playing it on a Mac myself. I have a 12" PowerBook Rev. A(867mhz G4, GF4 MX420, 640MB RAM, etc), and the performance I get out of WoW while on the road is abysmal(single digit frame rates when more than a handful of characters are on the screen). Just for comparison's sake, I loaded up WoW on a somewhat similar PC(AXP 2100+, GF2MX original, 768MB DDR, etc), and there's simply no comparison between the two; just eyeballing the FPS has the PC at well over 2x the performance. I even managed to isolate the CPUs in all of this, with the PowerBook almost never hitting 100% CPU utilization in this test(it hovered around 80% or so), meaning the PowerBook should have the edge over the PC, but as I stated before it was losing badly.I have a feeling a lot of this has to do with the fact that the Mac version is using an OpenGL renderer while the PC is using DirectX, but still 50% is insane. It beats not having WoW at all, but there's still some sort of large bottleneck in there, and I'm fairly sure it's all related to the graphics subsystems.
civilgeek - Wednesday, March 23, 2005 - link
I too would like to see a review like this for Everquest II. I don't play WOW and have not seen its graphics first hand but from the looks of the cards\resolutions that are being used in this review... Everquest is putting a lot more stress on the cards. I know that every video card I have seen thus far will crawl to a halt with the graphics turned up to Extreme Quality at 1280x1024. It would be very interesting to see where the bottle necks are and what card would do the best in this scenario.gotsmack - Wednesday, March 23, 2005 - link
When measuring fps, could you also show the most common lowest fps achieved?I want to see what the lowest will be, if I upgrade my hardware.