EVGA GTX-260 Core 216 FTW - Review

[Ali Tayyab]

1. INTRODUCTION

Since my last review (linked here for shameless increase in read count!) I have gone through no less than three cards (the card reviewed, its one gig variant, and now this Nvidia offering). As I said, I have a fetish for computer gizmos and usually cannot resist getting my hands on something my eyes take a fancy to!

This review will take you through why Nvidia had to take decisions it didn’t like, reduce prices when it didn’t want to and introduce a new card when it thought it wont have to. It will also provide you with all the necessary benchmarks (including those from GTA IV).

2. THE (TROUBLED) HISTORY OF THE GT200.

When Nvidia came out with 8800GTX/GTS, it surprised many. No one was expecting Nvidia to come out with a graphics-processing unit (GPU) with unified shader architecture. In fact, most interviews with Nvidia PR etc prior to the release of the card hinted that Nvidia was not going the unified way for this generation.

This, probably coupled with the fact that ATi had already one unified architecture GPU under its belt (the GPU inside the Xbox 360), led many to believe that ATi would probably have the upper hand in that particular round of GPU wars.

However, Nvidia did what no one expected it to do and did it really well. The 8800GTX was fast. It came at a premium price (but the fact that many did not note was that its little brethren 8800GTS was 30% slower while being 25% cheaper). For many generations, in fact until the launch of the GT200, the 8800GTX was one of the fastest cards around. In an industry where architectures are revised every six months and a new architecture comes out in about 12 to 18 months time this is an amazing achievement.

The expectations with the GT200 were so high (more than 2 times the performance of the fastest G80, free i.e. no frame rate deprivation 4x AA) that when the first cards hit the market at prices which were never heard of before (maybe with the exception of the voodoo 5 6000 and the notorious GX2s) people literally scratched their heads. The performance was not earth shattering. The humongous die needed a heat-sink shroud that covered both sides of the PCB (which definitely looked cool). Part of the reason was Nvidia’s philosophy that bigger is better coupled with the 65nm process used to manufacture the chip.

When ATi came out with the HD4870 the world took notice. Out of nowhere, this $299 card was challenging Nvidia’s $650 offering and often beating its $400 card.

The Microsoft Curse

I will digress a little here. I call it the Microsoft curse (I am thinking of trade marking this term, as I have never seen it used). The “FX” failed, as Nvidia had to deliver the NVx to Microsoft for the Xbox. The “FX” series was late, and was not up to scratch.

Likewise, ATi did the same for Xbox 360. The “2” and later the “3” series were no match for the “8” and “9” series Nvidia cards. In fact the broken Anti-aliasing engine of the “2” series cards meant that in effect using any level of AA made the frame rate nose dive. This was less prevalent in the “3” series; the engine was not (totally) broken anymore; it was not just very good.

This time Nvidia did not have the curse to blame. It was not working for Microsoft. It was gloating in its arrogance and (nearly) let ATi slip by. It did however deliver a chip called "RSX" to Sony for the PS3. However, this was not something Nvidia spent resources (engineers or money) on. This was just a G70, the chip that powered the "7" series cards. In fact, the chip was nearly identical to the one used in Nvidia's 7800GTX card. Thus, Nvidia did not really have an excuse to falter.

Literally turning green with envy, Nvidia (which ironically has green as its logo color), had to cut prices and introduce a completely new card: The GTX-260-Core 216.

It needed to do this not only to prevent customers from choosing its opponents offering, Nvidia did this to prevent its most “loved” partners from going the other way. (XFX recently officially stated that it would release cards based on ATi technology. Innovision (Inno3D brand) had already done so creating a line of cards under the moniker of “Force 3D”)

3. THE GT-200 ARCHITECTURE

Disclaimer: I am not a chip designer. My curiosity makes me dig deep into learning how things tick. Much like children like to break things to find out what they are made up of (akin to the people who use LHC to collide atoms to see what they are made of), I like to find out how things work. If I get stuff wrong, please point it out.

If you read the review of the ATi 4870, you will see how these modern GPUs are made up of identical smaller building blocks, much akin to Lego. Simply bunching up these units into larger sets gets you a very powerful chip. Of course this is an over simplification but it does help get the point across.

THE STREAM PROCESSOR (SINGLE PROCESSING CORE):

The basic building block is called a “Stream Processor”. Nvidia likes to call this a “Single Processing Core, so we’ll just call it that a “SPC”.

An “SPC” is made up of a single Floating Point Unit “FPU” and two Arithmetic Logic Units or “ALUs”. This “SPC” processes both Vertex and Pixel data (thus the unified architecture). You feed it loads of mathematical data and it will churn out more mathematical data for you. As it has a FPU it can do floating Operations (think fractions) and also work with Non-fractional numbers. As it has no cache, it cannot hold onto to any data. It must pass on its calculation before it can deal with the next set of numbers. Figure #1 shows what a typical SPC is made up of. (Please understand that this is a very basic presentation of what essentially is a SPC. It is by no means an accurate diagram of the unit)

THE STREAMING MULTIPROCESSOR:

Put eight of those “SPCs” together with some more processing logic and you get what is called a “Streaming Multiprocessor” or “SM”. Basically this refers to multiple single processing cores bunched together (thus “Multi”). Apart from these 8 SPCs the Streaming Multiprocessor also has some cache and a special unit called a “Special Function Unit” or SFU. There are 2 of these in a SM. An SFU is just a fancy name for a Floating Point Unit. These units specialize in Trigonometry functions for Anisotropic filtering etc.

THE TEXTURE PROCESSOR/ CLUSTER:

Keep on adding SMs till you have 3 of em and you get a “Texture Processor/ Cluster” or TPC. The TPC adds one Texture Unit and more control logic.
In effect a single TPC is made up of 8x3 or 24 SPCs and 2x3 SFUs. The G80 that powered the 8800GTX had 16 SPCs and 4 SFUs and thus was made up of 2 SMs.

THE STREAM PROCESSOR ARRAY:

Finally! Add 10 TPCs and you get the final stream engine of the GT200. Called a Stream Processor Array or “SPA”. This is 2 up from the 8 that the G80 had.
As 1 TPC has 24 SPCs, 10 TPCs would have 240 SPCs, thus the magic number in the GTX-280. Get rid of 2 of these units and end up with 192 SPCs, the number in GTX-260. Get rid of only 1 and you end up with the Core 216.

Briefly:
Basic Unit = Single Processor Core
Stream MultiProcessor = 8 SPCs + 2 SFU
Texture Processor = 3x SM (or 3x8 SPCs)
Stream Array = 10xTPC (or 24x10 SPCs)

As you can see the stream processors have gone up from 128 in the 8800GTX to about 240 in the GTX-280 (88% increase).

So now that we are done with the stream array we can move on to the Backend or Raster Operators. The GT200 has 8 ROPs, 2 up from the 6 that the 8800GTX had. Render Backends (RBEs) or, as Nvidia calls them, the Raster Operation Units (ROPs) handle the final graphical output.

Of course the Shader engine and the Render back ends form a part of the chip. There is memory control circuitry, the circuitry that feeds the shader array etc. Most of this is still based on the G80 with tweaks. The memory bus is now 512 bits (or 448bits in case of GTX-260) wide.

NEW POWER MANAGEMENT FEATURES

With a brand new architecture come new power saving features. The GT200 is not only able to control Clocks and Voltage it, it is also able to turn off components unit by unit. Nvidia's figures show that a G80 (powers the "8" series) at idle consumes about 80 Watts of Power while the G92 (powers the "9" series) consumes 45 Watts. The GT200 at idle consumes only 25 Watts. Not too shabby, eh?

But as with all good things there is a catch here as well. Many have found that, say for example while playing fall out 3, this power saving feature induces stuttering in games. This is down to the card clocking up from a lower state as the load on it begins to increase. With newer drivers this has been brought down to a minimum but it still affects many.

Unfortunately as yet there is no way to turn this "feature" off. Personally i've never really noticed this to an extent that it mars my playing experience. I did get some odd stutters in Prince of Persia. There was no Hard disk activity at the time that they occured, usually after a cut scene had played out.

Here is a Screen shot of EVGA Precision Overclocking utility which shows how the GPU changes its clock in reference to the game (Prince of Persia). More on this nifty tool in a little bit.

As you can see there are a few clock changes. The Changes within the game range between the maximum (670 Mhz) to 400 Mhz on the core, before settling down to ~300 Mhz which is idle Core clock speed. This change is reflected in RAM and Shader clocks as well.

GIMME MY PHYSICS!

Everyone probably knows that Nvidia purchased Ageia sometime ago. The idea was to accelerate Phys-X using Nvidia's stream processor. For those who totally missed what Phys-X is: Phys-X is "middle-ware" that works on dedicated hardware to accelerate Physics effects in a game. Good examples of physics effects include:
1. Smoke particle effects
2. Fluid motion effects
3. Cloth effects (like for example the Neurien demo)

Phys-X acceleration is available on many games, but there is no "killer" app that would make everyone go out and buy a Nvidia card to acclerate Phys-X.

Phys-X is not the only way to accelerate physics effects. Havok engine, which is more popular, does the same and runs off your CPU; no dedicated hardware needed. However the "density" and "realism" produced by the two engines differ quiet a lot. As Havok is now owned by Intel, it remains to be seen if Intel will accelerate Havok using its own Larabee GPU. (That should be possible as Larabee itself is based on x86 architecture off which Havok runs). A version of Havok, called HavokFX that could off load the calculations to a GPU (both ATi and Nvidia) was under development but has now been canceled.

Phys-X runs off any G80 class processor and beyond. Obviously the GT200 runs this better than its predecessors. Nvidia includes several basic examples of Phys-X effects with its drivers. There are dedicated Phys-X demos as well.

If you dont have Nvidia hardware, you can use this link to look at Phys-X effect videos:

YouTube - Nurien-Phys-X

Is Phys-X relevant now? Not really. Will Phys-X be relevant a year down the line? Maybe. Recently Nvidia got several developers onboard to support Phys-X. this includes EA, 2K Games and THQ. Mirror's Edge PC interation will have Phys-X acceleration as well. Click on the Video below to see Phys-X effects in the game. [Courtesy EA & Nvidia]

YouTube - ME-PhysX

The biggest question, however, is 'what will Microsoft do?' There will be some form of hardware assisted physics acceleration included with DirectX 11. Microsoft might license Phys-X from Nvidia (highly unlikely but possible), just as it licensed S3TC (a form of texture compression) from now defunct S3. Personally I wont buy an Nvidia card only for Phys-X just yet.

While we are talking about Phys-X,let me tell you that it is possible to use a secondary Nvidia card to offload Phys-X acceleration. Using a GTX-280 in combination with a 9600GT it is possible to offload Phys-X calculations to the less powerful card. The advantage here is that this will work on non SLI motherboards (read Cross-Fire ready motherboards). Phys-X also works with an SLI setup.

I DO AUDIO TOO!

Recent ATi GPUs have a dedicated sound engine on-board, allowing audio to work through the HDMI port. Nvidia also allows this albeit with a little difference. There is a Little S/PDIF connector under plastic cover next to the power connector. A pass through cable allows one to pass through audio when using the HDMI adapter. Thus there is no actual "hardware" to decode audio. The card only Passes-through audio to the HDMI port.

4. INTRODUCING THE EVGA GTX-260 CORE 216 FOR THE WIN EIDTION

EVGA is an Nvidia premier partner. It sells the largest number of discrete GPUs in United States based on Nvidia architecture. Two things set EVGA apart from most other discrete GPU manufacturers:

1. Life-time Warranty that is NOT voided by replacing the cooling solutions (As long as you keep the original and put it back on when doing an RMA)

2. A step up program. If you buy Card “A” and within 90 days a better card comes along and enters the step up program, you can return the card in good working condition, pay the difference and you get a brand new card. Provided your card also qualifies for the step-up program (as indicated on box )

Unfortunately this means nothing if you buy the card here as you don’t get either. Though if you do get the card from USA and register it to a USA address and do all the RMA./ Step-up business from there it is possible to avail all these features.

I am not sure what happens if you buy a card from galaxy and register (with EVGA) within 90 days to avail lifetime warranty.

Anyways the card was in direct response to the unthinkable (at least to Nvidia) performance numbers that the ATi’s 4870 was churning out. It is basically a GT200 that passes with only one TPC failing. Instead of 10 working TPCs, this one has 9, one up from the basic GTX-260.

It comes in a pretty standard affair box, nothing snazy, not as big as the Asus graphic card boxes, and not as small as the one for HD-4870. Though the latter did come with a nifty multi purpose tool.

The card itself looks exactly like the GTX-260 (plain vanilla edition). This one is based on the 65nm process. A newer GTX-260-Core 216 based on the 55nm process is now also available. It is pretty easy to tell the two apart. The latter has the back of its PCB bare as compared to the former which has its back shrouded in plastic. This is to get some airflow over the ram chips that are on the back of the PCB. The newer model allegedly has RAM chips only on the front, thus doesn’t need an elaborate cooling solution. Though those who have seen the 55nm part dont like the "shrunken" Cooler. The SLI connectors are under a plastic cover at the read of the card. This, being a "GTX" model, is TRI-SLI ready.

Whats in the box?

Other than the usual mix of cables, DVI to VGA converter and the Driver’s disc the card doesn’t come with a game bundle. At least that is what I thought till I registered the card!

It turns out that everyone who purchases a EVGA card and registers it, gets a 3Dmark Vantage Key FREE (yes I got mine ).

Not only this, you also get Farcry 2 for free if one wasn't shipped with your card.

The (For The Win) FTW edition is factory over clocked. The table compares the various iterations of the new Nvidia GPU series.

Riva Goes EVGA -The EVGA Precision Overclocking Utility

What happens when Riva Tuner meets EVGA? You get the Precision Overclocking Utility. Riva Tuner's developer, who goes by the alias of 'Unwinder', modified the interface so that it is easier to use. With the basic view you get all your monitoring info and the ability to modify Core, Shader and Memory clocks together with fan speed. The various features (as listed on EVGAs) website include:

*Up to 10 profiles, and ability to assign hotkeys to these profiles to allow in-game
*Ability to view temperatures in the system tray
*Core/Shader Clock Link/Unlink capability
*Fan Speed manual or auto adjustment
*Real-Time on-screen display support for ingame temp, framerate, and clock monitoring
*Logitech Keyboard LCD Display support
*In-app temp/clockspeed monitoring of GPU's (up to 4 supported)
*Information button shows device, driver, memory size, BIOS Revision and SLI mode
*Ability to choose different skins

The interface is doozy to use. It allows one to test the frequencies one sets along with fan speed. It also allows one to create profiles if one so desires. The graphic (temperature and clock speeds) display is also pretty neat. It opens as a part of the display, not as a separate window (as is the case with Riva tuner).


Here is a GPU-Z shot of the Hardware Spec:

A Little Note:
All Nvidia cards from the G80 generation and beyond have two different clocks for the GPU. One is called the Core clock and the other the Shader Clock. The "Stream Processor Array" runs at a higher frequency than the rest of the circuitry. The former is called the "Shader" clock, the latter the "Core" clock.

ATi runs the entire GPU at one clock speed, the "Core" speed.

Finally here is a tabular comparison as well as the closest local prices I could find (correct as of 28th December 2008)

*From Galaxy: URL: Galaxy .... Pakistan's Leading Computer Store (Home)
**There is no FTW edition locally available. A Super-clocked version is available for the listed price.

5. BENCHMARKS!

TEST BED:

CPU: Intel Q6600 (at 3.0Ghz)

MOBO:
Asus P5-b Deluxe [For all cards other than the GTX-260-Core 216-FTW]
Asus P5-q-E [For the GTX-260-Core 216-FTW]

RAM:
2 gigs ddr2-667

GPUs:
1. ATI HD4870-512 MB
2. ATI HD4870-1 GB
3. NVIDIA 8800 GTS-640
4. NVIDIA 8800 GTX
5. NVIDIA GTX-260-CORE-216-FTW

PSU:
Corsair HX620

OS:
Windows XP SP3

Driver:
Catalyst for ATi
Nvidia FW drivers

*As the cards were tested at different times, various driver revisions were used to test them out. Rest assured that the GTX-260-Core 216 was tested with the latest (and greatest) 180 series (aka Big Bang II) drivers

BENCHMARK SOFTWARE (Using Inbuilt Benchmarks)
1. 3DMark06
2. Crysis
3. World in Conflict
4. F.E.A.R.
5. GTA IV

The graphs this time are different. They are color coded for better readability.
ATi Card Tested using 1024*768 Resolution
ATi Card Tested using 1024*1280 Resolution
Nvidia Card Tested using 1024*768 Resolution
Nvidia Card Tested using 1024*1280 Resolution

This applies to all except the 3D-Mark Graph which was run at 1024*1280 resolution, and GTA IV.

1. 3D-Mark06

It isn’t a game, but it is still one of the most popular tools to benchmark video cards.
All the cards were tested using the “standard” 3D-Mark-06 test.

The GTX-260-216 is (a little) ahead here, but as I said, this isn’t a game. It is good for ego points but that is about it. The two 4870 cards are pretty much neck to neck. I suspect that the 1GiG variant is ahead here cuz of its slightly faster core clock. The ram clock is slower, but at the resolution that 3D-Mark06 defaults to isn’t going to put the memory bandwidth to much of a stress.

2. Crysis

The game that makes any single gpu card cry like a baby. Atleast if you want to run it at really uber high resolutions. The game was tested at both 1024*768 and 1024*1280 resolutions. All in game settings were set to "High".

The GTX-260-Core 216 is significantly faster than either of the 4870 cards. It can be argued that the GTX-260-216 is over-clocked significantly, but keep in mind that the 4870s are also over-clocked (as much as they could be)

3. World in Conflict

Two different resolutions tested, along with 4x AA. In-game settings were set to High. AF was set to 4X unless otherwise stated.

Surprise! The faster card here is the over clocked 8800GTX*. That is primarily due to the fact that this game is highly CPU limited. The graphs pretty much trail each other. The 4870s are marginally ahead. The game takes all the CPU power you can throw at it!


*Note: The 8800GTX and 8800GTS-640 test was done with sky-plane setting at 768. Please factor this in when taking into account its result. For the rest of the cards the settings were exactly the same.

4. F.E.A.R.

A pretty old game, but one of my all time favourties. The sequel is due early 2009, till then this will have to do. All in-game settings were set to maximum.

Again the GTX-260 is in the lead at both resolutions. No one is likely to notice this during game play as all cards (even the venerable 8800GTS) deliver good enough frames.

5. GTA IV

Last but not the least, Rockstar’s much awaited port. The game is a system hog. Infact in one interview rockstar said that the game is designed with the computers of the future in mind. Kind of what crytek said when crysis came out. Remember features that would only be enabled once GPUs powerful enough came out. I haven’t seen those features come up, or maybe the right GPU hasn’t come along yet!

Anyways the Game’s Display options were set to:

1. Resolution: 1024*1280
2. Textures: HIGH
3. RENDERER: HIGH
4. VIEW DISTANCE: 36
5. DETAIL DISTANCE: 79
6. VEHICLE DENSITY: 60
7. SHADOWS: 10

Unfortunately the game has no AA support. I didnt try to force any, given that the game doesnt run too well anyways!

The display settings exceed the visual experience that one gets from playing either the Xbox 360 or PS3 version of the game.


Looks can be deceiving. All 3 cards deliver somewhere between 50 and 54 frames on average.

6. POWER CONSUMPTION

The object of this test was to see the difference between idle and load power consumption of the entire system. As the load increases both the GPU and CPU start to crank up their respective use of power. By keeping all parameters the same and only changing the Video cards, it is possible to measure relative differential power consumption of various graphic cards.

The test system was again similar* among the GPUs being investigated.
The power draw was monitored using 'Powerchute' application which comes with APC UPSs. As this is the only device connected to the UPS, the power draw calculations are pretty much accurate approximations.

*The GTX-260-216 was tested using an Asus p5q-E motherboard. The rest of the system specs were the same. The p45 Chipset consumes more power than a p35 and x38 and is comparable to the p965.
This article compares power consumption of the afore mentioned chipsets. p965 Power consumption figures are alsoavailable. For all intents and purposes chipset power conversion is about the same for the two boards. If anything the p965 consumes less power as compared to the p45. Ineffect a GTX-260-Core-216 would consume even less power is a p965 powered board.

The GTX-260 cosumes less power at idle and load as compared to the HD4870 series.

7. BENCHMARK CONCLUSIONS:

The GTX-260-216 is faster as compared to either of the 4870 cards when CPU power isn’t a limiting factor. When the card first came out it was more expensive and slower than the 4870. However as mentioned before, Nvidia had to bear the pain. It drastically reduced prices and introduced this card.

Considering that locally the cards cost almost the same, the GTX-260-216 is a better card. Not only is it faster, you also get to use it with Phys-x whenever that hits mainstream.

Here is a summary of Performance comparison between the GTX-260-216 and the 4870 1GB

I place World in Conflict in its own corner. I believe the game is CPU bound even at low resolutions. Given the amount of calculations that must be done for every object that moves on the battle field this is probably very true.

8. REVIEW CONCLUSION

We already know that it is either faster than or at par with ATi’s equalent offering. We also know that it now costs about the same and that it has the potential to run Phys-x whenever it becomes important (sooner rather than later probably).

However the most compelling reason to get this, rather than an ATi offering, is that most games these days have a “The Way It Is Meant To Be Played” badge. This means that Nvidia has done all that it could to make sure the game runs better on its hardware. If the game has potential to run better on its rival’s GPU then that feature is either neutered (DX 10.1 Pathway in Assassin’s Creed first patch) or is taken away altogether (Benchmark feature in Angel of Darkness). I can cite other examples in lavish details about how this program manipulates designers but I’ll leave that for another day. Suffice to say that this is a very compelling reason to go green, if only to make sure you’ll get the best gaming experience out of 99% of all games that are out there.


ATi’s offerings here need to match their international pricing to become relevant again. There is this post on another local forum about availability of Ge-cube 4870-512MB cards for 19,500 Rupees.

If this is true, then the 512MB version would be a very strong buy at this price. The only game that really shines with a larger frame buffer is Farcry 2.

Right now as things stand you cant go wrong with this card (Nvidia).

And the best bit I saved for those who read to the last: I know of atleast one person who got a FTW edition from Galaxy rather than the SC edition. The FTW are the highest clocked editions out there (EVGAs). Order one and see if you get lucky!


PS. I didnt do any over clocking comparisons as this card is factory over clocked. The clocks can be further improved on (with the core hitting 700), but I really saw no reason to push them. As things stand the card is faster than over-clocked versions of its rivals. I never came across a situation where I saw the need for more oomph from the GPU. The minor gains from over-clocking beyond factory settings are not really worth it.

Click here for feedback and comments