 | |||
|
|
|
|
|
 |
eVGA e-GeForce FX 5700 Ultra Review October 23, 2003 Brandon Bell |
|
 |
|
| |
Summary: Based on NVIDIA's brand new GeForce FX 5700 Ultra GPU, the eVGA e-GeForce FX 5700 Ultra is poised to take on ATI's latest mainstream offering, the RADEON 9600 XT. The GeForce FX 5700 Ultra boasts a higher core clock frequency and even sports DDR2 memory at 900MHz. But that's not all, it can support up to 1GB of memory (including GDDR3) and is built at IBM's 0.13-micron facility. See how this card fares in today's review!
 Introduction | Page:: ( 1 / 20 ) |
It has been a long road to redemption for NVIDIA in the mainstream segment. After delivering the GeForce4 Ti 4200, a GPU that was highly regarded among the press and consumers for its incredible bang for the buck ratio (and still selling well to this day), NVIDIA has had a hard time exiting the GeForce4 Ti 4200’s shadow.
The GeForce FX 5600 and GeForce FX 5600 Ultra were intended to carry the mainstream torch from the GeForce4 Ti 4200 for NVIDIA, but the GeForce FX 5600 Ultra initially launched to lackluster reviews. In many cases, the GeForce FX 5600 Ultra was bested by its predecessor, leaving end user’s pretty jaded. After all, what’s the point of “upgrading” if you actually lose performance?
NVIDIA’s engineers quickly went back to the drawing boards and concocted a revised GeForce FX 5600 Ultra. New flip-chip packaging was implemented for its improved signal quality at higher clock speeds, while the core clock was cranked up to 400MHz, matching the specs of ATI’s RADEON 9600 PRO. When paired with its 800MHz DDR memory, the improved GeForce FX 5600 Ultra was much more competitive with ATI’s mainstream offerings, and just as importantly, the GeForce4 Ti 4200.
[image]
| <% print_image("01"); %> | <% print_image("02"); %> |
Unfortunately for NVIDIA, retail cards based on these revised specifications didn’t hit the market until late July, giving ATI another two months of essentially unchallenged dominance at the upper echelon of the mainstream segment. NVIDIA’s only answer was the GeForce FX 5600. This card was woefully underpowered in comparison to the RADEON 9600 PRO, as it was intended to compete with the RADEON 9600 at the $150 price point.
Board manufacturers did all they could to attempt to sell their GeForce FX 5600 cards, some added up to 256MB of memory to attract the eyes of consumers, while others used confusing product names that were misleading if you didn’t read the fine print. None of these cards could come close to the RADEON 9600 PRO in most situations, but since NVIDIA couldn’t deliver the hardware, they didn’t have many options.
[image]
| <% print_image("03"); %> |
What went wrong you ask? Based on what we can see, NVIDIA got stung by the same problem it had with GeForce FX 5800/5800 Ultra: manufacturing. Apparently, NVIDIA just couldn’t produce the higher frequency 5600 Ultra cores in the quantities it needed initially. Also remember that the revised flip-chip core was rushed into service after the initial 5600 launch. When you combine these two factors together, NVIDIA obviously had an enormous pair of obstacles to overcome.
While these issues have since been resolved, NVIDIA has been hard at work developing a follow-up to the GeForce FX 5600 Ultra: the GeForce FX 5700 Ultra. This card is based on an entirely new graphics core!
SIDEBAR: The official PR
| New core | Page:: ( 2 / 20 ) |
If you recall the design of GeForce FX 5600, you’ll remember that it is a derivative of NV30, NVIDIA’s GeForce FX 5800 family. NVIDIA essentially used NV30 as a building block for GeForce FX 5600 in much the same way an auto manufacturer builds multiple car lines off the same platform.
With approximately 125 million transistors, NV30 was expensive for NVIDIA to produce. They needed to come up with an alternative design for GeForce FX 5600 that was cheaper to manufacture in order to serve the mainstream market. In order to accomplish this, NVIDIA sliced off a TMU: while GeForce FX 5800 offers four pixel pipelines with two texture units per pixel pipeline (4x2), the GeForce FX 5600 family is built on a more conservative four pixel pipeline with one texture unit per pixel pipeline architecture (4x1). NVIDIA also modified GeForce FX 5600’s memory subsystem, GeForce FX 5600 employs dual 64-bit memory controllers versus four 32-bit memory controllers in GeForce FX 5800.
For GeForce FX 5700 Ultra, NVIDIA went back to this formula; only this time they used the GeForce FX 5900’s NV35 core as the new starting point. From there, NVIDIA made some adjustments to the core to make the GeForce FX 5700 Ultra less expensive to manufacture.
[image]
| <% print_image("04"); %> | <% print_image("05"); %> |
128-bit memory interface
Before you start drooling over the prospects of a mainstream card with a 256-bit memory interface, don’t. GeForce FX 5700 Ultra did not adapt the GeForce FX 5900’s 256-bit memory interface. Instead GeForce FX 5700 Ultra relies on a 128-bit memory interface (with two 64-bit memory controllers), just like the GeForce FX 5600 Ultra.
One key difference between the GeForce FX 5700’s memory controller and that of its predecessor is its support for new memory types. GeForce FX 5700 supports conventional DDR memory like the GeForce FX 5600, but also DDR2 and GDDR3.
GDDR3 (short for graphics double data rate SDRAM) is a new memory type developed specifically for graphics. ATI and NVIDIA worked closely with memory manufacturers in the design and development of GDDR3’s specifications. GDDR3 is designed to operate at significantly higher clock frequencies than previous memory types without consuming large amounts of power. Micron in particular made a splash in June of this year when it announced it had shipped its first GDDR3 samples to both ATI and NVIDIA. Its website currently lists modules as high as 700MHz (1.4GHz effective) that will begin shipping this quarter.
In the meantime, GeForce FX 5700 utilizes first generation DDR memory, while the GeForce FX 5700 Ultra employs DDR2, just like NVIDIA’s GeForce FX 5800 Ultra.
The GeForce FX 5700 Ultra’s DDR2 memory is clocked at 450MHz (900MHz effective) providing the 5700 Ultra core with up to 14.4GB/sec of memory bandwidth. Like the 5600 Ultra, the 5700 Ultra’s most popular memory configuration appears to be 128MB; although it’s possible a board manufacturer may chose to produce a 256MB card (the core can support up to 1GB of memory).
SIDEBAR: Micron’s GDDR3 memory
| The core (cont’d) | Page:: ( 3 / 20 ) |
Pixel pipelines
Besides the new memory controller, the GeForce FX 5700 Ultra core sports many changes. Remember, we’re dealing with a cost-reduced version of the NV35 core here, so NVIDIA is basically passing all of the new technologies first introduced in its latest flagship product down to the mainstream consumer. The only difference is performance, to ease manufacturing costs NVIDIA reduced the GeForce FX 5700/5700 Ultra to a 4x1 pixel pipeline architecture, GeForce FX 5900/5950 is a 4x2 design.
As a result of the changes, GeForce FX 5700 is composed of approximately 82 million transistors (versus 130 million in NV35); this is a slight increase from the 80 million transistors in GeForce FX 5600.
The core itself operates at a clock frequency of 475MHz, an increase of 75MHz from the GeForce FX 5600 Ultra, but 25MHz shy of the RADEON 9600 XT’s 500MHz core clock speed. This gives ATI a slight advantage in peak theoretical fill rate 2.0 Gigatexels/sec versus 1.9 Gigatexels/sec in GeForce FX 5700 Ultra.
New manufacturing partner
The GeForce FX 5700 is also the first NVIDIA product to be manufactured by IBM. NVIDIA previously relied solely on TSMC’s foundries. NVIDIA plans to continue its relationship with TSMC, but IBM will also be rolled into the mix, giving NVIDIA two manufacturing partners. It’s expected that NVIDIA’s next-generation part will also be manufactured by IBM. GeForce FX 5900 and 5950 cores will continue to be manufactured by TSMC.
Like the GeForce FX 5600, the GeForce FX 5700 is built on a 0.13-micron manufacturing process with copper interconnects, but IBM spices up the formula by fusing low-k dielectric material to the combination. As we mentioned in our RADEON 9600 XT article, low-k dielectric shields the copper circuits within the graphics core from each other, preventing electrical crosstalk from occurring. This allows NVIDIA to increase core clock frequency without wasting power. One interesting tidbit we picked up while talking with NVIDIA about the new IBM process is that they were able to go straight into full production from first silicon, just like NV35.
Features
Since the GeForce FX 5700 is based on the GeForce FX 5900/5950’s roots, you’ll see all of the same fundamental technologies incorporated in those cores integrated into the GeForce FX 5700 as well. This includes NVIDIA’s UltraShadow technology. UltraShadow is designed to improve performance when shadows are used extensively. The technology works by limiting all lighting calculations to specific areas that are predefined by the game developer, allowing the shadow to be rendered more quickly and with greater efficiency. We like to think of it as occlusion culling, but exclusively for shadows.
NVIDIA has also adopted the new compression algorithms in its Intellisample HCT technology into GeForce FX 5700. This should result in improved anti-aliasing performance at high resolutions. Finally, NV35’s CineFX 2.0 enhancements have been included. This allows pixel shader programs to be executed more quickly, enhancing shader performance. We’ll put this theory to the test first.
SIDEBAR: Remember the original GeForce 256? It was the first consumer card with DDR memory.
| ShaderMark 2.0 | Page:: ( 4 / 20 ) |
ShaderMark 2.0
Notes
As you can see, the GeForce FX 5700 Ultra’s new core is definitely improved in shader performance, but still trails ATI’s offerings significantly. Another problem that’s apparent is the lack of support for some shaders as well as features such as HDR.
SIDEBAR: The GeForce 4 contained approximately 63 million transistors
| Tomb Raider shader performance | Page:: ( 5 / 20 ) |
Tomb Raider
SIDEBAR: The codename for GeForce FX 5700 Ultra was NV36.
| The eVGA card | Page:: ( 6 / 20 ) |
The eVGA e-GeForce FX 5700 Ultra, like previous eVGA cards, is based entirely on NVIDIA’s reference design. NVIDIA actually handles the manufacturing (via Flextronics); eVGA then takes the card and slaps their sticker on it. This system has worked marvelously for both companies. eVGA doesn’t have to manage expensive manufacturing facilities, while NVIDIA is assured that the board maintains a certain quality level. At one point NVIDIA had developed a bad reputation for 2D display quality that was actually the fault of individual board manufacturers who skimped on board components. By producing boards directly, NVIDIA has better control over quality.
[image]
| <% print_image("06"); %> | <% print_image("07"); %> |
Another benefit is that eVGA is able to bring boards to market more quickly. eVGA was actually the first card manufacturer to have GeForce FX 5900 Ultra cards at retail and online outlets.
The board design is a remarkable departure from the GeForce FX 5600 Ultra. The PCB is 9” long, the same length as GeForce FX 5900/5900 Ultra and GeForce4 Ti 4600. You can also see that the GeForce FX 5700 Ultra has significantly more capacitors than the GeForce FX 5600 Ultra. An external power source is still required, which is located in the upper right corner of the board, parallel to the edge of the PCB. This makes installation a little easier than on the GeForce FX 5900/5900 Ultra.
DDR2 modules consume less power than conventional DDR, but tend to generate more heat. To combat this, a large heatsink covers the memory modules as well as the graphics core itself. We were a bit surprised to see NVIDIA go with a one-piece unit for the heatsink on the top of the card, as heat from the core is also spread onto the memory modules (and vice versa). This is why NVIDIA elected to use a two-piece unit for the GeForce FX 5900 Ultra. One benefit however, is that the surface area of the heatsink is increased.
[image]
| <% print_image("08"); %> |
Unlike the GeForce FX 5900 Ultra, the 5700 Ultra is a single-slot cooling design, so the PCI slot adjacent to the AGP interface isn’t consumed. A much larger fan is used to keep the graphics card cool, which is located slightly offset of the graphics core. The fan then blows air across the fins of the heatsink.
While the fan looks somewhat intimidating, it’s actually quieter than the fan used on the 5600 Ultra (or 5900 Ultra for that matter). Like previous GeForce FX products the GeForce FX 5700 Ultra operates dynamically, in 2D applications clock speeds are reduced to 300/900.
SIDEBAR: The e-GeForce FX 5700 Ultra card actually shipped at 906MHz on the memory, we have a feeling this is similar to the NV28 Ti 4200 cards that shipped at 513MHz, when the spec was officially 500MHz.
| More card details/bundle | Page:: ( 7 / 20 ) |
As you can see in the pictures, eVGA elected not to include the Philips decoder chip for video input support, as they’ve done on previous products. Quite frankly, we’re not surprised to see this feature go, as few mainstream cards have it. Consumers tend to prefer lower price over video input support.
You can actually see where NVIDIA silk-screened the PCB for the Philips decoder:
[image]
| <% print_image("09"); %> |
At this time, eVGA has no plans to introduce a GeForce FX 5700 Ultra variant with video input support (or memory configurations greater than 128MB). The card we’re reviewing today is the only 5700 Ultra SKU planned.
e-GeForce FX 5700 Ultra packaging
eVGA bundles the card with a DVI-to-VGA adapter and an S-Video cable, surprisingly, eVGA continues to omit a power adapter cable. Also included are two case badges, NVDVD 2.0, Ghost Recon, America’s Army, NVIDIA’s 3D technology demos, WindowBlinds, a demo copy of EarthViewer and a trial version of Roxio’s Photosuite 5.
eVGA continues to bundles its cards with its Automated Driver Management (ADM) software. ADM ensures that you have the drivers for your chipset and AGP GART drivers installed. Once it has verified that these drivers are in place, it then installs the NVIDIA Detonator driver. If they aren’t installed, ADM automatically installs them for you. This is important, as these drivers can have a significant impact on the performance and stability of your system if they’re not installed.
Price and availability
The e-GeForce FX 5700 Ultra is currently shipping and will be priced at $219 with a $20 mail-in rebate. This is right in line with NVIDIA’s official pricing for the GeForce FX 5700 Ultra, so the e-GeForce FX 5700 Ultra is priced competitively. Of course, prices tend to slowly come down as competition among retailers sets in, and online prices are usually a bit lower than retail. You shouldn’t have a problem finding cards online, while eVGA distributes some of its cards through Best Buy and Office Depot as well.
With the debut of the GeForce FX 5700 and GeForce FX 5700 Ultra, the GeForce FX 5600 family will slowly be phased out in favor of the GeForce FX 5700 family. Expect prices on these cards to fall quickly before they all disappear. The GeForce FX 5700 will officially be priced at $149, $50 less than the GeForce FX 5700 Ultra.
SIDEBAR: Expect board manufacturers to ship their GeForce FX 5700 cards somewhere in the 425 core/275MHz memory range.
| Test Systems | Page:: ( 8 / 20 ) |
System Setup
Intel Pentium 4 3.2GHz
ASUS P4C800 Deluxe
512MB OCZ EL PC3200 (DDR400) SDRAM
ATI RADEON 9600 XT
ATI RADEON 9600 PRO
ATI RADEON 9500 PRO
ATI RADEON 8500
ATI RADEON 9800 128MB (non PRO)
Driver version CATALYST 3.8
eVGA e-GeForce FX 5700 Ultra
eVGA e-GeForce FX 5600 Ultra
MSI GeForce FX5600-VTDR128
MSI GeForce FX5900-VTD128
NVIDIA GeForce4 Ti 4200-8X
NVIDIA GeForce4 Ti 4600
NVIDIA GeForce3
Driver version Detonator 52.16
30GB IBM Deskstar DTLA 307030 ATA/100 Hard Drive
Windows XP Professional
DirectX 9.0b
Benchmarks
3DMark 03
NASCAR Racing 2003 Season (Bristol custom demo)
Quake III: Arena version 1.32 (fscrusher demo)
Unreal Tournament 2003 (T2 custom demo)
IL-2 Sturmovik: Forgotten Battles (The Black Death track)
Splinter Cell (FS custom demo)
Tomb Raider: Angel of Darkness (Beyond3D custom demo)
SIDEBAR: The P4C800 Deluxe was running in Turbo mode for all tests.
| 3DMark 03 Fill rate | Page:: ( 9 / 20 ) |
SIDEBAR: Did think you’d see 3DMark 03 again did you?
| NASCAR Racing 2003 Season | Page:: ( 10 / 20 ) |
Nascar 2003: OpenGL (for NV cards)
SIDEBAR: Hmm, would you guys be interested in seeing NASCAR Thunder 2004 replace Papy’s NASCAR Racing Season 2003? Personally I’m a bit disappointed to see NASCAR and Papyrus part ways after so many years.
| IL-2 Sturmovik: Forgotten Battles | Page:: ( 11 / 20 ) |
IL-2 Sturmovik: FB: OpenGL
SIDEBAR: Mmmm, Lock On: Modern Air Combat.
| Quake III 4xAA 8xAF | Page:: ( 12 / 20 ) |
Quake III - OpenGL
SIDEBAR: We’re still using our custom demo for testing with Quake 3.
| Unreal Tournament 2003 4xAA | Page:: ( 13 / 20 ) |
Unreal Tournament 2003 – Direct3D
SIDEBAR: You’ll want to check out the AF benchmarks with UT after seeing these numbers.
| Splinter Cell | Page:: ( 14 / 20 ) |
Splinter Cell – Direct3D
SIDEBAR: Chris is working on another image quality shootout article, you’ll see that tomorrow.
| Tomb Raider | Page:: ( 15 / 20 ) |
Tomb Raider – Direct3D
Notes
We’re using Reverend’s (of Beyond3D) custom Prague3a demo, which is also used at Beyond3D. Quite frankly, we were astonished by the results, after trailing ATI by a huge margin, they’ve actually managed to catch up in this test. Mind you, we disable features like depth of field, and glow in our testing, both of which may run faster on ATI hardware when enabled in the game. We’ll definitely be looking into our settings and how the affect framerates on ATI/NVIDIA hardware as there are certainly a plethora of options available, but it’s clear that NVIDIA’s new driver has definitely improved performance in this title.
SIDEBAR: Depth of field is disabled in all Tomb Raider tests.
| IL2:FB 4xAA | Page:: ( 16 / 20 ) |
IL-2 Sturmovik: FB
SIDEBAR: We use the perfect quality settings for testing with IL-2.
| UT 2003 4xAA 8xAF | Page:: ( 17 / 20 ) |
Unreal Tournament 2003
Notes
While we don’t want to spoil the results of Chris’ upcoming Image Quality Shootout article, which just missed today’s launches, we did want to let you know that NVIDIA’s Detonator 52.16 driver still disables trilinear filtering in UT2K3, so we went ahead and manually adjusted the RADEON cards to a similar filtering level, which turned out to be the “Performance” setting.
As you can see, the RADEON cards are actually still competitive with NVIDIA’s pseudo-trilinear even in default settings, but we didn’t want to give the impression that GeForce FX 5700 Ultra was “faster” per se in this test, as clearly it does less work than ATI by default.
SIDEBAR: GeForce FX 5700 Ultra cards will hit retail this Sunday.
| Overclocking | Page:: ( 18 / 20 ) |
Quake 3
Unreal Tournament 2003
SIDEBAR: We were actually able to overclock the board much higher, but we couldn’t get the texture corruption to go away regardless of cooling.
| Ballistics Report | Page:: ( 19 / 20 ) |
Pros
Performance: When it comes to performance, the e-GeForce FX 5700 Ultra is certainly no slouch. As you saw in our performance testing, the e-GeForce FX 5700 Ultra is quite a performer, running neck-and-neck with its closest competitor, the ATI RADEON 9600 XT in many benchmarks.
Sure, it’s not going to take down a GeForce FX 5900 or RADEON 9800, but with only four pixel pipelines and a 128-bit memory interface, it certainly performs well. This is largely thanks to its blazing DDR2 memory, which provides up to 14.4GB/sec of memory bandwidth. That’s an improvement of nearly 5GB/sec over RADEON 9600 XT, which operates 300MHz slower. NVIDIA does give up a little bit of fill-rate to ATI, but when you combine this with the performance enhancements found in NVIDIA’s Detonator 52.16 driver, the GeForce FX 5700 Ultra performs very competitively.
Price: It may not come with a free voucher for Half-Life 2, but the e-GeForce FX 5700 Ultra is still priced to move at $200. When we heard that NVIDIA would be shipping the GeForce FX 5700 Ultra with high-end 900MHz DDR2 memory, we figured it would be expensive. After all, the GeForce FX 5800 which used similar RAM was officially priced at $400 when it launched.
Of course, GeForce FX 5700 Ultra isn’t a 130 million transistor chip, but we’re still looking at a card with a very expensive board design.
In any case, it’s great to see NVIDIA insert the GeForce FX 5700 Ultra right in the slot occupied by the GeForce FX 5600 Ultra, despite the higher manufacturing costs. By the way, if $200 is still too rich for your blood, prices on GeForce FX 5600 Ultra cards should continue to drop as 5700 Ultra boards hit retail. Eventually these cards will disappear, so they may be a pretty tempting upgrade if the price is right.
Availability: Unlike the GeForce FX 5600 Ultra, which was practically stillborn upon its re-launch in May, GeForce FX 5700 Ultra boards will actually be available. GeForce FX 5700 boards should slowly trickle onto the market as well, but the 5700 Ultra’s are in the final stages of shipping as we speak, if they haven’t already (they’re set to hit retail this weekend).
eVGA support: All too often you see examples of video card manufacturers who don’t stand behind their product. Fortunately, eVGA isn’t one of those companies. eVGA demonstrated this last summer during the GeForce FX 5900 Ultra launch. Many end users had problems with their boards in 3DMark 03 that were the cause of a bad BIOS. Once eVGA tracked the problem down, a fix was issued immediately, but in the meantime eVGA was promptly replacing cards for those who had problems. It’s this kind of excellent customer service that’s so hard to find in the industry these days. We have no doubt that eVGA earned countless repeat customers as a result of this.
Cons
DX9 game support via drivers: As it stands now, NVIDIA’s in a position where they’re having to optimize specifically for DirectX 9 games. This isn’t a problem if the game you’re currently playing has been optimized by NVIDIA, but what about the titles of tomorrow or that obscure first-person shooter they haven’t looked at? In these cases, we just don’t know.
It’s this kind of uncertainty that has the whole industry guessing what’s going to happen next. NVIDIA is working on a compiler that should help the situation, and they’re working with many game developers to optimize their code for GeForce FX hardware from the start. But one thing’s for certain, it’s going to be more important than ever to make sure you keep your drivers up to date if you have an NVIDIA DX9 board.
SIDEBAR: Oops, almost forgot the link to eVGA’s website.
| Final Verdict | Page:: ( 20 / 20 ) |