3DNews Vendor Reference English Resource -
All you need to know about your products!
Digital-Daily.com
Digital-Daily

NVIDIA GeForce FX5700 Ultra Review

Author: Andrey Kuzin
Date: 28/01/2004

NVIDIA FX5700 Ultra
VPU chip NVIDIA NV36
Memory DDR2 128 Mb; 2.2 ns
Frequencies: 475/950 MHz
Category: Middle level `2004
Price: 240$

From the very second of examination, the card arouses vague suspicion... Isn't it too "cool" for a mid-end video cards? FX5600 Ultra and Radeon 9600Pro/XT look like tiny dwarfs near it. Its full-sized bulk, huge power stabilizer unit ("My God! The capacitors insides are worth at least 10 bucks!" - exclaimed one of ATI employees when I showed that video card to him), plus a massive radiator - these are just primary exterior attributes of FX5700 Ultra which make it related to Hi-End. The secondary signs are more interesting - the video card offers DDRII (!) 2.2 ns memory onboard, with the 128 MB arranged at 128-bit access.


NVIDIA FX5700 Ultra 128 Mb

As our test session "FX5600Ultra vs Radeon9600Pro" last summer showed, NVIDIA practically lost the previous competition round in the mid-end pricing range. This is especially seen from the first revision of the NV31 core. The chip had to be urgently re-made, with the frequencies raised, and the next version already caught up with RV350.

Seems like NVIDIA decided to secure itself against making mistakes.


I. The NV36 (FX5700) chip

Let's start with the NV36 chip that has come take the place of NV31 and win back the crown to NVIDIA in the mid-end pricing range.


NV36
NV36 / FX5700Ultra Sample (*802x741; 152 kb)


Version of A1 chip

The chip is made following the 0.13 mk process technology and sports formidable dimensions. On the reference card, there is the "Sample" option which nevertheless didn't prevent it from getting a 120% overclock from its nominal 475MHz to 570 MHz.

And here comes the surprise:


VGA The chip Core/
memory speed
Memory type Pixel
pipelines /
Texture units
Process technology Version of
pixel
shaders
FX5950 Ultra 256Mb NV38 475/950 MHz 256-bit 8 (4) / 1(2) 0.13 mk 2.0
FX5900 Ultra 256Mb NV35 450/850 MHz 256-bit 8 (4) / 1(2) 0.13 mk 2.0
FX5700 Ultra 128Mb NV36 475/900 MHz 128-bit 4 (2) / 1(2) 0.13 mk 2.0
FX5600 Ultra 128Mb NV31 ~400/800 MHz 128-bit 4 (2) / 1 0.13 mk 2.0
FX5200 128Mb NV34 250/400 MHz 128-bit 2 //1 0.15 mk 2,0

The very first glance at the tables results in the evident finding - NV36 is in fact NV38 (FX5950) with the number of pixel pipelines halved. That is, both the latest NVIDIA chips - NV38 and NV36 - have been produced on the base of the same core. It's a mystery why no one has yet noticed that :-)

Things with the memory bus are more straightforward - as long as the chip supports 256 bit, it's no problem connecting 128-bit memory to it. That is, the memory bus has been cut-down forcedly. As our tests will show, that was a wise move. The performance of the reference card exceeds all the thinkable limits for today's mid-end video cards and approach the recent high-ends which leaving the rival Radeon 9600XT well behind.

That the chip is unified is quite logical. ATI chose that way a bit earlier. It's too costly to support production and development of three chips at a time for all the pricing ranges. It looks like Hi-end and Mid-end have been merged.

Such approach for the manufacturer suffers from one shortcoming - something has to be done about those overly smart users seeking ways to turn $240 video cards into something worth of a $500 video card :)

Unfortunately, now way.


nVidia FX5700 Ultra nVidia FX5950 Ultra
Unsoldered FX5700 Ultra (left) and FX5950 Ultra (right)

The unsoldering of FX5700 Ultra and FX5950 Ultra differs greatly. As you see, there aren't too many conclusions.


II. NVIDIA FX5700 Ultra 128 Mb Video Card Features Reference board


NVIDIA FX5700 Ultra Front
NVIDIA FX5700 Ultra Front (*2000x1160; 406kb)

NVIDIA FX5700 Ultra Back
NVIDIA FX5700 Ultra Back (*2000x1182; 478kb)

Well, done with the chip.. Now a few words about the card, with the most exciting at the end :-)

Cooling system:
On both sides of the card, there are NVIDIA's standard radiators - calling them "standard", we mean their exterior... or the "spirit" of the solution. Every company caring about its image and brand-awareness, installs cooling systems that offer some sort of continuity in its exterior. Remember the coolers on ASUS or GIGABYTE video cards. So NVIDIA's reference boards have their own spice - radiators with black fins as ever. Some manufacturers preserve this cooling option, some might play around with alternative solutions. Time will tell.

The solution applied provides a very effective cooling of the board. Even with the peak overclocking, you could hold your hand on it without fear of getting burnt for whatever long time.

The card is very easy to dismantle: just unscrew 6 through studs and you are there - the cooling system is removed.
*1200x786; 254 kb
*1200x658; 221 kb

Power:
Like in previous mid- and hi-end cards by NVIDIA, the board offers an additional power connector. This time it's been positioned in the right way - horizontally. With vertical positioning, the risk of damage to the board rises essentially (e.g., one might pull up the connector by the roots while disconnecting the cable). The power supply module is made following the 2-channels scheme and has 10 capacitors for smoothing load in the peak power consumption mode.


Memory:
Samsung 2.2 ns DDR2 made up of 8 chips (4 on the front, and 4 from the rear side). The memory bus offers 128-bit organization. The nominal memory frequency is 450(900) MHz.


Samsung 2.2 ns DDR2

VIVO:
The board offers a space for installing a VIVO chip, but the chip itself in the version we got is missing.


Test configuration

CPU P4 2.4Mhz 800FSB (Northwood D1)
Mb Epox 4PDA2+ (i865PE)
Memory PC2700 2x256Mb 400Mhz in the dual-channel mode
Timings - 2:6:3:3
Video cards NVIDIA FX5700 Ultra
ATI Radeon 9600Pro
ASUS Radeon 9600XT
OS WinXP + SP1
Drivers Detonator 53.03
Catalyst 3.10

Test software:

    Synthetic benchmark:
  1. 3DMark2003 v340;
  2. 3DMark2001SE v320;
  3. Village Mark
  4. Codecreatures v1.0.0 (a DirectX 8.1 application, shaders on, Hardware T&L);
  5. RightMark Video Analyzer v0.4
  6. AquaMark 3 (DirectX 9.0, Vertex Shaders 1.1/1.4/2.0, Pixel Shaders 1.1/1.4/2.0, Hardware T&L, AquaMark3 Triscore mode);

    Old games:

  7. Comanche 4 Bench
  8. Unreal Tournament 2003 Demo (Direct3D, Hardware T&L, vertex shaders, Dot3, cube texturing.);
  9. SSAM The First Encounter (OpenGL);

    New games:

  10. Gun Metal Benchmark 2 v1.20s (a DirectX 9.0 benchmark, Vertex Shaders 2.0, Pixel Shaders 1.1, Hardware T&L);
  11. X2: The Threat Demo (Direct3D, multitexturing, Dot3, running in the benchmark mode embedded in the demo version);
  12. Final Fantasy XI Official Benchmark 2 (a benchmark for assessing the performance in the future game Final Fantasy XI. The developers haven't presented any data on the gaming engine);
  13. HALO: Combat Evolved 1.2 (DirectX 9.0, Vertex Shaders 1.1/1.4/2.0, Pixel Shaders 1.1/1.4/2.0, Hardware T&L);

ASUS Radeon 9600XT, NVIDIA FX5700 Ultra, ATI Radeon 9600Pro
Cards under test: ASUS Radeon 9600XT, NVIDIA FX5700 Ultra, ATI Radeon 9600Pro

Overclocking

The video card was overclocked with the RivaTuner utility. The operation stability in overclocking was verified with the comprehensive benchmark Aquamark3. First, the memory was overclocked. Once distortions appeared on the screen ("broken" pixels, stripes, lines etc.), then we pushed the memory frequency down. Upon getting the highest stable memory operation frequency, we started a sequential overclocking of the core. No additional cooling was used.

Therefore, we were able to overclock the NVIDIA FX5700 video card to the following clock speeds:


Video card Core/memory
standard
Core/memory
Max Over
Excess
NVIDIA FX5700 Ultra 475 MHz / 900 MHz 570 MHz / 1080 MHz 120/120%

This is quite substantial overclock. In this benchmark, FX card traditionally loose to respective Radeon cards. And FX5700 Ultra was no exception. We wanted to find out how it would be possible to raise it in this benchmark:


Aquamark3
AFx4 1024x768
900 DDR2 950 DDR2 1000 DDR2 1050 DDR2 1080 DDR2 1090DDR2
475 MHz 29288 29630 30043 30381 30573 artifacts
500 MHz



31503
520 MHz



32639
540 MHz



33215
560 MHz



34533
570 MHz



34953
580 MHz



artifacts

Again, the same 20% of free advantage. The same table made up as a 10-stage 3D graph looks like this:


Aquamark 3

Superposing the results for the 21st card gives us the general idea of FX5700 Ultra "ranking".


Aquamark 3

Very nice results :). ABIT FX5900 128Mb currently costs 320$. Right behind it, there goes overclocked FX5700 Ultra at $240 with a 1500 tri-score points (1.5 fps/sec).

Seems like this time NVIDIA overdid with the performance of its mid-end solution. Then where should sellers put their old stores of FX5900? There is no other way than reducing the prices :-).

Benchmarking: Synthetic tests


Traditionally, we start with the comprehensive 3DMark 2003. First off, we get the overall score with which we can approximately estimate the board's performance level. Of course, as applied to this benchmark we should mention all the reservations at which we repeatedly made accent in our materials. However, we can't ignore a detailed analysis of performance for the boards in this test :=).

As we see from the graph, NVIDIA GeForce FX 5700 Ultra is a sure leader at all the resolutions. The direct competitor, ASUS Radeon 9600XT built on the ATI Radeon 9600XT chip, lags behind NVIDIA GeForce FX 5700 Ultra, but the lag is negligible enough. Anyway, let's look at the detailed results for 3DMark 2003.


The fillrate is one of the key characteristics of the video card that affects the performance in gaming applications. As we see, in the single-texturing mode NVIDIA GeForce FX 5700 Ultra is second to none. This should affect in the best way in games that do not use pixel and vertex shaders of the DX 9.0 generation.


On the contrary, the most topical mode of multi-texturing demonstrates a definite leadership of the ATI Radeon 9600XT chip. We can only guess that in this case it was the specific configuration of pixel rendering pipelines and texture units in NVIDIA GeForce FX 5700 Ultra that played a leading part, which can take both 4x1 and 2x2 scheme. To the benefit of ATI Radeon 9600XT, it is the higher GPU speed. To fully find out the matter of fillrate in the multi-texturing mode (as well as compare the produced values with the theoretical possible) for both these and some other cards, we'll soon carry out additional tests with a few more benchmarking packages.


At the programmable T&L, the NVIDIA chip does a better job than ATI chips.


The sore spot of all NVIDIA's FX cards is the operation of version 2.0 pixel shaders. NVIDIA GeForce FX 5700 Ultra proved no exception this time even with ForceWare 53.03 installed that re-enabled NVIDIA's proprietary compiler that had been disabled with the version 340 patch in ForceWare 52.16 drivers!


The test in the 3DMark 2003 package itself is quite interesting. The test exerts immense loads on not only the video card (each of the trolls jumping into the abyss is an object made up of 5500 polygons, plus the test uses version 1.4 shaders, and the necessity of using just this version is highly questionable, but anyway let's leave on the conscience of FutureMark =). The results of our video cards are even more interesting =). At the low resolutions, ATI Radeon 9600XT rules. But as the resolutions go up, the difference becomes leveled, and at 1600õ1200 the NVIDIA GeForce FX 5700 Ultra takes a lead over its rival. Hard to say which factor was decisive: the speed of processing the pixel programs or the fillrate, but the fact remains: at the low resolutions ATI is the leader, at high resolutions NVIDIA rules.


This is the first test of the 3DMark 2003 package which take part in the overall rating for the video card. The test makes little use of pixel programs. As a result, NVIDIA GeForce FX 5700 Ultra is a sure leader at all the resolutions.


The second test leaved the palm with NVIDIA GeForce FX 5700 Ultra.


The test is more difficult than the previous one, although the versions of shaders used are the same (1.1 and 1.4, that is, the test is in fact DirectX 8.1-oriented). At that we no longer see the superiority of ATI Radeon 9600 XT in the low resolutions. However, with the rise of resolution the NVDIA GeForce FX 5700 Ultra wins clearly, although the difference in absolute fps values is negligible.


That is the most demanding shader-oriented of all the 3DMark 2003 tests. Used are shaders of versions 1.1, 1.4, 2.0. Here, the results for synthetic benchmarks just make themselves felt (NVIDIA GeForce FX 5700 Ultra didn't do a good job at version 2.0 pixel programs), and we can see the known "dislike" of NVIDIA's FX cards to version 2.0 pixel programs. The NVIDIA GeForce FX 5700 Ultra lags behind at all the resolutions and at performance is approximately in the interim position between ATI Radeon 9600XT and ATI Radeon 9600 Pro.


The overall score in 3DMark2001 SE shows results already familiar from 3DMark 2003. The latest solution for the Middle-End video cards based on NVIDIA GeForce FX 5700 Ultra is a leader at all the resolutions.




The leadership of NVIDIA boards in the Codecreatures test despite the quite active use of version 1.1 pixel shaders can be explained by the intensive work NVIDIA has done at optimizing the test for FX boards. The test is also strongly dependent on the fillrate of the accelerator.


The test is old enough, however is allows to most effectively verify how well the cut-off of hidden surfaces is implemented. Well.. As we see, ATI's technologies at that are not so much better than NVIDIA's: the ATI Radeon 9600XT chip takes a lead.


The leadership at AquaMark is also with ATI Radeon 9600XT. NVIDIA's board lags behind, albeit insignificantly.


At the RightMark package, the leadership ambiguous again. On the one hand, NVIDIA GeForce FX 5700 Ultra rules at the low resolutions. On the other, as the resolutions go up the ATI Radeon 9600XT together with ATI Radeon 9600Pro win, and their results fully coincide.

Now we are moving on to real-world gaming applications.

Final Words

In our previous material regarding comparison of ATI and NVIDIA mid-end video cards ("Comparative test: FX 5600Ultra versus Radeon 9600Pro") the NVIDIA board (built on the NVIDIA GeForce FX 5600 Ultra chip) showed not the best of itself. In fact, the struggle for the Mid-end sector that time was lost by NVIDIA. Of course, by that time NVIDIA hadn't produced the new series of ForceWare drivers which essentially increased the operation speed of FX cards (we'll look into this issue in one of the forthcoming reviews), and 2 revisions of the NVIDIA GeForce FX 5600 Ultra core had been released. Nevertheless, the overall alignment of forces at THAT time was definitely not in favor of NVIDIA.

Today, there is a new leader in the Middle-End sector. NVIDIA GeForce FX 5700 Ultra beats ATI Radeon 9600XT at most tests. Yes, NVIDIA GeForce FX 5700 Ultra still suffers from "childhood diseases" in some way like poor handling of version 2.0 shaders (which nevertheless does not prevent it from beating ATI at games and applications making active use of these shaders). But, what do we have in terms of prices? The approximate price for FX 5700 Ultra is $240, ATI Radeon 9600XT ~200$, i.e. it's up to you to decide... For now, we can only state NVIDIA's victory in the Middle-End sector and monitor prices for boards built on NVIDIA GeForce FX 5700 Ultra and ATI Radeon 9600XT chips. If NVIDIA's pricing policy remains as aggressive as it is now toward ATI, then the leadership is going to be undisputable.

The GeForce FX 5700 Ultra video card was presented by nVidia.

Read more on this topic:

MSI GeForce FX 5950 Ultra
ATI VGA Roundup `2003
3DMark2003 build 320 vs 340: A rare moment of truth?
Tests of ForceWare 52.16: FX 5900 versus Radeon 9800Pro
ASUS RADEON 9800 XT: a turning point
Tests of ATI Radeon 9800 PRO
FX 5900 versus Radeon 9800Pro
FX 5600Ultra versus Radeon 9600Pro

Copyright © 2005 Digital-Daily. All Rights Reserved.
contact - info@digital-daily.com