ATI and Nvidia chips

Posted by Robert Dow on July 28th 2008 | Comments Closed
Categories:
Tags:

Alex Gorvoi, Robert Dow, Jon Peddie, and Scooter

Now that the dust and hysteria have settled, we thought this was a good time take a look at the two new competing architectures from the leading GPU suppliers.

Actually it’s taken us this long to collect all the information and run the tests. We ran so many tests and got so many results we stand in awe of the web boys who manage to pump them out in a day or two. Just getting all the eight- pin power connectors worked out ate up hours of time. Nothing is simple anymore.

Fortunately, we had some extra help from Scooter the benchmarking kitty, seen below performing incoming inspection.

We decided to look at the chips in a slightly different way, using key major categories rather than a litany of all their features.

  • Products: We’ll look at four boards (and their GPUs).
  • Performance per Pixel (dots): We’ll use Vantage, no games because of the difficulty of getting anything to run on 64-bit Vista.
  • Package (dimensions): This will be chip and board physical comparisons.
  • Power (Watts).
  • Price (dollars).
  • Pressure (decibels): We would have liked to come up with a way to report scientifically, quantitatively, on noise but lack the room and measuring equipment, however, we do have a sound meter.
  • Parameters: And lastly we’ll talk about some of the special features in both architectures like AA, DP, and tessellation.

Clockwise, the HD4850, GF9800GTX+, GF280, and HD4870.

Products

We are going to examine two boards from ATI and two from Nvidia. Specifically they are:

  • ATI Radeon HD 4850 (RV770).
  • ATI Radeon HD 4870 (RV770).
  • Nvidia GeForce 9800 GTX+ (G92).
  • Nvidia GeForce 280 (GTX280).

Performance

This is what you came for right?

And for those who prefer to look at numbers, here’s the data

As the Figure 1 and Table 2 show the GeForce 280 was the clear winner. In terms of price performance, i.e., benchmark points per dollar, the Radeon HD4850 was the clear winner, as shown in the benchmarks results Table 3.

Product Clock MHz Cores Process Memory Watts Price
HD4850
625
800
55nm
GDDR3
110
$199
HD4870
750
800
55nm
GDDR5
160
$299
GF9800
760
120
55nm
GDDR3
155
$220
GF280
602
240
65nm
GDDR3
236
$449

TABLE 1:Comparison of ATI and Nvidia graphics AIBs.
(Source: Jon Peddie Research)

Figure 1: Benchmark results for ATI and Nvidia’s newest AIBs.
(Source: Jon Peddie Research)

 

HD 4870 HD 4850 GTX 280 9800 GTX+
2650 x 1600 4 AA
1,240
971
3,371
Failed
2650 x 1600 No AA
2,724
2,051
3,670
1,772
1600 x 1200 4 AA
5,243
4,045
6,930
3,518
1600 x 1200 No AA
5,654
4,366
7,357
3,974
1600 x 1200 No AA
5,643
4,360
7,370
-
1280 x 1024 4 AA
7,284
5,722
9,578
5,107
1280 x 1024 No AA
7,876
6,181
10,200
5,870

TABLE 2:Comparison of ATI and Nvidia graphics AIBs.
(Source: Jon Peddie Research)

HD 870 HD4850 GTX280 9800GTX+
Benchmarks per dollar
26.3
30.9
22.7
26.7

TABLE 3: Price-performance: Benchmarks per dollar.
(Source: Jon Peddie Research)

Board Length Width
HD4850
9.25
Single
HD4870
9.25
Double
GF9800GTX+
10.5
Double
GF280
10.5
Double

TABLE 4:AIB sizes.
(Source: Jon Peddie Research)

Package

The AIBs vary in size quite a bit with the HD4850 being the smallest and the Nvidia boards the largest.

The longer Nvidia boards got in the way of the SATA cables on a couple of our test systems, annoying when installing, but not crippling.

Power

The manufacturers have an average power specification, which we’ve recorded in the accompanying Table 5.

We first measured the power consumption of the PC without any graphics board in it. That was 99 watts, we used that as the tare weight.

Interestingly, the Nvidia GTX has the lowest wattage static power consumption, which could be argued is important since most computers aren’t running games of other high-demand graphics applications.

AIB Idle W Under test W
HD4850
85
104
HD4870
121
162
GeForce 9800GTX+
74
151
GeForce 280
71
201

TABLE 5:Power consumption of the AIBs.
(Source: Jon Peddie Research)

AIB Idle Under test
HD4850
-8db
-8 db
HD4870
- 8db
-7db
GeForce 6800GTX+
-8 db
-6db
GeForce 280
-8 db
- 5db

TABLE 6:Power consumption of the AIBs.
(Source: Jon Peddie Research)

Pressure

By far the Radeon HD4850 was the quietest. The system with no graphics AIB in the PC the ambient was -10db, and we used that as the tare. The biggest noise factor was the disk. The relative measurements for the AIBs is shown in Table 6.

We were surprised by how quiet all the boards, and the PC was. We have a couple here that absolutely scream when running and increase their fan speed (and noise) when under test.

Parameters

Both sets of AIBs have unique features which may or may not be called on depending upon the test and/or application. For example, both companies have built-in dedicated hardware video accelerators that are somewhat useful for video cut scenes in games, and very useful for watching real videos (not YouTube Flash.)

ATI has a new AA technique called custom filter anti-aliasing (CFAA) and we discuss that in more detail in this issue where we have the write up on the ATI AIBs.

Summary

We put all our parameters, Price, Performance, Power (max W), and Pressure into Table 7. And if we take the product of them we get P4 values.

So the Radeon HD4850 is the best overall deal, but most likely it’s the GeForce 280 you’re going to want.

HD 4870 HD 4850 GTX 280 9800 GTX+
Price
$299
$199
$449
$220
Performance
7,876
6,181
10,200
5,870
Power (Max W)
162
104
201
151
Pressure (db)
-7
-8
-5
-6tv

TABLE 7:Price, Performance, Power, and Pressure.
(Source: Jon Peddie Research)

Figure 2: P4 factor of ATI and Nvidia’s latest AIBs.
(Source: Jon Peddie Research)

Meet Alex Garovi

Welcome the newest member to the team, Alex the benchmarker. We dumped the work for most of the tests run on the boards and systems on Alex’s desk and told him he could go home when they were finished.

That didn’t work out too well as a threat since it took over three days and three of us to get all the tests run, and even then we didn’t get everything done that we had hoped to.

 

 

 

Discuss this entry

Commenting is not available in this section entry.

Comments