facebook rss twitter

Review: AMD's ATI Radeon HD 4850 and 4870: bloodying NVIDIA's profits

by Tarinder Sandhu on 25 June 2008, 11:56

Tags: AMD (NYSE:AMD), Sapphire, ATi Technologies (NYSE:AMD)

Quick Link: HEXUS.net/qanuu

Add to My Vault: x

HEXUS.bang4buck, temps, power-draw, overclocking



HEXUS.bang4buck

In a rough-and-ready assessment of the cards' bang per buck, we've aggregated the 1,920x1,200 frame-rates for the four games, normalised them* and taken account of listed the cards' prices.

We're being cruel on the Sapphire Radeon HD 4850, frankly, by evaluating its HEXUS.bang4buck at the 1,920x1,200 setting, yet if you want to play with the big boys...........

But, even so, there are more provisos than we'd care to shake a stick at. We could have chosen three different games, the cards' prices could have been derived from other sources and pricing tends to fluctuate daily.

Consequently, the table and graph below highlight a metric that should only be used as a yardstick for evaluating comparative performance with price factored in. Other architectural benefits are not covered, obviously.

Graphics cards Sapphire Radeon HD 4850 512MiB Sapphire Radeon HD 4850 XF Sapphire Radeon HD 4870 512MiB Sapphire Radeon HD 4870 XF Sapphire Radeon HD 3870 X2 1024MiB BFG GeForce GTX 280 1024MiB ZOTAC GeForce 9800 GX2 1024MiB BFG GeForce 9800 GTX+ 512MiB BFG GeForce 9800 GTX 512MiB BFG GeForce 9800 GTX SLI
Actual aggregate marks at 1,920x1,200 180.99 285.9 225.68 330.47 202.02 309.02 306.75 203.28 192.13 321.41
Aggregate marks, normalised*, at 1,920x1,200 126.51 206.9 162.71 238.22 151.29 224.86
228.33
141.51 134.17 238.63
Current pricing, including VAT £125
£250
£175
£350 £239
£439
£299
N/A***
£179
£358
HEXUS.bang4buck score at 1,920x1,200 1.01 0.83 0.93 0.68 0.63 0.51
0.76
N/A 0.75 (1.07**) 0.67
Acceptable frame rate (av. 60fps) at 1,920x1,200 No (ET, Crysis, LP) No (Crysis, LP) No (Crysis, LP) No (Crysis, LP) No (Crysis, LP) No (Crysis, LP) No (Crysis, LP) No (ET, Crysis, LP)
No (ET, Crysis, LP) No (Crysis, LP)


* The normalisation refers to taking playable frame rate into account. Should a card benchmark at over 60 frames per second in any one game, the extra fps count as half. Similarly, should a card benchmark lower, say at 40fps, we deduct half the difference from its average frame rate and the desired 60fps, giving it a bang4buck score of 30 marks. The minimum allowable frame rate is 20fps but that scores zero.

As an example, should a card score 120fps we treat it as 90fps as only half the frame rate above 60fps is counted for the bang4buck - this is the formula: (120-((120-60)/2)). Similarly, should it score 30fps, we count it as only 15fps: (30+((30-60)/2)).

The reasoning behind such calculation lies with playable frame rates.

Should card A score 110fps in a benchmark and card B 160, then card B would otherwise receive an extra 50 marks in our bang4buck assessment, even though both cards produce perfectly playable frame rates and anything above 60fps is a bonus and not a necessity for most.

Similarly, without our adjustments, the aggregated bang4buck total for two very different cards would be identical if, in a further benchmark, card A scored a smooth 70fps and card B an unplayable 20fps. Both would win marks totally 180, yet the games-playing experience would be vastly different.

A more realistic (and useful) assessment would say that card A is better because it ran smoothly in both games - and that view would be accurately reflected in our adjusted aggregation, where card A would receive 150 marks (85+65) and card B 100 (100+0).

In effect, we're including a desired average frame rate, in this case 60, and penalising lower performance while giving frame rates higher than 60fps only half as much credit as those up to 60fps. If this doesn't make sense or you have issue with it, please hit the HEXUS community.

**  assuming a price of £125; we cannot confirm this until it happens.

*** it would be wrong of us to conjecture on the value of a product that isn't hard-launched for three weeks.

Here's the HEXUS.bang4buck graph at 1,920x1,200.



The graph divides the normalised score by the price.

What we see is that, with value taken into account, the aggressively-priced Radeon HD 4850 takes top honours, followed by the Radeon HD 4870 and then, surprisingly, the HD 4850 in CrossFire.

We've not accounted for NVIDIA's price repositioning, and it's only right to do so when it happens. Should the 9800 GTX fall to £125 it will look the best of the bunch.

Whatever the case, ATI has produced performance and value from its HD 4800 series. Our choice, based on acceptable frame-rate, would be the HD 4870, though.

Temperature musings

We perform our testing in an open test bed, with a 120mm fan simulating case airflow. 

Graphics cards Sapphire Radeon HD 4850 512MiB Sapphire Radeon HD 4870 512MiB Sapphire Radeon HD 3870 X2 1024MiB BFG GeForce GTX 280 1024MiB ZOTAC GeForce 9800 GX2 1024MiB BFG GeForce 9800 GTX+ 512MiB BFG GeForce 9800 GTX 512MiB
Ambient temperature 21°C 25°C 19°C 21.5°C 22°C N/A 21.5°C
Idle temperature 71°C 71°C 56°C 47°C 60°C N/A 52.5°C
Load temperature 81°C 83°C 80°C 74°C 80°C N/A 67°C
Ambient-to-load delta 60°C 58°C 61°C 52.5°C 58°C N/A 45.5°C

We've ignored the pseudo GeForce 9800 GTX+'s readings because the real-deal card will be based upon power-efficient 55nm technology.

Looking at the rest, the idle temperatures of the Radeon HD 4800-series cards are way high and, as mentioned earlier, will probably need a BIOS fix to cure the problem.

Power-draw

Minimising the skew from different platforms, all cards were run on the ATI platform detailed on page 15, other than having a different CPU, QX9650, and differing memory, 4GiB DDR3-1,333.    

The Canyon test from 3DMark06 was run at 1,920x1,200 with 4xAA and 16xAF.

Graphics cards Sapphire Radeon HD 4850 512MiB Sapphire Radeon HD 4870 512MiB Sapphire Radeon HD 3870 X2 1024MiB BFG GeForce GTX 280 1024MiB ZOTAC GeForce 9800 GX2 1024MiB BFG GeForce 9800 GTX+ 512MiB BFG GeForce 9800 GTX 512MiB
Idle load 123 149 119 127 174 N/A 141
Full load 203 229 289 279 313 N/A 223

As you'd expect, the Radeon HD 4850 is the most frugal of the bunch. The HD 4870 draws around the same power as a GeForce 9800 GTX, matching paper specifications. All others area wattage hogs, frankly.

It will be interesting to see where the pukka GeForce 9800 GTX+ jumps in.

Overclocking

With both the HD 4850 and HD 4870 fans not pitching up to full speed when under load, overclocking is for reference only. We managed to push the HD 4850's clocks from 625/2,000MHz to 680/2,060MHz. This pushed ET:QW 1,920x1,200 performance from 51.57fps to 59.3fps - almost matching a 3870 X2.

The sample HD 4870 would barely go any higher on either the core or memory. We only got a couple of extra fps above its stock performance, and we'll look back at it once the fan-speed issue has been fixed.

Wrapping it all up now.