Radeon HD4800 series of graphics cards is real refreshment in these days. Every new generation of GPU brought a big shift in performances and ATi this time showed us that top model graphics chip (RV770) doesn’t have to be expensive. Based on experience from previous HD3850 and HD3870 models, we expected that new model HD4870 will dominate in middle-high class, with lower price and higher performances than other competitors. But, we were wrong… Tests showed something quite unexpected. Still, before benchmark results let’s see what new GPU offers.

New Series

As unwritten rule says, the one that first masters a new manufacturing process, later will rule on the market. As a matters stand now, nVIDIA isn’t still introduced its 55nm GPU, while AMD/ATi already has second generation graphics chip made in 55nm technology process. This made possible for ATi to produce chip with 260mm2 surface (half less from surface of one NV GTX 280 ). Still, ATi managed to pack even 953 million transistors. Number of stream processors is, as expected, increased and now amounts 800 (previous generation had 320). Number of texture units is significantly enlarged and now it counts 40 which is rather large number if we consider that previous generation had only 16 (maybe that was the biggest fault of older series) while number of Raster operations stays the same, number of ROP is 16.
Consumption in case of Radeon HD4850 model was somewhere around 110W and if we compare it with 105W for HD3870 it is obvious that in HD4870 was put a lot of effort. For the sake of truth, HD4870 consumption is around 150W, a little too much just on first sight… HD4870 works at 750MHz and HD4850 at 625MHz, so from this fact resulted main differences for higher, but still optimal consumption of HD4870. Both HD48xx models communicate with memory through 256-bit bus. What really distinguishes these two graphics cards is memory. For HD4870 is used, for the first time GDDR5 memory as HD4850 uses GDDR3 standard. New GDDR5 memory brought bigger memory bandwidth which enabled flow from 115.2GB/sec (63.4GB/sec is for HD4850). We should also mention that clock for HD4870 is 1800MHz and for HD4850 is 993MHz.

Premier specimen of HD4870 in our test lab was signed by Force3D brand (this is first product from Force3D that we saw so far). It isn’t that important who signed graphics card that ATi designed, much important is how it works. HD4870 is completely different story, comparing it with HD4850, when it comes to cooling solution. Everything that was wrong on HD4850, on HD4870 is corrected. We didn’t mind to demount cooler from graphics card and to comment what we found under the hood. But before that, let’s take a look on PCB of HD4870 model.

Qimonda GDDR5memory chips arranged in “L” way, i.e. two rows by 4 chips and every chip has 64MB, which in total sum gives total of 512MB of video memory. There are some announcements for HD4870 models with 1GB video memory that should give better results in high resolutions.

Cooling is really bulky and it is based on two large copper heatpipes that go through dense aluminum ribs. Responsible for transfer of heat on aluminum ribs is turbine type fan. Turbine itself is very efficient and heat is efficiently transferred out of case. We noticed that cooler sits perfectly on GPU, but memory chips don’t. More precisely, ATi decided to use thermal tapes, not the best but the cheapest solution. This choice definitely had effect on price: by using thermal tapes costs for finishing cooler body are less. However it may be, it turned out that these tapes, as well as whole cooling solution is, very efficient. Turbine or fan is up to 15% of max RPM noiseless (but card is very warm) while on 100% you cannot stand the noise, but the cooling is still top-level. Still, we found optimum on 35% of maximum RPM. On these setting graphics cards wasn’t noisy, and proper cooling was still provided.