ATI memory

[mr.max-amd]

i have never owned an ati but im an amd fanboy but i just dont get why they wont make their busses on the higher end cards higher
i mean GDDR5 on a 256bit bus does ok against ddr3 on a 448bit bus
i just dont get this one
why not gddr5 on a 512bit bus?
that would destroy nvidia
i mean thats the only reason i wont buy an ati card
id love to buy a 4890 but i want the better mem of a gtx 260 or a gtx 275
anyone know why no bigger busses for ati?

[Kendle6661]

Well, it's pretty simple, it's all about money. AMD/ATI just can't justify the added expense for using gddr5 on a 512 bit bus line. Not many people are willing to spend 600-700 on a video card and AMD knows this

[Badbonji]

There are two ways for high bandwidth on gpu's atm, GDDR5 and 512 bit bus. These are expensive, so to keep the cost down they only use one of them, and the bandwitdh is still enough as to say increasing it doesn't gain much fps (To simulate this I think you can overclock the memory, and see how much it improves - in my experience it doesn't do as much as increasing core/shader)

[Rattle]

512bit wont help 4series and they are not far behind gtx200 series with 256bit.

[mr.max-amd]

ok, so if there equal in memory performance
what about stream processors? i mean doesnt 1 nvidia sp equal like 4 ati
basicaly i think you guys can probably tell its time for an upgrade and i cant decide between the gtx 260 or maybe the gtx 275 and the 4870 and the 4890
basicaly what performs better a gtx 260 or a 4870 and a 275 or a 4890
if i wait for the 4890 and 275 it may become a money issue
so what would be better
4870 or gtx 260
and
4890 or gtx 275

[Firestrider]

Nvidia has fewer faster clocked shaders, and ATI has more slower clocked shaders. Theoretically the 4000 series cards have better single-precision and double-precision floating point performance than the GTX 200 series, but this is not represented in games or Folding@home.

The reason why ATI went with a 256-bit bus is because they could get the performance targets they wanted thanks to GDDR5 without the added cost and complexity of a wide bus. ATI has already tried using a 512-bit bus with the 2900 series cards which resulted power hungry beasts (the 2900XT has the highest TDP of any graphics card to date) with large die sizes that still didn't reach the performance of Nvidia's 8800 series cards.

From what I recall a 4870 is equivalent to a GTX 260-192, and a 4850 is equivalent to a 9800GTX+. You won't see much of an improvement over your current 9800GT, so I say wait.

[mr.max-amd]

im definetly going to see some more power
and my card is giving me problems
so gtx 260 216 then?
maybe the gtx 275?

[Farinorco]

Regarding memory buses, there's another reason to not use a wider bus (apart from cost and circuitry complexity): the wider the bus, the bigger has to be the chip (the GPU in this case). The NVIDIA's G200 (chip in GTX260 and GTX280) and G200b (the 55nm version of the same chip) are both mammoth-sized enthusiast oriented chips (576mm2 for G200 I think, I don't know about G200b but it should be around 412mm2) and they can hold a giant 512bit bus perfectly (448bit after the "cutting" process in GTX260).

On the other hand, ATi's RV770 reaches those performance levels with a much smaller size. RV770 is 256mm2 (less than half a G200 chip), and I don't think such a little chip could have a much bigger than 256bit memory bus. Fortunately to ATi the use of GDDR5 gives the same bandwidth to the memory bus with a width of 256bit than what NVIDIA has with their 448/512bit and GDDR3, ending with a value oriented chip that performs competitively with the enthusiast oriented NVIDIA chip...

Regarding Stream Processors, ALUs or whatever anyone likes to name them, they're really not so comparable between two different architectures. Saying one of them has more but slower units and the other less but faster units is not really enough. They are different processing units, with different instruction sets, grouped in a different way and even with a different internal hierarchy, and they could perform very differently in one or other situation.

If you want to really compare shader processing power, you should look for synthetic shader-based benchmark results on both. On the other hand you can simply ignore shader processing power and focus on general processing power and real world benchmarks (games and so). Comparing Stream Processors count doesn't have a sense except between 2 chips that share the same architecture, of course (for example to compare differences between a HD4670 and a HD4870).

Finally, regarding what you should buy: HD4870 vs GTX260 is a hard one. Both are similarly featured videocards, and both are around the same price tag. ATi has the edge in performance : price, while NVIDIA has it in performance : power consumption. ATi offers DX10.1 support while NVIDIA offers CUDA support (which includes anything that uses CUDA to work, such as their new PhysX version), both very little used technologies (DX11 retrocompatibility may end up making useful the DX10.1 though, while CUDA support will be nothing more than anecdotal once we have OpenCL and/or DX11 drivers running). AFAIK, ATi drivers are more regularly updated (in a periodic monthly base to be exact) and any necesary fixes or optimizations are done really fast, BUT they don't use to pay attention enough to alternative platforms such as OpenGL and Linux, so I would give the drivers point to ATi if using the usual Windows/DirectX platform, and to NVIDIA otherwise...

Anyway, I'd clearly wait at least to the HD4890 (will be selling in 2 days, but I think they don't allow reviews until 6th) to see what it exactly brings and how prices are going to stay. It seems to be like a new HD4870 with the ability to reach way higher clocks than old one, so depending on pricing it could be the one to get...

[mr.max-amd]

what will do better in eye candy games like crysis?

[Farinorco]

what will do better in eye candy games like crysis?

Isn't it all about eye candy when talking about videogames graphics? I don't get what you mean here...

If you want to know how the game Crysis (specifically) runs in each videocard you can search for reviews, most of them include Crysis as one of their real world benches... last time I did Crysis ran more or less the same in a HD4870 than in a GTX260.

[mr.max-amd]

didnt know im more of a cpu guy
also would having a gig of memory benefit at higher resoloutions...with it im probably getting a 26" asus monitor
and iv heard memory has alot to do with performance at higher resoloutions
also whats the power draw difference from a overclocked 9800gt to a gtx 260 at about 650 core?

[largon]

I have no idea why you would think the wider the memory bus the better.
Increasing the bus width is nothing more than one of the several ways to increase available video memory bandwidth. And since 8×32bit bus of GDDR5 is more than enough for HD48-series there is no reason to make the card more expensive than what it is by going higher than that, considering higher bandwidths would yield very little in performance.

[Rattle]

forget about the buss and forget about crysis. get yourself either a 4870 1gb or a gtx 260 216 and go play some real games !

[Farinorco]

Or a HD4890 if this (http://www.gamestar.de/hardware/tests/grafikkarten/pcie/1955014/test_radeon_hd_4890.html) is not a fake or otherwise flawed review...

[Neuromancer]

Or a HD4890 if this (http://www.gamestar.de/hardware/tests/grafikkarten/pcie/1955014/test_radeon_hd_4890.html) is not a fake or otherwise flawed review...

Interesting. 285 seems to perform better with AA than the 4890 (moot point for me since I am a high res CRT gamer and do not need AA), but makes me wonder, are they STILL using CSAA? (EDIT: nm I see they still are)