WE ARE FIVE days from launch of Graphzilla's most important line of products for 2007. The 8800 series may have been a success with quite a large stain (Vista drivers blunder), but the mainstream and low-end product lines are where volume, and thus revenue is made. This is the first article that discloses details about 8300, so read on.
For starters, we have to congratulate Graphzilla’s engineers, who really outdid themselves by creating happy, clockable chips. We are hitting clocks of well above 800MHz with bog-standard cooling, which speaks of the phenomenal performance of TSMC’s 80nm process.
Thanks to the extreme overclockability of the G84 and G86 chips, channel partners abandoned Nvidia's own reference designs and pushed forward with their own custom products, clocking the GPUs well above reference spec. We also got final details about the GPUs themselves, so we’ll just give you a snip at the final parts, as specified by Nvidia. Bear in mind that partners are tweaking things a lot in a bid to differentiate themselves as much as possible, but a lot of first and second tier AIBs will come to market with products of the following spec.
G84-400/403/405 a.k.a. GeForce 8600GTS comes at a default clock of 675MHz for the GPU and 1GHz DDR (2GHz) for GDDR-3 memory. The GPU consists of 289 million transistors, putting it right in G71, or 7900 territory. These 289 million transistors are spent on 32 Scalar Shader units clocked at 1.45GHz each. Power draw with default clock should not exceed 71 Watts, but theoretical headroom is 125-150W (45-75W from PCIe slot, 75W from 6-pin PCIe power connector), hence the overclocking competition between the companies.
G84-300/303/305 a.k.a. GeForce 8600GT comes with a near-identical GPU, but the clock is different. Thirty-two Scalar Shader units work at 1.18GHz, while the rest of the GPU works at 500/550/600 MHz. Video memory is GDDR-3 and it will be clocked at 700MHz DDR, (1.4 GHz). Power consumption is 43 Watts, only three Watts more than 8500GT (but overclocking is limited by the stable power your motherboard can provide).
G86-300/303/305 a.k.a. GeForce 8500GT is basically half a G84 chip. This el’cheapola packs only 16 Scalar Shaders in 210 million trannies, which does not bode well for performance in shader-intensive applications. The market will decide is this enough. Shader units are clocked at 900MHz, while the rest of the chip is set to work at either 450, 500 or even 600MHz, with this last being the most used option. Video memory is your typical DDR2 system memory, regular PC2-6400 (DDR2-800) memory chips.
G86-303/305 a.k.a. GeForce 8300GTis the only product not launching on April 17th. This product is set for introduction after AMD launches its R(V)630 chip, and it is the same chip as one on the 8500GT board. So the GeForce 8300 will pack 16 Scalar shaders as well (unless nV disables them in drivers, but hardware-wise we are talking about the same chip). But, this chip will be placed on 7300 PCBs, meaning we have a case of castration. Yes, the 8300GT and GS will feature a 64-bit memory interface, so you can guess what the performance will be. This baby is packed with either 128MB or 256MB of DDR2 memory, again PC2-6400 or something slower. Power envelope is around 35W, but this depends on the final clock. Nvidia did not set the clock yet, because it awaits performance results from AMD’s own chips.
It will be interesting to see how Nvidia's simplistic, but highly effective scalar design cope with less, but more complex vector units hidden inside R610/630/600/650 chips.
For overclockers, there is little or no doubt at all. 8600GTS can be overclocked using the shipping single-slot cooler to 750-775MHz with no problems at all, putting better air or even water-cooling and eventually a voltage-mod will result in clockspeeds of over 1GHz for the core. GDDR-3 memory performs brilliantly, since you can easily run it at 2.2 or even 2.35GHz, ending up with bandwidth of 37.5GB/s.
News source: THEINQUIRER
For starters, we have to congratulate Graphzilla’s engineers, who really outdid themselves by creating happy, clockable chips. We are hitting clocks of well above 800MHz with bog-standard cooling, which speaks of the phenomenal performance of TSMC’s 80nm process.
Thanks to the extreme overclockability of the G84 and G86 chips, channel partners abandoned Nvidia's own reference designs and pushed forward with their own custom products, clocking the GPUs well above reference spec. We also got final details about the GPUs themselves, so we’ll just give you a snip at the final parts, as specified by Nvidia. Bear in mind that partners are tweaking things a lot in a bid to differentiate themselves as much as possible, but a lot of first and second tier AIBs will come to market with products of the following spec.
G84-400/403/405 a.k.a. GeForce 8600GTS comes at a default clock of 675MHz for the GPU and 1GHz DDR (2GHz) for GDDR-3 memory. The GPU consists of 289 million transistors, putting it right in G71, or 7900 territory. These 289 million transistors are spent on 32 Scalar Shader units clocked at 1.45GHz each. Power draw with default clock should not exceed 71 Watts, but theoretical headroom is 125-150W (45-75W from PCIe slot, 75W from 6-pin PCIe power connector), hence the overclocking competition between the companies.
G84-300/303/305 a.k.a. GeForce 8600GT comes with a near-identical GPU, but the clock is different. Thirty-two Scalar Shader units work at 1.18GHz, while the rest of the GPU works at 500/550/600 MHz. Video memory is GDDR-3 and it will be clocked at 700MHz DDR, (1.4 GHz). Power consumption is 43 Watts, only three Watts more than 8500GT (but overclocking is limited by the stable power your motherboard can provide).
G86-300/303/305 a.k.a. GeForce 8500GT is basically half a G84 chip. This el’cheapola packs only 16 Scalar Shaders in 210 million trannies, which does not bode well for performance in shader-intensive applications. The market will decide is this enough. Shader units are clocked at 900MHz, while the rest of the chip is set to work at either 450, 500 or even 600MHz, with this last being the most used option. Video memory is your typical DDR2 system memory, regular PC2-6400 (DDR2-800) memory chips.
G86-303/305 a.k.a. GeForce 8300GTis the only product not launching on April 17th. This product is set for introduction after AMD launches its R(V)630 chip, and it is the same chip as one on the 8500GT board. So the GeForce 8300 will pack 16 Scalar shaders as well (unless nV disables them in drivers, but hardware-wise we are talking about the same chip). But, this chip will be placed on 7300 PCBs, meaning we have a case of castration. Yes, the 8300GT and GS will feature a 64-bit memory interface, so you can guess what the performance will be. This baby is packed with either 128MB or 256MB of DDR2 memory, again PC2-6400 or something slower. Power envelope is around 35W, but this depends on the final clock. Nvidia did not set the clock yet, because it awaits performance results from AMD’s own chips.
It will be interesting to see how Nvidia's simplistic, but highly effective scalar design cope with less, but more complex vector units hidden inside R610/630/600/650 chips.
For overclockers, there is little or no doubt at all. 8600GTS can be overclocked using the shipping single-slot cooler to 750-775MHz with no problems at all, putting better air or even water-cooling and eventually a voltage-mod will result in clockspeeds of over 1GHz for the core. GDDR-3 memory performs brilliantly, since you can easily run it at 2.2 or even 2.35GHz, ending up with bandwidth of 37.5GB/s.
News source: THEINQUIRER
