Sapphire Shows Us its Radeon HD 7970-based Vapor-X 6GB
By , - Source: Tom's Hardware IT
|
41 comments
Better than reference, says Sapphire.
At Computex, Sapphire shows us its Radeon HD 7970-based Vapor-X 6GB, which gets it name from its dual-fan vapor chamber cooling. It has an eight phase VRM for the GPU and is made from a 12 layer PCB and Black Diamond Chokes, which allow the card to run cooler and more efficient.
apphire claims that its product has improved stability, more overclock room and lower peak temperature compared to the reference model. This, of course, makes room for a button for instant overclocking to 1100 MHz for the GPU and 6000 MHz for the memory.
Discuss
Ask a Category Expert




You'd have 24GB of VRAM total, but each GPU can still only use its own VRAM. Graphics cards can't share data on a high enough bandwidth connection in order to share memory capacity between the GPUs and CF/SLI can't change that. They would need to be able to talk to either the other GPU or the other GPU's memory at an equal speed to their own and this is almost definitely not something that would work well in if tried.
Theoretically, AMD/Nvidia could do this on dual GPU cards where they could link the two GPUs to each other and/or to all of the memory on the card with very high bandwidth and minimal latency, but chances are that this would require a modified or completely new RAM interface and I think that it might need double the input/output ports to reduce the chance of a two GPUs fighting each other over RAM access. This would mean more complex and expensive RAM chips would be needed and the memory controllers might need to be modified. However, the card would not need each GPU to have its own huge reserve of memory, so it could be done cheaper if fewer of these chips were used than would be used by a more conventional dual GPU card today.
Regardless, GPUs on separate cards would need some sort of breakthrough to be able to share memory at high speed enough bandwidth and with acceptable latency. The latency might be an almost easy thing to do because GPUs don't seem to be the most latency limited devices, but the bandwidth problem doesn't seem like an easy thing to solve cheaply.
I'm pretty sure that here in the first word of second paragraph, there should be a capitalized letter s before apphire.
No this is Vapor-X. Vapor-X is a cooling design, think heatpipe but with a chamber right above the GPU core. Toxic is normally their overclocked edition, the top one at that as they have one called the Dual-X (I have it) that has a small overclock and a second BIOS overclock to 1000MHz.
The Toxic tends to be clocked higher as well as overclock better.
6GB of Vram to play Max Payne 3!
You'd have 24GB of VRAM total, but each GPU can still only use its own VRAM. Graphics cards can't share data on a high enough bandwidth connection in order to share memory capacity between the GPUs and CF/SLI can't change that. They would need to be able to talk to either the other GPU or the other GPU's memory at an equal speed to their own and this is almost definitely not something that would work well in if tried.
Theoretically, AMD/Nvidia could do this on dual GPU cards where they could link the two GPUs to each other and/or to all of the memory on the card with very high bandwidth and minimal latency, but chances are that this would require a modified or completely new RAM interface and I think that it might need double the input/output ports to reduce the chance of a two GPUs fighting each other over RAM access. This would mean more complex and expensive RAM chips would be needed and the memory controllers might need to be modified. However, the card would not need each GPU to have its own huge reserve of memory, so it could be done cheaper if fewer of these chips were used than would be used by a more conventional dual GPU card today.
Regardless, GPUs on separate cards would need some sort of breakthrough to be able to share memory at high speed enough bandwidth and with acceptable latency. The latency might be an almost easy thing to do because GPUs don't seem to be the most latency limited devices, but the bandwidth problem doesn't seem like an easy thing to solve cheaply.
The problem is that cards can't use each others VRAM because each card is producing their own frame. A quad-crossfire with four of 6 GB VRAM will still only be recognized as a 6 GB of VRAM, but paired with four 7970 GPUs.
Now if GPUs could use each others' VRAM, that would require significantly upgraded CF/SLI and memory bandwidth (GDDR 6 1024+ bit) to handle the data going between the cards.
The result? Any cost reductions from reduced VRAM would be overshadowed by the need of greater bandwidths.
Gigabyte's new GTX 680, this Sapphire Radeon 7970... Are the graphics card manufacturers trying to tell us something like like make sure that your case has excellent air flow or else? They don't seem to interested in getting the hot air out of the case with these supposedly top-end models. Granted, people buying these would probably have incredible cases with great air flow, but still, it's a little worrisome.
I had a dream of using two 100mm centrifugal fans on a GPU with a reference heatsink, one end is blocked off with a sheet of aluminum. The fans' casing are also shaped to avoid air from spilling out of the heatsink.
The first fan blows air into the heatsink, and the second fan pulls the air out of the heatsink and blows it out of the case.
Although turbulence within the heatsink would hurt efficiency of the fans, at least none of the hot air is spilling into the case.
At this rate I won't be surprise when video card reach crazy VRam amounts... not like 6GB VRam is already a lot.
I think you are mostly right. They seem to be solely focused on moving the heat from the card, regardless of what effect it has on the rest of the case. But it's a lot easier and a lot cheaper nowadays to get a decent airflow case. There are a lot more cases at the mid-ATX form factor that have both side and top fan mounts that can be used for exhaust, and front fans for intakes, as opposed to what used to be just a box of metal with one or two fans. They're just playing to a changing reality, I think.
I'm pretty sure that a video card's usage of VRAM is not limited by the maximum addressable RAM capacity of the operating system running on the CPU of the computer except maybe for system RAM shared between the GPU(s) and CPU.