VR-Zone on G70 and R520

[bkswaney]

Next Gen GPUs Info : R520 & NV50/G70

ATi and NVIDIA are certainly working hard to push out their next generation GPUs for Computex launch in June. NVIDIA is particular tight-lipped about their NV50/G70 GPU even to their partners while ATi has not demo a working sample to their partners till now so things aren't looking too rosy. None the least, we heard that NV50/G70 is a dual-core monster with 32 pipelines and is based on TSMC 90nm process technology. ATi R520 is also 90nm based and contains 32 pipelines. The reference cooler design for NV50/G70 looks huge with 4 copper heat-pipes and a large fan. We have also heard about an interesting strategy that NVIDIA is going to equip their next generation cards ranging from low-end to high-end with 2nd gen. Turbo Cache technology.

[gram_vaz]

what's turbo cache exactly? i'm thinking it's the pci-e equilivant of agp aperture.

[jAkUp]

It uses system memory. A turbo cache card only has typicall 64mbs of memory, the other 64mb it uses is your system memory. The only advantage to this is price, I don't see why they would put this on high end card.

[bkswaney]

It uses system memory. A turbo cache card only has typicall 64mbs of memory, the other 64mb it uses is your system memory. The only advantage to this is price, I don't see why they would put this on high end card.


I'm at a loss on this one myself.

[wEEt]

Dual-Core is useless. And G70 is not NV50. 32 Pipelines are defenitely too much. :(

[retsam]

i dont know, maybe turbo cache is more then we think it is..maybe is has something more todo with dual core or sli then we think ....

[zoomy942]

Dual-Core is useless. And G70 is not NV50. 32 Pipelines are defenitely too much. :(


there's no such thing as too much :) :)

[Alaa]

when can we buy Nvidia 2nd generation?

[JoKeRr]

when can we buy Nvidia 2nd generation?

You want a TNT2?? :lol2:

[Alaa]

i want a G70!

[Banko]

You want a TNT2?? :lol2:
Well actually wouldn't be a TNT, since the Riva was technically NV's first gen card? :p

[grimreefer]

i want a G70!
i need one, still using a 5950.

[Red_Shift]

It uses system memory. A turbo cache card only has typicall 64mbs of memory, the other 64mb it uses is your system memory. The only advantage to this is price, I don't see why they would put this on high end card.
Turbo cache is not dependant on local memory, you can throw whatever amount of memory you want to the graphics card and still implement turbo cache tecnology. For example there are 256mb 6200 cards with turbo cache ( I know XFX have those).

Turbo cache can use way more than 64mb of system memory.
When turbo cache arrived on the 6200 I immediattly thought that it would be cool to implement it on high end cards, cause it can cut costs significantly and don't loose performance, turbo cache is meant to use system memory very quick, it does much quicker than agp aperture or something, maybe this 2nd gen turbo cache can do it almost as quick as using local vga memory. Imagine for example a high-end gpu on a card with 128mb of ddr3 (or somethig faster) and with turbo cache, maybe this way you would not be memory bottlenecked in most cases; this card should cost less than 350$ and have a great performance. 256mb cards w/ this 2nd gen turbo cache would be preferable to 512mb cards too.

[Kickus_assius]

I think dual core is a great idea. It makes the core be much cooler and although it's not as fast as the single core equivalent, it's still pretty fast. The dual core intel cpu's with 2x2GHz are nowhere near as fast as a single core 4 GHz, but the dual core cpu is much cheaper isn't it and runs at probably half the temperature and the speed is equal to what, like one single 3.4 GHz. I think it will work very well for graphics cores since heat is a major issue. The only thing is I think ATI will kick ass with their single core being 32 pipes and Nvidia's card will be like having two 16 pipe cards in SLI, wouldn't it (except with the full 16X bandwidth)?

[jolle]

It uses system memory. A turbo cache card only has typicall 64mbs of memory, the other 64mb it uses is your system memory. The only advantage to this is price, I don't see why they would put this on high end card.

Carmack and Sweeny has been pushing for Virtual RAM on graphics cards for a long time, I read somewhere long ago.. dunno if its still viable..
3Dlabs supported it on VP10 a few years ago, their implementation was for AGP and seemed to work well, not only for Textures but other things aswell..
TurboCache and Hypermemory seems to be similar but on a faster bus.
If used right it could prolly allow developers to put more textures in scenes, think it was something about drawing mipmaps from the textures in systems RAM, doesnt require too much bandwidth since you dont get the entire texture, but only the smaller mipmap..

With some good management it might work out nicely, keep the most bandwidth critical stuff onboard.. dunno..
You got 4gb/sec bi directional, so you could prolly substain like a 60mb stream @ 60 fps (if you need to send a 60 mb of textures to the card every frame that is).. if that counts for anything..

[Subtestube]

The big thing that makes using main memory viable on PCIe is that it's full duplex and it's fast. That's really about it. If you were clever about what you kept in on card memory at a given time (besides the ubiquitous framebuffer), I figure that Turbocache/Hypermemory could actually get you a LOT more effective memory.

[Raje]

You got 4gb/sec bi directional, so you could prolly substain like a 60mb stream @ 60 fps (if you need to send a 60 mb of textures to the card every frame that is).. if that counts for anything..

With 3Dc texture compression (and probably other lossless compression), whatever bandwidth the cards have becomes 4 times as much with 3Dc... which both ATI and NV (G70/G80 supposedly) are using now.

G70/G80 = 3Dc (http://www.theinquirer.net/?article=22734)

[Raje]

I think dual core is a great idea. It makes the core be much cooler and although it's not as fast as the single core equivalent, it's still pretty fast. The dual core intel cpu's with 2x2GHz are nowhere near as fast as a single core 4 GHz, but the dual core cpu is much cheaper isn't it and runs at probably half the temperature and the speed is equal to what, like one single 3.4 GHz. I think it will work very well for graphics cores since heat is a major issue. The only thing is I think ATI will kick ass with their single core being 32 pipes and Nvidia's card will be like having two 16 pipe cards in SLI, wouldn't it (except with the full 16X bandwidth)?

You cannot compare a Dual Core CPU with a Dual Core GPU. The reason is that Graphics are infinitely parallelable. You could have a GPU for each pixel and you could see a linear improvement in graphics processing power. There are of course limitations to this: bandwidth and memory for all the processing units, etc.

I'm confident that if Nvidia has chosen to go with two physical units for their next GPU they did it because they could better cool the two GPUs, they could scale clocks on the overall simpler GPUs higher than on one 32 pipe GPU, and because they will be able to blow away the 32 pipe single GPU as far as performance is concerned.

Another consideration for a dual GPU configuation is manufacturing. 2 smaller GPUs would have better yields than one larger GPU.

[bkswaney]

I hope the G70 is dual core. :D

All this dual core madness. :p

[MUYA]

Why not just stick 8 quad blocks anf stick 12 VS units or 6 quad blocks and 8 VS units and call it a refresh? ;)

[harl]

I hope the G70 is dual core. :D

All this dual core madness. :p

Dual core in grafics it's very diferent from dualcore CPU's
EJ:
Making a card with 32 pipes it's expensive, but rumors says
tha Nvidia will use two separate cores of 16 in the same chip
That makes the card cheaper not faster

[jolle]

Someone mentioned pipeline stalling somewhere..
and using a dual core GPU could lessen that, since all pipelines wouldnt stall when it happens, only half of them, those in the other GPU..
something about 2 exuecution engines (2x16) would have some benefit over one gpu with 32 pipelines...
dunno, read it somewhere and forgot where.. anyone got any idea if there is any truth in it?

[Domell]

Well, if G70 is dualcore it could be veery nice :P But is any different between one core 32-pipe GPU and dualcore 2x16-pipe GPU when we talk about performance?????...dualcore CPUs aren`t faster than singlecore CPUs in games because they need software support...

[nutball]

GPUs are already multi-threaded. Dual-core has no meaning for GPUs. In the sense that new CPUs are dual-core, GPUs have been dual-and-more-core for years. I don't see why this is such a challenging concept. The software support for multi-threaded GPUs already exists - - it's happening in front of your face as you read this post.

What people are talking about is G70 being either dual-die, or dual-chip. Not dual-core. Different things.

[SH64]

G70 to require more power ??

WE HEARD THAT Nvidia asked PCI-Sig, the PCI and PCIe standardisation body, to provide some more juice for Nvidia's next generation cards. It turns out that 75W from external connector and 75W from motherboard is not enough. Nvidia wants an additional 75W for its G70 card.

We also heard that Nvidia asked PCI-Sig to let it use two power connectors for the card, each providing 75W of power. If the Sig agrees, it seems that Nvidia next generation card might end up drying 225W of power. It's simple maths, 75 W from PCIe motherboard connector, and two times 75 from a six pin power connector.
http://www.theinquirer.net/?article=22917

bring on your 600W PSUs :p