Indigo Renderer Forum

galinette wrote:Oops, I just bought a X4 965 two month ago

since its quite new, throw it on eBay and buy the 6 Core.... you should get a good price for your "old" CPU

Well. I finally started building a new machine, something I've never ever done before, and to kick things off got an Antec CP-850 PSU and an Antec 183 case. Once the local stores get more GTX480s in stock, I'm dropping a lot of cash for one. Not sure what I'll be doing for a motherboard (suggestions more than welcome!) but I'm pretty set on getting an i7 920 or 930, I think. Might as well invest in the higher end stuff if this is what will feed me for the next while. That stuff will come later, I think. RAM for this thing will be pricey, so building this will take a while and several paychecks. Maybe I'll have it ready in time for GPU Indigo!

The Antec CP-850 DID look like it was going to run me over when I pulled it out of the box. It looks big, bad and very dangerous. If this thing doesn't work out and I can't pay rent, I can always move into the Antec case and live there, there's plenty of room. Maybe even get a roommate.

EDIT: But, relative to the topic, which is why I posted in the first place, the point of this was actually to wonder how an AMD or Intel 6-core monster would fit into that kind of set up? I imagine you need a pretty non-standard motherboard...

I'd strongly suggest you get a 470 instead if you absolutely must have one now... new Fermi-based GeForce cards are right around the corner, they're working madly to fix the manufacturing problems that made the current ones so slow (relative to how it should be) and hot.

lycium wrote:I'd strongly suggest you get a 470 instead if you absolutely must have one now... new Fermi-based GeForce cards are right around the corner, they're working madly to fix the manufacturing problems that made the current ones so slow (relative to how it should be) and hot.

You think we'll see some new ones released soon? I can afford to wait, because I won't get this system up and running until I get a CPU, mobo and good (but expensive...) RAM, so I can leave the card for last, I suppose. Any link to news about these new ones?

Anyway, you're the boss when it comes to this, tell me what's best!

right now it's all rumours, but here's what i've heard:

http://www.fudzilla.com/content/view/18624/1/ (geforce 460, cheaper but still a proper fermi chip)
http://www.fudzilla.com/content/view/18608/1/ (castrated cache, probably not good for gpgpu)

i'd say hold on until we've tested our super improved opencl kernel with radeon 5xxx series; it's looking like cuda will be dropped actually!

lycium wrote:it's looking like cuda will be dropped actually!

Well, developing the same stuff in 2 different "languages" doesn't make to much sense anyway...
Just imagine the development time, and debugging for openCL AND Cuda versions ^^

Double precision performance of the consumer fermi cards reduced to 168Gflops
In TESLA (same chip gf100) perfomance non reduced (515Gflops).
maybe, today ATI is best choise. (5870 ~ 500Gflops)
http://www.geeks3d.com/20100329/geforce ... rformance/

Attaining AMD's peak FLOPs rate is more difficult than with nVidia because their shader cores are individually SIMD rather than scalar; at least the situation is better than we have on x86 with SSE because GPU SIMD is way more flexible: we have swizzling, dot products, cross products, vector normalisations, etc in the ISA.

Also, we're not currently using double precision in our GPU acceleration kernels, so GeForces 4xx cards will still kick ass

StompinTom wrote:Well. I finally started building a new machine, something I've never ever done before, and to kick things off got an Antec CP-850 PSU and an Antec 183 case. Once the local stores get more GTX480s in stock, I'm dropping a lot of cash for one. Not sure what I'll be doing for a motherboard (suggestions more than welcome!) but I'm pretty set on getting an i7 920 or 930, I think. Might as well invest in the higher end stuff if this is what will feed me for the next while. That stuff will come later, I think. RAM for this thing will be pricey, so building this will take a while and several paychecks. Maybe I'll have it ready in time for GPU Indigo!

The Antec CP-850 DID look like it was going to run me over when I pulled it out of the box. It looks big, bad and very dangerous. If this thing doesn't work out and I can't pay rent, I can always move into the Antec case and live there, there's plenty of room. Maybe even get a roommate.

EDIT: But, relative to the topic, which is why I posted in the first place, the point of this was actually to wonder how an AMD or Intel 6-core monster would fit into that kind of set up? I imagine you need a pretty non-standard motherboard...

Nice choice on case and powersupply. They go well together are are really quiet, I'm getting the same ones eventually

MB wise, those chips fit the standard Intel and AMD wise for their respective socket type.
I would go for a 930 over the 920. It can overclock higher and tends to be cheaper, unless you go looking 2nd hand.

---

Cuda will die just like Glide did to OpenGL, and then D3D took over.

Won't be long till Microsoft's version of Cuda comes along and kills off both

There is an article about AMD Fusion and discrete GPUs on AMD blog:
http://blogs.amd.com/play/2010/08/06/am ... usion-apu/

It says:
"So, to finally answer the question whether discrete GPUs will die, the answer is: Hell No.

AMD relies on our ability to generate world-class discrete graphics cores and then leverages them in our APU product lines. You should also expect AMD to innovate when an AMD discrete graphics is plugged into an AMD APU, the mythical word of “synergy” will be realized. We are expecting to improve both graphics and compute performance."

I just didn't understand this "AMD discrete graphics is plugged into an AMD APU" part
Does this mean APU + GPU instead of CPU + GPU?
Or may be this "Hell No" means that discrete GPUs will continue "to live" inside APUs?

Edit: Never mind. I've understood it by myself. "Plugged into" is just a figure of speech to describe APU, it's not something else instead of PCI Express card for discrete GPU.

Apparently (based on Google search), AMD has something called Hypertransport interconnect technology associated with HTX connector (vs PCIe).
http://en.wikipedia.org/wiki/HyperTransport
http://www.hypertransport.org/docs/pres ... -22-06.pdf

And this HTX connectivity "Eliminates Bandwidth Bottlenecks of Speed-Critical, Compute-Intensive,Server Clustering and Co-Processing Applications".

Could it be the answer to my own question?

Just to correct myself, PCI-e 3.O is actually going to support HyperTransport protocol too:

http://www.xbitlabs.com/news/other/disp ... _2011.html

"In 2008 Advanced Micro Devices and Hewlett Packard proposed a number of extensions for PCI Express 3.0. One of the extensions is protocol multiplexing, a feature that would allow chips to dynamically switch between seven different protocols in addition to PCIe using the shared set of pins. This would allow creation of chips that would be compatible with PCIe, HyperTransport, QuickPath Interconnect, Ethernet and other buses at the same time. Another extension is called lightweight notification and would allow co-processors or peripheral chips to talk to each other through system memory using a PCIe transaction without interrupting a host processor."

The big win for integrating GPU into CPU is the data transfer!
The CPU can feed the GPU directly without need to transfer over PCI-E lanes.
This should result in a 100% GPU load if used for RayTracing...

Hello all, sorry for reviving this old topic, but I have one question.

Has anyone tried render using AMD A series CPUS with Indigo using OpenCL (GPU render)? Does it work? Or maybe it doesn't support OpenCL in AMD APUS?

Why I am asking? I want to build small renderfarm using few pcs with APUS.

Thanks in advance