• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

nVidia/AMD Geforce/Crossfire Video/RAID Controllers

L

lightp2

Gawd
Joined
Oct 28, 2009
Messages
672
Moderator please move this to Data Storage Forum if you felt it is more appropriate. I am sorry this is a rehashed idea and no new suggestion.

I do not need any credits. I welcome and urge anybody with direct contact to make similar suggestion to nVidia or AMD if they are able to execute it.

1. By today, in generic term we are already aware nVidia does not have a majority chipset business anymore. It has ION chipset and mostly entry-level AMD chipset still in production in case OEMs want them. It also has Tegra-line for mobile.(which is looking good, I would definitely appreciate a Tegra2 long battery-hour tablet if very affordable, I am hoping there's no production problem on this one)

2. It implies that for now, pc chipset is less of a concern to them. It also implies revenue is much less.

3. To create new revenue,

3a. Since NVIDA pc chipset division is gone. NVRAID is no longer a concern.

3b. nVidia please consider this.
3b1 - Even the cheapest entry level VGA card now has 512MB RAM. Many have 1GB.
3b2 - nVidia has good software expertise and 3rd party interface with software development groups over many segments.
3b3 - We are talking about CUDA/OpenCL current and Fermi-derived future cards.
3b4 - Please dedicate your best team and write an CUDA/OpenCL-enhanced new software RAID where processing can be offload to Display Card.
3b5 - Every nVidia VGA Card now instantly become RAID Controller with 512MB cache. Just take a look at real hardware RAID controller to understand how expensive it can be.

4. If nVidia must protect its new source of income.
4a. Build a custom Tegra/Geforce integrated PCI-e RAID controller. Tegra controls the generic processing code and Geforce offload the CUDA-based processing. Now I must say I am not software person and do not know how difficult it could be. Expert please suggest yourself to nVidia.
4b. If nVidia does not want to deal with SATA interface at all, the a hybrid system can be devised.
4b1 - AMD/VIA/anybody can build a 10-SATA port custom premium motherboard. Why? Since a lot of people can buy even more expensive controller, buying a 10-12 port premium motherboard should be possible. So SATA port will be addressed by other vendors.
4b2 - You can also use ADD-In many-ports SATA non-RAID card to provide ports, but they are usually equally expensive.
4b3 - The Motherboard SATA ports provide physical connection for disk, and after that nVidia Tegra/Geforce RAID controller will take over. You must remember this special RAID card is talking over PCIe x16 ports with 512MB-1GB local VRAM with hundreds or thousands of shaders/core so resource is not a concern for it. If it ever works, I cannot imagine the speed and beauty of a GTX285-powered, (with battery-protected CACHE) 1GB-2GB RAID controller. With that much cache and processing power, a lot of disks can be grouped under it with superb performance
4b4 - You can contemplate the SLI powered storage powerhouse. :) The same idea works equally well for Crossfire NAS.

5. Some users complain nVidia chipset is crap and it has history of abandoned chipset software features so they are not willing to try their luck on nVidia.
5a. To provide good will to them, nVidia please hire one or more reputable Linux Kernel developers to build a common interface/RAID software to the custom nVidia TegraGeforce RAID controller and open source it. The users will be happy to have such a useful piece of hardware and have Linux take care of the software problem. I am neutral so Windows/BSD/Solaris/AIX/VMware all can benefits from this if somebody is willing to invest resource on them.

6. Off-course you may not need to do all of these if Linux DMRAID and Linux MD software experts developed CUDA optimized software raid for RAID5/6. However, now it could be a problem for all the performance seeking Hardware RAID controller kits since instantly 10s of millions of low-cost high performance RAID controllers suddenly appears on scene without warning.

There could be many more detail, knowledgable people can fill in the blank.

All the ideas are yours, appreciate your life.

Cheers
 
Not all calculations are ideal for GPU processing. I honestly don't know if the parity calculations used for RAID fall into this category or not. However, it's not just a magical speed increase by moving something from CPU to GPU. If that were the case, you'd have an "Nvidia Nside" badge over the old "Intel Inside" one.

Second, using the GPU for processing uses up some of its resources. I run the dnetc client on my CPU and both GPUs, and I can say for a fact that running a very intensive CUDA core actually makes the plain desktop slow. Running dnetc and a game simultaneously, each one basically gets half of the processing power. I don't think most gamers would want to give up half their GPU to keep their RAID array going, and I'm not sure non-gamers would want to buy an Nvidia GPU to do RAID.

Since Nvidia already has experience with GPUs, chipsets, CUDA, etc., it seems like they would've already run into this if it's a viable option. Not that they're the only ones who can come up with good ideas, it just seems like this is taking some very of their very common products and combining them.

While I'd love to get great RAID performance from a $50 video card I have sitting around, I don't think I'd buy a $300 video card using all sorts of clever trickery to avoid putting a proven $400 RAID card in my server for critical data (which I'm pretty sure is a much bigger market than enthusiast storage).
 
this is a short generic thinking as I am defintely not up to task on technical details.

Assuming this concept can work

1. About worry of performance drain.

Reply : Most Storage Hardware Controller let you decide how much cache ram percentage for read and write. example 75% read 25% write. I assume same idea can also apply where you specify ratio for ram and processing power on the nVidia custom TegraGeForce card. If you can adjust this dynamically it is even better! When graphic heavy session you tell the card 80%GPU/20%RAID. Since a Tegra is on-board maybe it can do more work and let geforce concentrates on graphic processing. (remember OpenCL dynamic flow between CPU and GPU) If dual-Core Tegra 2 then you have even more processing power avaialble.

2. Intel/AMD can implement the same. it is just that since chipset business is still available and quad/hexa/octo/... CPUs are functioning, there little need for such concept.

3. nVidia has nothing to lose since majority chipset is gone. It has bunch of knowhow but no where to practice. To me, might as well put all the knowledge into such attempt. TBH, it does look like it possesses all the necessary techs except the argument about GPGPU suitability for such task. However, we must not forget Tegra is a more general purpose processor. Since some people express question about its market outlook, if it really a problem, nVidia might as well use them for something else. It's between no revenue or some revenue.

4. Finally, the way I look at it since many Intel/AMD motherboards have fakeraid function, it is pointless to redo the same with slight enhancement. The most sensible option will be a high-power dedicated Video/RAID that target enthusiast and enterprise.(especially for users who complaint entry-level HW RAID-6 taking days to rebuilt 2TB hard disk, considering we going to see bigger capacity disk down the road) However, Linux MD shows that it can beat out many hardware cards using software raid so we could be surprised as well if someone is successful in building such gpu-enhanced solution.

Cheers
 
one minor note

A. borrow concept from lenovo U1

1. This custom TegraGeforce Video/RAID controller normally provide GPU/RAID functions.

2. (If this is possible, I am not sure) During boot-up, press the correct key and you boot into the Tegra On-Board environment, together with all the custom software environment (I think mostly Linux-based/or Windows-CE) with access to 512MB or 1GB Geforce GTX backend.) imagine it as a glorified Tegra tablet, if you insists on light gaming I think the geforce portion should be up to task as well.
 
You must log in or register to reply here.
Back
Top