As an Electronics Technician, computer/gaming nerd and amateur physicist when I first heard about Larrabee I was extremely skeptical, but mostly because of misinformation of the Larrabee GPU being mode of entire CPU cores.
After reading all of the current relevent information I did a 180 in my opinion and now think it has incredible potential.
Untested? Yes and No. It isn't out there yet but the technology it's based on is well understood and games have already been modded to run on it. They have a pretty good idea of how it will scale.
Scalable? Yes. Very.
How is coding? Writing code for this is easy. In fact, it's the biggest draw. This is very important as it not only brings down game development cost but it can decrease the time to market.
I was recently thinking, how would Intel bring Larrabee to market? It's supposed to be fully compatible with games but has to be specifically coded to for full optimization. Coming out on the XBox would be a perfect way to transition onto the desktop and servers.
In my humble opinion, Intel has the clout to get Larrabee working. The ease of coding will greatly assist its adoption and the ONLY cause concern is power efficiency.
I really like the fact that features that normally are done in hardware can be done on the Larrabee in software. This might seem like a step backwards but if it can be done efficiently it isn't. Physics encoding? Yes. Transcoding? Yes.
SLI and Crossfire. Adding a 2nd card in SLI or Crossfire mode for more gaming power can be hit and miss. The game has to be specifically coded for it as well as having driver support. Larrabee on the other hand isn't quite the same. We can get 2X the performance with two cards. There's plenty of room for discussion here but the bottom line is games need to get away from SLI type coding and code so that GPGPU's like Larrabee's design can scale almost perfectly by adding more cores.
If Intel can get Larrabee on the XBox I think it has a strong chance of dominating the PC GPU market.
(Just to be clear, and it was mentioned before, Larrabee is a GPU. A traditional CPU will be in a console or desktop.)
How is the coding going to be easy? Intel haven't released an SDK for this yet, so there is no way to tell. Programmers today have a hard enough job effectiovely using two or four processors, yet alone a 32 core GPGPU. Just because it is an x86 instruction set, doesn't automatically mean it will be easy to program. Going massively parallel does seem to scale well, acording to Intel's figures, but the associated coding is by no means easy. Have a chat to a PS3 developer and ask about how hard it is to program for a Cell processor and it's 7 SPEs.
Offloading hardware specific features into software is an interesting way of doing things, and it is clear that Intel intend the Larrabee to be more than just a GPU. However, this extra layer of software translation comes at a significant performance hit, and adds another layer of complexity to coding.
As far as SLI and Crossfire, the performance would not double. SLI or Crossfire does not allow the Larrabee chips to share their coherent cache. It is that coherent cache that is a key part to the scalability of the cores on the Larrabee die. Because the chips do not share that high speed cache (and associated interconnects), and because there is some overhead to both SLI and Crossfire, the performance will not simply double.
Now that we've thrown 'em off the trail, use the form below to get in touch with the people at Engadget. Please fill in all of the required fields because they're required.
Larrabee in next XBOX? Good chance it's YES.
As an Electronics Technician, computer/gaming nerd and amateur physicist when I first heard about Larrabee I was extremely skeptical, but mostly because of misinformation of the Larrabee GPU being mode of entire CPU cores.
After reading all of the current relevent information I did a 180 in my opinion and now think it has incredible potential.
Untested? Yes and No. It isn't out there yet but the technology it's based on is well understood and games have already been modded to run on it. They have a pretty good idea of how it will scale.
Scalable? Yes. Very.
How is coding?
Writing code for this is easy. In fact, it's the biggest draw. This is very important as it not only brings down game development cost but it can decrease the time to market.
I was recently thinking, how would Intel bring Larrabee to market? It's supposed to be fully compatible with games but has to be specifically coded to for full optimization. Coming out on the XBox would be a perfect way to transition onto the desktop and servers.
In my humble opinion, Intel has the clout to get Larrabee working. The ease of coding will greatly assist its adoption and the ONLY cause concern is power efficiency.
I really like the fact that features that normally are done in hardware can be done on the Larrabee in software. This might seem like a step backwards but if it can be done efficiently it isn't. Physics encoding? Yes. Transcoding? Yes.
SLI and Crossfire.
Adding a 2nd card in SLI or Crossfire mode for more gaming power can be hit and miss. The game has to be specifically coded for it as well as having driver support. Larrabee on the other hand isn't quite the same. We can get 2X the performance with two cards. There's plenty of room for discussion here but the bottom line is games need to get away from SLI type coding and code so that GPGPU's like Larrabee's design can scale almost perfectly by adding more cores.
If Intel can get Larrabee on the XBox I think it has a strong chance of dominating the PC GPU market.
(Just to be clear, and it was mentioned before, Larrabee is a GPU. A traditional CPU will be in a console or desktop.)
How is the coding going to be easy? Intel haven't released an SDK for this yet, so there is no way to tell. Programmers today have a hard enough job effectiovely using two or four processors, yet alone a 32 core GPGPU. Just because it is an x86 instruction set, doesn't automatically mean it will be easy to program. Going massively parallel does seem to scale well, acording to Intel's figures, but the associated coding is by no means easy. Have a chat to a PS3 developer and ask about how hard it is to program for a Cell processor and it's 7 SPEs.
Offloading hardware specific features into software is an interesting way of doing things, and it is clear that Intel intend the Larrabee to be more than just a GPU. However, this extra layer of software translation comes at a significant performance hit, and adds another layer of complexity to coding.
As far as SLI and Crossfire, the performance would not double. SLI or Crossfire does not allow the Larrabee chips to share their coherent cache. It is that coherent cache that is a key part to the scalability of the cores on the Larrabee die. Because the chips do not share that high speed cache (and associated interconnects), and because there is some overhead to both SLI and Crossfire, the performance will not simply double.