How are IBM capabilities (Research wise) compared to Samsung and Intel?
The AMD Execution Thread [2007 - 2017]
- Thread starter overclocked_enthusiasm
- Start date
-
- Tags
- amd
- Status
FrameBuffer
Banned
lat I recall.. 2nd to none
IBM seems to be as good...possibly better on fundamental research. Probably weaker on the development side, and definitely weaker on anything involving high volume.
But the issue is that R&D for low volume products sold in million dollar servers can be very different than for high volume products sold into hundred dollar desktops. For example, there was an article about how IBM's partners were trying to force IBM to abandon gate-first for HKMG and instead focus on gate-last (like TSMC, Intel and everyone else).
As another example, IBM's eDRAM cache in POWER7 is impressive...but it may not be feasible for high volume, high performance products like Shanghai/Istanbul because of the increase in cost (due to extra manufacturing steps).
As a last example, consider SOI. It increases the wafer cost, but gives you a slight edge in performance...but not that much. Maybe 5-10%, and I hear it's more like 7%. Nobody uses it for really cost sensitive products like mobile phones, embedded controllers, etc. It's only used on high performance microprocessors, and it's not even clear that it makes sense for really high volume CPUs. If you assume SOI gives you a 7% performance benefit and adds 7% wafer cost, you might be better off just spending more money and time on design (which would be a fixed cost rather than a variable cost like more expensive wafers) to achieve that 7% perf. gain and using bulk wafers.
The real point to take away is that as a rough approximation, Intel and IBM seem to be on par. But IBM has very little experience with high volume manufacturing, while Intel has been doing that for 2-3 decades. And the scale difference is big. IBM is struggling to keep one fab on node N full. Intel has 3 full fabs on node N and 3 full fabs on node (N-1). Quantity makes a big difference in how you optimize.
David
DK, your are right about SOI on cost sensitive applications but I don't think cell phones fall in this category, not anymore, now that all cell phones are smart phones performance matter and it's hard to achieve 7% more performance with 7% more area (oh, and power, this is important), if SOI does that, fine, a better IPhone (or IPhone-killer), also their chips are already very small and the final product is not so chip, there is any point on spending $2 on the processor and $100 on the chip?
At least ARM thinks there is such a makert: http://www.arm.com/products/physicalip/silicon-on-insulator.html
At least ARM thinks there is such a makert: http://www.arm.com/products/physicalip/silicon-on-insulator.html
They say that the benefits of SOI decrease with better manufacturing process, but this article is claiming 20% higher speed with SOI at 45nm.
The silicon results show that 45nm high-performance SOI technology can provide up to 40 percent power savings and a 7 percent circuit area reduction compared to bulk CMOS low-power technology, operating at the same speed. This same implementation also demonstrated 20 percent higher operating frequency capability over bulk while saving 30 percent in total power in specific test applications.
The silicon results show that 45nm high-performance SOI technology can provide up to 40 percent power savings and a 7 percent circuit area reduction compared to bulk CMOS low-power technology, operating at the same speed. This same implementation also demonstrated 20 percent higher operating frequency capability over bulk while saving 30 percent in total power in specific test applications.
That was basically a PR stunt. I don't think ARM expects many customers to use SOI.
David
aaronspink
Veteran
Lets see, crap foundry process vs IBM's top end process. Its report is vapid fabrication at best. It holds about as much weight as me claiming that my unobtaining technology can decrease power by 10000000% while increasing performance by 101413453%.
Thanks for the answer.
If I may, How come IBM manages to pay for a research for the next generation "Node", build a Top Class factory with being one of the smallest Semi Conductor company?
Its Semi division is something 15 times smaller than Samsung.
The bulk of IBM's cash comes in on the services and software side of the systems its sells. If it weren't for that cash flow, I am not sure its microelectronics group would be self-sustaining.
3dilettante pretty much nailed it. There are some other issues though:
1. The US government probably subsidizes IBM. SOI is very useful for electronics in space, and the government doesn't want to fab chips outside of the US.
2. IBM has an incredibly lucrative client base (mainframes) that generate huge margins (which is the heart of what 3dilettante said).
3. IBM semiconductor does a lot of custom engineering work and other odd stuff...think about BlueGene, PS3, etc. That revenue (and some of the good PR) can be used to justify having a fab.
4. IBM products use a lot of chips that are fabbed by IBM. That may not show up as revenue for IBM semi.
5. IBM licenses their technology to others, and it's not clear how that is accounted for. For instance, I suspect most IBM process tech partners pay IBM money...but that may be treated as licensing, rather than as revenue for semiconductors.
6. Pride, and the fact that IBM isn't losing money.
David
iirc I heard that bulldozer will include them.
itsmydamnation
Veteran
i remember reading it will have AES instructions as well, so if theres nothing apart from AVX between sandy bridge and westmere then for once AMD could be at the same place as intel in terms of supported instruction sets when bulldozer is released.
Assuming of course that ppl won't use Intel compiler or Intel removes vendor check from it, which disables (most?) extensions on non-Intel CPUs
No text in the article section of that page for me.
- Status
