Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
@Flum
You are just cherry picking what seems to support your statement, but you get the full picture in the article and I posted its conclusion. When it comes to actual circuits, like SRAMs, Globalfoundries 7nm is more dense than Intel 10nm. We can argue about a few nm pitch here and there, where one will win over the other, here and there. All this doesn't even remotely support your statement that Intel would have a lead. If anything, it is a wash and Intel will be late, so no lead.
And don't tell me about some 10nm chips that are out in the field by Intel. If that would indicate a mature process, they wouldn't delay volume production. It's all about volume production and that is going to be this year for the foundries on 7nm and next year for Intel 10nm. No lead for Intel here, nowhere and certainly no 3 years lead as often claimed by Intel.
Now you can stare at the chart and look for information about the future of this company, like those other monkeys.
We'll have to wait and see. I expect foundry 7nm to be denser in actual designs. In any case, Intel will be late with it and have no process lead anymore. That's new to their business. We'll see how they cope with it.
Here's an interesting comment/wrapup of the actual densities of Intel vs foundries.
A feature size, like SRAM cell size, is a much better indicator of actual achieved density than some pitch values, since it is a matter of many factors, how much you can integrate on a given area.
That's what happened to Moore's Law:
AMD announced in the earnings call that it is going to release a new server CPU on Global Foundries 7nm process by the end of this year. That's quite a big step from the current 14nm process and I wonder, how this CPU will compare to Intel's server CPUs by that time. To me, it seems like a smart move from Lisa Su to target the server market first with this 7nm CPU on the smallest process node, since Intel tends to switch to new processes late for its server portfolio. That made sense, as long as there was little competition, just to wait for the yields to mature, before switching those large, low volume and fast clocking server CPU dies to smaller nodes. That opens a window of opportunity for AMD, which they are obviously trying to benefit from.
Good results from Intel. Main driver was the data center with 25% growth! So far, AMDs EPYC didn't seem to make a dent. It's a conservative business, though, major shifts are done carefully and slowly. AMD will have to prove that it is here to stay before larger groups will consider a major shift from Intel.
In general, the PC market finally seems healthy. Smartphones slowing down, tablets basically dead and people are still using their PCs (especially at work). More competition from AMD will lead to better and cheaper products, hopefully leading to increasing sales. Something that both, Intel and AMD, will benefit from. Intel will likely face lower margins and its lowered expenses into R&D are not particularly what they need now, to stay ahead of the crowd.
AMD did fine in its last call, by the way. Again a quarter in black, net debt isn't very high and revenue climbed significantly. AMD increased R&D spend by 26% vs the YoY quarter, which partly offsets the increase in gross result. I think it's a sensible move. Competitiveness is more important than short term results at this point. Still a long way to go and Intel likely isn't going to sit still in the meantime. Going to be very interesting. Intels next generation CPU vs Zen 2 may be a major indicator how this turns out.
Samsung said to be early with its 7nm process:
First Zen+ Benchmarks released.
The 12nm CPU with slightly improved architecture and slightly higher clock speeds reaches close to Intel's single thread benchmarks and excels even further in multithreaded applications. In most games, it is now on par with Intel or even faster. Very few arguments left for the more expensive Intel CPUs.
Will be interesting how Zen 2 announced for next year, which is said to clock up to 5 GHz, will compare to Intel's offer by then. Intel's answer to this will be crucial. A 10% improvement won't be sufficient this time ...
The movement of a stock price in the past contains no information, whatsoever, about its movement in the future. A question of belief - for some.
TSMC pumping 24 billion, in words - twenty four billion, into the development of new processes. That is to fend off Samsung (Intel not so much). They plan for 5nm chips in 2020.
With all those crazy numbers, I wonder how high the risk of overspending is in all this. I can hardly imagine a scenario which makes sense economically with such high spending. Who is going to pay for those chips? Who is paying those huge mask costs? How is that killing the economies of scale in semiconductor, when shrinking by 40-50% costs you so much in depreciation of the fabs? Who is out there for only the performance benefits and willing to pay accordingly? Most chips are not highest performance class anyway. This is going to be very interesting.
Some words about AMD's competitive road map that has been presented at CES. Seems like a very well focused portfolio. Especially the integrated graphics part should give AMD quite an edge due to the higher graphics performance at competitive CPU performance, power efficiency and lower cost. Looks like excellent execution on AMD's side.
Redesigned 12nm Zen parts with 12nm as power efficient as Intel 14nm+, according to AMD, will likely close the single thread performance gap further, with higher multithread performance than Intel offers. 7nm foundry then on par with 10nm Intel. We'll see who'll is first. Wouldn't bet on Intel. Losing market share and increasing capital expenditure for coming nodes is a poisonous combination for Intel. Could become a target for a nice short even.
In the meantime, Intel management is busy dumping its shares and let their HR guy lead the security team. Ridiculous, don't you think?
Intel CEO insider selling to minimum before security issue published. I am not one of those who thinks that insider selling tells you much about a companies future, since there are a lot of rules that have to be followed by those owners. Still, I think reducing his share count to the absolute minimum he has to keep doesn't really show signs of much belief in the company he runs.
It is less about this security issue, of which the impact on the company is just unknown. In most cases, those things don't have much effect on sales in the mid to long term. Still, the CEOs action is a sign of overvaluation of the company. He knows best what impact AMD will have on Intel's numbers.
No opinion about the HBM or other integrated options/interfaces of Stratix 10. I know first hand, though, that Stratix 10 is only delivered as engineering samples so far. It's going to be available in quantities soon, I suppose, but that still makes it an almost 4 year delay! In addition, the highest density parts are shifted to 2019. There are even rumors that those will be shifted to 10nm altogether. Intel still seems to have major issues building them. In the meantime, Xilinx got the chance to overtake and they now have access to more advanced nodes than Altera/Intel. Looks like another failed acquisition from Intel.
What would bother me the most is the fact that Intel makes announcements that are that far off and then even doesn't justify in front of clients and investors. That is a style where a company quickly loses customer's and investor's faith. Intel talked a lot in recent years but delivered little. Not sure if or how that is going to change.
Intel compares Xeon with EPYC. Pretty comprehensive comment in this article:
EPYC based HP server with SPEC results released.
The 32 core EPYC seems pretty much on par with the 28 core Intel highend Xeon. It is a bit faster in floating point and a bit slower in integer, so basically a wash. Price is a lot different though, whereas the Xeon sells at 10k$, the EPYC sells at less than 5k$. It would be interesting to know the production costs since AMD will have presumably significantly better yields with its 4 die solution and the overall die area is pretty similar for the two cores.
Performance per watt will be crucial as is total cost of ownership, before datacenters will switch in larger numbers. Intel will go down in price before that happens, for sure. We'll see how much of their cash cows they are willing to sacrifice. Intel needs a lot of money for the (already very late) transition to 10nm.
Adding to the discussion, here's a die size comparison between some Intel and AMD processors:
To recall: The 28 Intel core pretty much matches in terms of performance with the 32 core Epyc. The server processors are best to compare since they don't integrate graphics and such. The Intel CPU is single die, whereas the AMD CPU is multi-die (4 dies on interposer). Intel seems a bit more dense, but that is not by much (AMD's is 8% larger). But: Using 4 dies instead of 1 takes more area, since you need some additional area for pads and sawing etc. The advantage of AMD's 4 die solution, on the other hand, is cost. Mask costs spread across higher volume (multiple SKUs per die) and yield is better for smaller chips. Again, I don't see any process advantage for Intel as of today.
Qualcomm published benchmarks of its server SoC Centriq 2400. The Spec benchmarks look good compared to Intel Xeon, at least considering price and power. Power seems to be about 40% less than Xeon with comparable performance (according to Qualcomm, that is). Performance per core is quite lower than the Xeons though, so it needs more cores for a similar result. It is manufactured on Samsung 10nm process which seems quite capable, also for high performance applications. The Centriq 2460, the highest end model, integrates 60 MByte cache, which seems quite a lot. Will be interesting to see how it compares in terms of die size to Intel 14nm and Globalfoundry 14nm with AMDs EPYC.
Price seems significantly lower (almost half for the high end one) but this is only when compared to single socket systems from Intel. Taking a dual socket Intel based system with cheaper processors, but ending up in the same performance ballpark, the resulting price, at least for the chips, is pretty similar.
The biggest question will be total cost of ownership, where platform cost and power consumption will be crucial, as well as software issues, since that is an ARM based design. My gut feeling is that it's not quite there yet, especially since Intel just got some real competition from AMD and will likely be forced to lower prices, making the Qualcomm solution even less attractive. It will also be interesting how a similar solution from AMD compares power-wise, when real platforms are available from all three architectures.
Very interesting, indeed. Not sure of Intel's "strategy" of outright buying each possible customer for its foundry business. Their track record with Altera seems not great so far.
It would lead to a powerful position of Intel in the mobile market. There would be a lot of redundancy with its existing comms unit. Intel wasted billions to become a force in mobile. They could have bought Qualcomm in the first place and save all that money. Seems a bit desperate to me, if true.
Here are some initial benchmarks of the new Intel+Radeon combo. Seems decent, though I don't have a comparison to what good discrete GPU solutions, but still not so clunky laptops, provide as benchmark results. Maybe someone can enlighten me.
Edit: And here are some results from Ryzen Mobile. Seems pretty much on par, meaning that AMD can basically deliver the same results with (really) integrated GPU as Intel does with its new combo package. It is a bit early though, so it isn't possibly to compare them in real life applications and also costs.
It could end up that AMD actually provides similar performance in the same form factor as Intel can at a few hundred dollars lower price point (Intel solution said to be for 1200-1400$ laptops, AMDs Ryzen Mobile for 800$ ASPs). Still, AMD would benefit from each of those combo packages Intel sells. Seems like a good deal for AMD.
Yup, that's also how I understood it from the article I have linked to. I also got it wrong initially, since many headlines claim that it's integrated. The GPU is not integrated as IP but just a separate die on the interposer. As to AMD, no licensing involved, just selling the bare dies to Intel so they can make a tighter package. In the end it helps Intel with smaller form factors for performance laptops in the discrete graphics class and it helps AMD gain market share in that segment since NVIDIA can't provide the same package themselves. Looks like NVIDIA pissed Intel once too often.
This is also the reason why it isn't going agains Ryzen mobile. That's a different price segment (about half as much as those performance monsters). An integrated GPU is much cheaper but also quite slower. It should be less power hungry also, so this Intel+AMD solution is above 15-25W TDP I'd suspect.
Update: This article explains in more detail what the AMD+Intel deal seems to be about. It is not a licensing of AMDs graphics IP but the use of discrete AMD high performance mobile GPUs on the same chip carrier with an Intel CPU, including integration of the graphics memories in form of HBM. This replaces high end solutions in clunky notebooks with discrete graphics, since the package is thinner and more compact. This is good for both, Intel and AMD, since Intel can provide a high end solution for compact notebooks and AMD will likely sell more of its GPUs in high end laptops than NVIDIA. Win-win. Only loser being NVIDIA here.