USA Today: http://www.usatoday.com/story/tech/2016/04/19/intel-layoffs-12000-11/83242832/
They even beat their Q4 earnings estimates, which is nuts.
They even beat their Q4 earnings estimates, which is nuts.
Intel announces layoffs for 11% (~12,000) of their workforce
- Thread starter Albuquerque
- Start date
-
- Tags
- intel intel layoff
Intel is desperate for growth and plowing good money after bad by investing in a mobile division that's not making them money at the moment. This announcement, the departure of senior executives recently and the poaching of a Qualcomm exec means they're trying to change the course of a massive ship. They're trying to invest in 5G technologies but this yet to be established standard may not reward any one player as it did in the past with Qualcomm, and I think this puts Intel at a relative disadvantage.
Managing a healthy profit in the 2020's (when 5G is set in stone) will probably come down to actually making superior chips rather than relying on patent protected revenues as dominant companies did in the past. Intel kept its licensing very narrow with x86, only officially licensing to AMD and fighting tooth and nail against reverse engineered products from Cyrix, Centaur et al. It used very underhanded rebates to crush AMD when it had inferior engineering; this succeeded and they ultimately manufactured almost all x86 products and kept most of the profit. They're ceding ground in this current mobile computing generation to fabless players like Qualcomm which licensed its mobile interface technologies to phone manufacturers in return for a small percentage of device wholesale price (even if Qualcomm chips are not used in the device) which yielded them a healthy profit once expensive premium smartphones like the iPhone became standard. These extremely lucrative licensing rules might be undergoing a change as the IEEE demanded that fees be limited to the added value at the component level instead of as a fixed percentage of device wholesale, and Qualcomm is naturally resisting these new rules ratified by its competitors. The rules won't apply retroactively to existing tech like 3/4G but I think these rules will probably apply to 5G; it will be tougher for Qualcomm but it could get especially tough to eke out a profit in this environment for a second tier mobile hardware player like Intel.
Intel's fab investments are also yielding less cost benefit each generation; I think their cancellation of Fab 42 ground breaking in 2014 and their Altera acquisition is really a sign of their need to fill existing capacity with higher margin and more in demand wafers now that people are taking 5+ years for a PC upgrade cycle. They will also need to defend their ultra high margin big iron data center CPUs from assault by the likes of Qualcomm and ARM players who don't have fabs upkeep costs and have more expertise in customizing whole SoCs for customer needs. This demand for "heptathlete" multi-discipline performance is in contrast to Intel's focused Usain Bolt-like sprint dominance when it comes to the narrow discipline of designing CPUs, which is generally yielding diminishing returns since SandyBridge (maybe even Nehalem). Having Intel's historically dominant assets is becoming less relevant today in the era of open CPU architecture licensing from the likes of ARM (with very competitive performance / watt implementations from the likes of Apple) as well as more flexible programming languages and virtual machines; hot new algorithms like unsupervised machine learning are much more the realm of GPU-like architectures as well.
I think Intel will have a tough time in the 5G era managing a healthy profit as a distant second place mobile player. Although they still make a healthy profit off server CPUs that they can dump into unfruitful mobile R&D, so do their competitors and changing licensing rules for 5G might make it an extra bloody fight instead of favoring one leader with a wide moat. I wouldn't be surprised if they next spun-off and opened up their fab operations as they forced AMD to years ago; it's so expensive to maintain fabs, IBM recently paid GloFlo to take theirs off of their hands.
Managing a healthy profit in the 2020's (when 5G is set in stone) will probably come down to actually making superior chips rather than relying on patent protected revenues as dominant companies did in the past. Intel kept its licensing very narrow with x86, only officially licensing to AMD and fighting tooth and nail against reverse engineered products from Cyrix, Centaur et al. It used very underhanded rebates to crush AMD when it had inferior engineering; this succeeded and they ultimately manufactured almost all x86 products and kept most of the profit. They're ceding ground in this current mobile computing generation to fabless players like Qualcomm which licensed its mobile interface technologies to phone manufacturers in return for a small percentage of device wholesale price (even if Qualcomm chips are not used in the device) which yielded them a healthy profit once expensive premium smartphones like the iPhone became standard. These extremely lucrative licensing rules might be undergoing a change as the IEEE demanded that fees be limited to the added value at the component level instead of as a fixed percentage of device wholesale, and Qualcomm is naturally resisting these new rules ratified by its competitors. The rules won't apply retroactively to existing tech like 3/4G but I think these rules will probably apply to 5G; it will be tougher for Qualcomm but it could get especially tough to eke out a profit in this environment for a second tier mobile hardware player like Intel.
Intel's fab investments are also yielding less cost benefit each generation; I think their cancellation of Fab 42 ground breaking in 2014 and their Altera acquisition is really a sign of their need to fill existing capacity with higher margin and more in demand wafers now that people are taking 5+ years for a PC upgrade cycle. They will also need to defend their ultra high margin big iron data center CPUs from assault by the likes of Qualcomm and ARM players who don't have fabs upkeep costs and have more expertise in customizing whole SoCs for customer needs. This demand for "heptathlete" multi-discipline performance is in contrast to Intel's focused Usain Bolt-like sprint dominance when it comes to the narrow discipline of designing CPUs, which is generally yielding diminishing returns since SandyBridge (maybe even Nehalem). Having Intel's historically dominant assets is becoming less relevant today in the era of open CPU architecture licensing from the likes of ARM (with very competitive performance / watt implementations from the likes of Apple) as well as more flexible programming languages and virtual machines; hot new algorithms like unsupervised machine learning are much more the realm of GPU-like architectures as well.
I think Intel will have a tough time in the 5G era managing a healthy profit as a distant second place mobile player. Although they still make a healthy profit off server CPUs that they can dump into unfruitful mobile R&D, so do their competitors and changing licensing rules for 5G might make it an extra bloody fight instead of favoring one leader with a wide moat. I wouldn't be surprised if they next spun-off and opened up their fab operations as they forced AMD to years ago; it's so expensive to maintain fabs, IBM recently paid GloFlo to take theirs off of their hands.
Last edited:
Wasn't this re-org way overdue ? The PC market has been in decline for the past five years.
Intel's revenue from PC parts are now less than 60% of total revenue, yet revenue, gross margin and gross profits are intact. Reducing head count by 11% is only going to improve margins.
While Intel has a hard time cracking the mobile market, the reverse is true for the PC and server markets to an even higher degree. Considering the margins in the mobile market vs PC/server, that's not a bad place to be for Intel.
I'd short Qualcomm a long time before shorting Intel, that's for sure. The only credible threat to Intel in its dominant markets is AMD, - if they survive and don't fuck Zen up. Qualcomm on the other hand, faces an onslaught of chinese manufactures with corresponding immense competitive pressure. High end Android phones are going to be cheap and margins razor thin.
Cheers
Intel's revenue from PC parts are now less than 60% of total revenue, yet revenue, gross margin and gross profits are intact. Reducing head count by 11% is only going to improve margins.
While Intel has a hard time cracking the mobile market, the reverse is true for the PC and server markets to an even higher degree. Considering the margins in the mobile market vs PC/server, that's not a bad place to be for Intel.
I'd short Qualcomm a long time before shorting Intel, that's for sure. The only credible threat to Intel in its dominant markets is AMD, - if they survive and don't fuck Zen up. Qualcomm on the other hand, faces an onslaught of chinese manufactures with corresponding immense competitive pressure. High end Android phones are going to be cheap and margins razor thin.
Cheers
All the handset manufacturers ultimately end up paying licensing fees to Qualcomm and will continue to do so for the foreseeable future as long as 3G/4G are being used; it doesn't matter much if they're using Samsung or Mediatek chips. Although 3/4 of Qualcomm's revenues are from selling chips through its Qualcomm CDMA Technologies division, 3/4 of its profits are from the remaining 1/4 of its revenues from its Qualcomm Technology Licensing division. It's easy to misunderstand how Qualcomm makes its money, I would go as far as to say that its chip division's primary business purpose is to feed its licensing arm with more patents to license as it just doesn't generate all that much profit.
I've no special love for Qualcomm, they use equally strong-armed tactics as Intel in the mobile space to get to where they are, but I still think Intel has more to lose than Qualcomm in the parts of the industry where it's dominant due to a fab heavy and no out-licensing model. In the age of rising and fast moving Asian fabs funded by revenues from big customers likes Apple and Samsung able to ramp 7nm tech for 2018 as well as open CPU instruction sets and core designs from ARM licensing, Intel's dominance in the world of CPUs is quickly being eroded down to its data center products; but even here customers like Google are eager to replace them. Intel's recent moves in light of further slowing momentum smack of desperation and are evidence that as it stands, it is less well positioned to seize any meaningful or profitable piece of the growing mobile market.
Last edited:
Problem is that Intel is part of the problem for the cause of the PC market decline over last 3-5 years.
They have not delivered a worthwhile product to convince both consumers and professional market to upgrade, compounded that they seem intent on trying to claw back money on Broadwell that should had been a lesson learned project and move forward with Skylake-Kaby Lake with EDRAM in a timely fashion rather than half hearted as it seems for now.
While Intel have a near "monopoly" on consumer and workstations, they have not realised that market demand is what actually controls growth or at a minimum sustains that market sector and you cannot just shovel out a marginal product and expect it to do well.
The problem now is that again Intel are making a decision in not admitting to themselves the problem they have selling their recent CPUs to consumers and businesses, and their solution is to take their eye even further off this ball.
They may have identified these other markets, but oh boy they have their worked cut out getting mobile and HPC-cloud clients in a large and meaningful way.
Especially as NVIDIA is now close with IBM,/CRAY/Google/Etc one side of the market that makes for some convincing wins in the HPC sector, and Intel competing against Qualcomm/Apple/Google/etc on the other with mobility.
Cheers
The PC market has been in transition in the past decade. Performance increases are miniscule compared to past performance, that's physics, - nothing can be done about it. It is also a saturated market.
The average age of PCs is increasing because there is little impetus to upgrade if it isn't broken. As long as the average age is increasing, sales will continue to decline. My home PC has a i7 920, - a seven year old CPU.
Some PCs are replaced by other devices (phones/tablets). That further shrinks the market.
The market isn't shrinking because Intel hasn't put EDRAM on every SKU. The lack of SKUs with EDRAM is because of lack of market demand !! It's an expensive way to add bandwidth and is best used as a power saving option which is why we see it on low TDP high end mobile chips (like the i7 used in the Surface Pro 4).
Intel would love to sell you a CPU with a huge GPU subsystem but the memory subsystem needed to sustain it, would price it out of the market. That might change when (or if) we see a transition to high bandwidth memory (HBM/2 or HMC).
Cheers
I'm much more excited about the performance and power usage on modern chip design bringing amazing functionality to smaller devices. That's where the money and development is now, with the PC full desktop market shrinking to a point eventually for enthusiast builders only.
Intel has messed the markets up.
The problem is that there are aspects to Skylake that are good, as there is with EDRAM.
However due to Broadwell and it seems Intel trying to claw back cash they have decided to eek out as much time for each product as possible, there is no reason for EDRAM not to be on consumer Skylake, same reason no reason not to release Kaby Lake sooner....
Apart from Intel decided to release Broadwell and trying to sell it to markets.
EDRAM is one of the better innovations they have done recently and it works well, but they are restricting what they put that on to create a false market.
More issues than that tbh but these are the most obvious.
Cheers
There can't be demand for nonexistant products, so that's a circular argument. Intel could easily offer an unlocked skylake-k with eDRAM for gamers, but they just fucking refuse to. That's unbelievable and stupid. Meanwhile, we get laptops that offer similar or higher CPU throughput on certain workloads than high-end overclocked gaming rigs, thanks to the eDRAM. Imagine what the gaming rig could have done with the same facilities! ...Except of course, that's all we can do. Imagine, because Intel is being stupid, and lack of competition isn't forcing them to stop.
We're sort of back at pre-Athlon 64 days, except the CPU we got isn't the Pentium 4, thankfully.
The demand for pimped out, super performing products that were just as possible in theory in the 90's and early 2000's (as an eDRAM packed Skylake today) came from quickly escalating software demands, everything from OS release cycles to 3D games putting more pressure on hardware year after year. So all these just out of reach but imaginable products were quickly incarnated and superseded in practice. Things on that side have cooled off since; 3D games are now mostly limited by budgets for asset production rather than hardware capability.
Maybe when next-gen killer apps emerge (potentially based on the emergent capabilities unsupervised learning neural net algorithms enable), we'll see the next virtuous cycle of hardware demand from software requirements. (This specific type of algorithm will likely not be Intel's strong suit though, I think nVidia is clear better situated at the moment.)
The other part of the former Wintel juggernaut is also experiencing a gradual waning, with an earnings miss this afternoon. A lot of profit was made by Microsoft by riding the hardware / OS upgrade cycle, but operating systems are definitely starting to ramp down as a big profit driver at Microsoft; now it's mostly driven Office, and SQL server despite what their hip and trendy press releases and publicity would have you believe about what their focus is. This sunsetting of a critical leg of Microsoft profit follows a trend made by the competition of giving their OS' away for free since really, few customers want the hassle or choice of dealing with a separate OS product from their hardware now that hardware is relatively inexpensive. With PCs costing sub $500, whatever markup from windows the retail cost takes will be a much higher percentage than in the days of old when a typical system cost over $1k (not inflation adjusted). New OS' also serve a subversive purpose as walled gardens, which are a great way of earning licensing fees from participants; Microsoft's is giving users a free upgrade to Windows 10 is really a sneaky way to direct more users into their incarnation of a closed software ecosystem which will be could more profitable for them if it takes off.
Nadella might be able to turn-around this large software company though; it must be complicated but it seems more feasible than doing this with a large hardware company laden with high upkeep fabs like Intel.
Last edited:
D
Deleted member 13524
Guest
For mobile BGA processors, EDRAM was great back in 2013 until late 2015.
Starting late 2015 (Skylake's release), Intel would have been much better off just using LPDDR4. At 51.2GB/s in a 128bit configuration, their largest 48 EU model would probably be more than well served, while drawing substantially less power than the dedicated EDRAM module and coming out cheaper too. EDRAM has proven to be quite the power hog when looking at battery time comparison between Surface Pro 4's i5 and i7 models.
Now Samsung, Qualcomm and apple have released SoCs whose GPUs can best Intel's much more expensive Core M offerings when faced with 1080p and higher resolutions, in a good part due to the memory bandwidth available. And within that power range, Atom is practically a no-show, with the fastest X7 hardly keeping up with 28nm ARM SoCs from 2014.
There seems to be some difficulty moving up towards newer technologies for Intel, especially for mobile devices. Apollo Lake will be the first to support LPDDR4 and the first devices with it (Surface 4?) probably won't come before the end of 2016, about 1.5 years after the release of LG Flex 2 and Galaxy S6, both with LPDDR4 3200MT/s.
The smartphone Atom Z line seems to have been killed off ts R&D since Moorefield is 2 years old. SoFIA to my perception was a failure because performance is dirt low, which is probably why you can't find them anywhere and all chinese low/mid-ends are populated with Mediatek SoCs.
Their alien technology together with paying a bunch of guys to think how to better castrate fully functioning chips in order to create a larger portfolio would be just great for mobile devices if they could stomp over ARM competitors through illegal monopoly practices like they did with AMD.
I hate to hear about people losing their jobs, but I love to see that Intel's grand scheme is finally not working.
Starting late 2015 (Skylake's release), Intel would have been much better off just using LPDDR4. At 51.2GB/s in a 128bit configuration, their largest 48 EU model would probably be more than well served, while drawing substantially less power than the dedicated EDRAM module and coming out cheaper too. EDRAM has proven to be quite the power hog when looking at battery time comparison between Surface Pro 4's i5 and i7 models.
Now Samsung, Qualcomm and apple have released SoCs whose GPUs can best Intel's much more expensive Core M offerings when faced with 1080p and higher resolutions, in a good part due to the memory bandwidth available. And within that power range, Atom is practically a no-show, with the fastest X7 hardly keeping up with 28nm ARM SoCs from 2014.
There seems to be some difficulty moving up towards newer technologies for Intel, especially for mobile devices. Apollo Lake will be the first to support LPDDR4 and the first devices with it (Surface 4?) probably won't come before the end of 2016, about 1.5 years after the release of LG Flex 2 and Galaxy S6, both with LPDDR4 3200MT/s.
The smartphone Atom Z line seems to have been killed off ts R&D since Moorefield is 2 years old. SoFIA to my perception was a failure because performance is dirt low, which is probably why you can't find them anywhere and all chinese low/mid-ends are populated with Mediatek SoCs.
Their alien technology together with paying a bunch of guys to think how to better castrate fully functioning chips in order to create a larger portfolio would be just great for mobile devices if they could stomp over ARM competitors through illegal monopoly practices like they did with AMD.
I hate to hear about people losing their jobs, but I love to see that Intel's grand scheme is finally not working.
That was just EU .. they way they seem to be suing every major US tech company these days, you can't really take them too seriously.
Yeah, they could give it away for free as well. They just refuse to, *shakes fist at Intel*.
They already tried the highend SKU+EDRAM with Broadwell. It didn't sell !!
The 5775c is more expensive than the 6700K, and is as expensive as the 6 core socket-2011 5820K.
EDRAM's raison d'etre is to provide bandwidth for a bigger GPU part, which makes sense for price insensitive highend mobile chips, but not for highend desktop/workstation CPUs because everybody has a discrete GPU anyway. It then provides a modest performance boost (10-15%) but at much higher cost. And it is really the cost issue that prevents Intel from making a Skylake-K w. EDRAM. I would expect the 5775c to be more expensive to produce for Intel than the bigger die 5820K.
Cheers
Because Broadwell as I mentioned should not had been released and they could had just moved forward with Skylake, this is what quite a few analysts thought they would do.
However it seems senior management is locked into the project and cannot let it go while not committing properly to it anyway for consumers limited to a maximum of a 65W product.
The reason Broadwell did not sell is that for consumers it is not offering much of anything over Haswell, for those on earlier processors who did not go for Broadwell are waiting for Skylake with EDRAM or Kaby Lake with EDRAM...
And yeah there are quite a few potential buyers including myself.
So Intel management by refusing to drop Broadwell project from production is stalling other viable products, this is compounded that Intel did not distribute a large amount of the Broadwell I5/I7 with EDRAM for consumers.
With the right build (and game for consumer) it has been clearly shown that EDRAM benefits business setups-servers but also importantly consumers and also games - at times beating the Haswell 84W part that are higher clocked (some reviews show negligible differences but others do show notable improvements).
And even now they are only just committing to the business-server side for Broadwell with better focused Xeons, which lets be honest can be done now with Skylake if they wanted to.
This is forcing some HPC-server clients to go with the new Broadwell processors rather than continue waiting for Xeons they expected.
From a business strategy it is pretty clear that Intel is forcing Broadwell onto both consumers and business whether they want it or not, by delaying optimum Skylake and later Xeons, limiting EDRAM to a 65W enthusiast Broadwell part (which is not ideal for enthusiast product), limiting Skylake server products,etc.
The only reason I can think they would do this is management cannot accept to write off Broadwell.
Cheers
Bingo !
The exact same would be true for desktop Skylake SKUs with EDRAM. It doesn't add enough performance for the price (and more importantly to Intel, the cost).
EDRAM only makes sense when it is used in conjunction with the integrated GPU, hence only highend mobile SKUs have it.
Cheers
Which (in this globally generalizing form) is just not true. Broadwell (especially i5-5675C and even i7-5775C) does very well against Skylake (6700K/6600K) in games. From what I've seen so far, it seems as especially games with large particle and/or physics systems seem to just LOVE the extra amount of faster-than-RAM cache. 5675C for example is in our benchmark suite decidedly faster than 6600k while also consuming less power (probably due to lower clocks).
I would rephrase the above notion into either "The reason Broadwell did not sell is that did not want to cannibalize Skylake sales" or "The reason Broadwell did not sell is that Intel did not even try and sell it to customers".
I would rephrase the above notion into either "The reason Broadwell did not sell is that did not want to cannibalize Skylake sales" or "The reason Broadwell did not sell is that Intel did not even try and sell it to customers".
My point is that Broadwell is mostly a die shrinkage with limited changes that also taken too long to make it worthwhile to take into production, while also critically limiting enthusiasts to 65W and lower clock speeds.
Even then as I pointed out it outperforms an 84W higher clock enthusiast Haswell part in quite a few games (depends upon the reviewer and game), but that is not enough of an incentive to upgrade.
Skylake actually offers rather more technical benefits for consumers than Haswell or Broadwell, and the EDRAM would had been the icing on the cake for enthusiasts considering the benefits it can be seen having on Broadwell.
And with the right review-hardware and games Skylake does offer notable performance over Haswell.
Broadwell is splitting Intel's business strategy-focus and not in a good way, as can be seen with the persistence to go forward this & next quarter with Broadwell Xeon rather than Skylake with bringing forward Kaby Lake to a better production timeframe.
Intel just announced their plans to get out of SoCs for phones and tablets entirely:
http://anandtech.com/show/10288/intel-broxton-sofia-smartphone-socs-cancelled
Their attempt to leverage their foundries and CPU expertise came too little too late now that sophisticated OoO ARM implementations fabbed on leading edge fab processes (that are generally very close to Intel's best) are common; this is now a separate, independent supply chain fueled by its own profits entirely out of Intel's control.
It looks like their new focus in the mobile space will be 5G thin modems and IoT modules where they hope to gain traction based on new rules that they along with smaller wireless patent holders waved through last year:
http://www.eetimes.com/author.asp?section_id=36&doc_id=1325706
http://www.wsj.com/articles/patent-holders-fear-weaker-tech-role-1423442219
These new rules base royalties on added value of components instead of as a percentage of the whole sale price of the entire device. I generally think these rules are good for consumers and make sense from a fairness perspective, but Qualcomm has already vowed not to follow them:
http://www.bloomberg.com/news/artic...wi-fi-standard-rules-unfair-may-not-take-part
and others have complained loudly:
http://www.eetimes.com/author.asp?doc_id=1326144
It's still not an easy path for forward Intel in mobile; this sounds like a much lower margin space where they don't have the best expertise or any intrinsic advantages from their existing strengths. They could conceivably make another pass at SoCs in the future if they see an opportunity better aligned to their strengths, but I think the best they can hope for is to maintain relationships in the industry and make a push for IoT data to go through big data-centers where they're still making bank.
I also think Qualcomm and other ARM SoC manufacturers have more to gain in the data centers than Intel does in modems based on their willingness to tailor their SoCs with custom blocks and craft entire racks for customers who now might write most of the custom software stacks for their systems. The old model of supplying chip supply to a system manufacturer who sells to customer who then installs existing server software is from a bygone era when hardware was bigger and more expensive plus open source/open licensing for both hardware and software was way less prevalent. There's plenty of slack at every step of the way that adds up to massive inefficiencies in data centers, a more custom but mostly sealed rack (with soldered memory instead of DIMM slots and more FPGAs) could pack more compute and memory than existing designs. Intel's superior absolute single-threaded performance may not matter for some loads where hitting the knee of the curve performance/watt wise is more important, and newer algorithms will leverage FPGAs and GPUs architectures better. (Intel's acquisition of Altera is part of this trend, although Altera is the second place player to Xilinx and Altera was purchased because of an existing relationship w/ Intel's foundries.)
http://anandtech.com/show/10288/intel-broxton-sofia-smartphone-socs-cancelled
Their attempt to leverage their foundries and CPU expertise came too little too late now that sophisticated OoO ARM implementations fabbed on leading edge fab processes (that are generally very close to Intel's best) are common; this is now a separate, independent supply chain fueled by its own profits entirely out of Intel's control.
It looks like their new focus in the mobile space will be 5G thin modems and IoT modules where they hope to gain traction based on new rules that they along with smaller wireless patent holders waved through last year:
http://www.eetimes.com/author.asp?section_id=36&doc_id=1325706
http://www.wsj.com/articles/patent-holders-fear-weaker-tech-role-1423442219
These new rules base royalties on added value of components instead of as a percentage of the whole sale price of the entire device. I generally think these rules are good for consumers and make sense from a fairness perspective, but Qualcomm has already vowed not to follow them:
http://www.bloomberg.com/news/artic...wi-fi-standard-rules-unfair-may-not-take-part
and others have complained loudly:
http://www.eetimes.com/author.asp?doc_id=1326144
It's still not an easy path for forward Intel in mobile; this sounds like a much lower margin space where they don't have the best expertise or any intrinsic advantages from their existing strengths. They could conceivably make another pass at SoCs in the future if they see an opportunity better aligned to their strengths, but I think the best they can hope for is to maintain relationships in the industry and make a push for IoT data to go through big data-centers where they're still making bank.
I also think Qualcomm and other ARM SoC manufacturers have more to gain in the data centers than Intel does in modems based on their willingness to tailor their SoCs with custom blocks and craft entire racks for customers who now might write most of the custom software stacks for their systems. The old model of supplying chip supply to a system manufacturer who sells to customer who then installs existing server software is from a bygone era when hardware was bigger and more expensive plus open source/open licensing for both hardware and software was way less prevalent. There's plenty of slack at every step of the way that adds up to massive inefficiencies in data centers, a more custom but mostly sealed rack (with soldered memory instead of DIMM slots and more FPGAs) could pack more compute and memory than existing designs. Intel's superior absolute single-threaded performance may not matter for some loads where hitting the knee of the curve performance/watt wise is more important, and newer algorithms will leverage FPGAs and GPUs architectures better. (Intel's acquisition of Altera is part of this trend, although Altera is the second place player to Xilinx and Altera was purchased because of an existing relationship w/ Intel's foundries.)
Last edited:
Intel's difficulties in the mobile space seems at least partially tied to their insistence on making their SoCs x86 compatible for strange and incomprehensible reasons. Nobody cares about x86 in the mobile space. It's not important, and it's not an advantage. In fact it's the opposite of an advantage. Intel is mistakenly confused (ever since the Itanic debacle became self-evident even to Intel themselves) that their business model is selling x86 processors, when in reality it is to manufacture and sell chips period.
I'm unsure if this move is really advantageous for Intel though. Seems they're locking themselves into their traditional desktop/server market (which might not exist for all eternity), rather than trying to identify and fix whatever it is that is holding them back from succeeding in the mobile space. The market could slowly be moving away from them, and by sticking to their proven strategies they would hence be selling to an increasingly shrinking customer-base.
There's a fair chance that Intel and its current products might go as obsolete one day in the future as kodak 35mm film is today...
I'm unsure if this move is really advantageous for Intel though. Seems they're locking themselves into their traditional desktop/server market (which might not exist for all eternity), rather than trying to identify and fix whatever it is that is holding them back from succeeding in the mobile space. The market could slowly be moving away from them, and by sticking to their proven strategies they would hence be selling to an increasingly shrinking customer-base.
There's a fair chance that Intel and its current products might go as obsolete one day in the future as kodak 35mm film is today...