The best (and worst) of SemiAccurate

Our Track Record

Below is a short listing of some of our articles that were anywhere from days to years ahead of anyone else in the industry. Please pay special attention to the date of publication, both ours and the next public source we were able to locate. This is not an exhaustive list by any means although we did the best we could to research non-SemiAccurate sources.

Over the years there is a constant refrain from commenters and various other technology news sites about the accuracy of SemiAccurate stories. They are all very quick to criticize but rarely if ever correct their own work or apologize for inaccurate or libelous posts. As you might expect our frequent retort of, “compile our track record and you will see” tends to fall on deaf ears, that is the wont of an internet troll.

Because of this we decided to compile some of it ourselves, mainly for fun when we got things right, wrong, and something in between. That list grew silently in the background and one day we finally decided to formalize it. Hunting down links and documents to get it worthy of public scrutiny took a bit longer than planned but it is finally here now. We hope you enjoy it.

We consider this to be a ‘living ‘ document and plan on adding, subtracting, and changing parts of it as news becomes available. We will not call out these changes like our other stories because it is not a story, it is just a running commentary on what we see and have seen. If you have something we missed, good or bad, please email editor @ semiaccurate dot com.

Our Journalism Standards

We do not pull articles. We do not silently edit articles, unless there’s a typo or similar minor error. Our track record speaks for itself, mistakes are left to stand. Updates, additional information, and corrections are added to articles with the date and time of such edits noted at the end of the article along with the content of the edit.

Our news stories about unreleased products and non-public industry events are always confirmed from at-least two independent sources before they are published, sometimes more. Review content comes from samples sent to us by manufactures and we abide by the embargo dates that these companies set.

We do not take kindly to the unethical games that some unethical companies like to play with technology journals that use an ad-based revenue model. This is one of the major advantages of our subscriber-based model; we have no monetary incentive to bring you anything other than what we believe to be the truth about a given product or company. More to the point we have no fear of calling out such behavior when we hear about it, we are not afraid of some company pulling ads.

Note: These stories are posted in no particular order and this list is not intended to be comprehensive. There are some stories that have still not become public knowledge yet and we are tracking those until evidence exists publicly.

ARM and AMD:

One example of financial analysts pulling from SemiAccurate is RBS’ Didier Scemama and ‘his’ June 24, 2011 note entitled, “ARM” “Getting closer to AMD”. Read this SemiAccurate story, then this one, then compare that to what Didier wrote in ‘his’ analysis if you have access to a copy. Please note that Mr Scemama did not contact SemiAccurate about our stories and was not at the conference in question. ‘His’ note is effectively the two articles information expanded upon with hilarious results. We say hilarious because where he ‘expanded’ upon our analysis he got important details painfully and humorously wrong. Please note that PCI is not an ARM interconnect nor is it absent from the x86 CPU world. Several calls and email to RBS concerning this note were not returned.

Other analysts were openly sceptical including some literally sitting next to the author at the talk. It took AMD over a year to admit to having an ARM core development program but it turned out to be almost exactly what SemiAccurate described far earlier.

Time: June 2011 to October 2012 ~1 year 4 months lead.

Microsoft drops Nokia name:

Microsoft recently dropped the Nokia brand name to the surprise of most onlookers. SemiAccurate subscribers heard about it in June of 2014. Others only found out unofficially in late October of the same year and officially much later. We were over four months ahead of the next source on that one.

Time: June 2014 to October 2014 ~4 months lead.

Intel and Rockchip and Radios:

Intel has made investments in several Chinese mobile SoC vendors like Rockchip and Spreadtrum. Most of the press was quite sure that Intel would invest in Mediatek, with most saying it would be an outright purchase.

This prediction went on for months with many variations on the theme but all were dead certain of the tie up. They were also dead wrong. SemiAccurate was the one lone source who called the Intel investment in Rockchip correctly and we did it in early April, 2014.

Almost two months later it was officially announced but no one had the complete story. Important parts that SemiAccurate told you about like the modem integration were not public until November of 2014. More importantly almost nine months later several key pieces of the story have still not found their way to anyplace other than our analysis.

Time: April 2014 to May 2014 and November 2014 ~2 months and  7 months lead.

Nvidia’s Denver core:

When SemiAccurate wrote up Nvidia’s Project Denver core down to the pipelines in August 2011, people laughed, mocked, and said we were crazy. We weren’t but it wasn’t until two and a half years later at CES 2014 that Nvidia first outed a few hints about the architecture. It was obvious that SemiAccurate was dead on right. Believe it or not we are really happy that the Denver team persevered and got the project out, congratulations on a long hard job.

Time: August 2011 to January 2014 ~2 years 5 months lead time.

Samsung’s mobile GPU project:

Then there is the tale of Samsung building an internal GPU as reported by Fudzilla in September of 2014. We reported the same thing a year and a half before and have followed it up with more details since that are again still not public.

Time: February 2013 to September 2014 ~1 year 7 months lead time.

Nvidia becoming a patent troll:

The technical world was shocked by Nvidia’s patent trolling turn, something that SemiAccurate said would happen long before anyone else. We first told you that their so called Kepler License was a smokescreen. We later analyzed the situation and told subscribers quite directly that Nvidia was going on a patent trolling rampage against the entire mobile industry.

That was in June and July of 2013 respectively. On September 4, 2014 Nvidia partially confirmed our news, we say partially because most of the targets are not public yet, just Qualcomm and Samsung. SemiAccurate went into far more detail, you can read the summary here. We were once again over a year ahead of the official news with more detail much of which is still not found elsewhere.

Time; June 2013 to September 2014 ~1 year 3 moths lead time.

Texas Instruments and OMAP:

When SemiAccurate first wrote up that Texas Instruments was trying to sell off it’s OMAP division, TI got a tad annoyed. By “a tad” we mean more than peeved enough to lie to EETimes’ Dylan McGrath and get him to write-up a lilly-livered and completely false story about OMAP not being for sale.

Roughly two weeks later when we named the suitors and added more details, TI finally got in touch with us directly to deny the sale. PR was also being lied to but SemiAccurate doubts they actually knew about the deception. Although we published their statement we knew from multiple sources that our story was dead on right.

Over a year later when all of the deals fell through for the reasons we laid out, Texas Instruments shuttered their OMAP division. Since EETimes didn’t even name us in their attack piece, we didn’t expect much in the way of an apology or acknowledgement from them. We got that and no more.

Time: July 2011 to October 2012 ~1 year 2 months lead time.

Maxwell on 16nm:

July 1, 2014 we published information about Nvidia skipping 20nm and moving to 16nm production long before anyone believed they could or would. On August 25, 2014 Economic Daily News wrote up a portion of our news above. The short piece is that TSMC is beginning volume production of 16nm in 2015. Digitimes later wrote that TSMC would be pulling in 16nm to early 2015, both stories were almost two months after SemiAccurate and far less complete. The GPU names reported in our story were correct at the time but later changed for marketing reasons.

Time: July 2014 to August 2014 ~2 months lead.

Nvidia does better than expected:

Nvidia reported good numbers for it’s Q3/2015 sales, a surprise to everyone but SemiAccurate readers. Apologies for being a quarter early, the deals we told you about came in a few weeks later than planned moving the bump into the next quarter. Sorry.

Time: May 2014 to November 2014 ~6 months lead.

Nvidia’s Maxwell process choice:

There was much debate over Nvidia’s process choice for Maxwell during 2013 with definitive data going public only in early 2014. Tweakers was the first to break this information but countless others wrote up baseless and wrong speculation earlier. SemiAccurate however was first by over a year, we reported that Maxwell would be both 28nm and mid-range first in late 2012.

Time: December 2012 to January 2014 ~1 year 1 month lead.

Faked Fermi boards:

In October of 2009, Nvidia’s CEO got up on stage at their GTC conference and lied to an audience full of customers, reporters, and analysts. Only SemiAccurate said the emperor had no clothes and not only that, we proved it.

Now that the Fermi saga is long past, only SemiAccurate was correct about the problems, products, and lies surrounding it. We are still waiting for the corrections on other sites but we don’t actually expect they will apologize.

Time: N/A

Surface and HP:

When SemiAccurate said that HP was dumping Microsoft over Surface, some people laughed, some people broke their own written site ethics rules to copy our story and not credit, and a few did the right thing. Unfortunately no one understood what we meant by this and most still don’t. The idea is simple, Microsoft shattered their partner ecosystem and it is gone for good. Chromebooks are now ~20% of the US retail market and Microsoft is toothless and unable to stop them. Their dominance is over and this was the catalyst, now do you get what we were saying?

Time: Dirt doesn’t age appreciably.

Apple and GPU supply:

SemiAccurate has a long history of calling Apple’s GPU supplier moves correctly and long before anyone else. This is just one example.

Time: Various but many months minimum.

Facebook and ARM servers:

When SemiAccurate said that Facebook would jump into ARM servers, people laughed. OK some people laughed but we got the most annoyed phone call from an Intel rep that we have ever gotten, they were incensed and demanded that we retract our ‘false’ story. Others were far less ethical. We didn’t retract anything because we were right and we knew it.

Over two years later it started to leak that Facebook was and had been trialling Calxeda servers in their data centers just like we said. Again no apologies, corrections, or anything else, not even credit where it was due. But that isn’t enough for SemiAccurate, we kept on the story and are still following it, see?

Time: August 2010 to September 2012 ~2 years 1 month lead.

Intel and GPU memory:

Remember when SemiAccurate said that Intel would have custom memory on their CPUs for graphics purposes? We were also the first to name Crystalwell as the technology in question and were months if not years ahead of the next source. No one believed us but we nailed it, or came really close. It ended up not being on an interposer and it was more general purpose than we thought but the main points were dead on.

Time: Various.


From before day one, SemiAccurate called Intel’s Ultrabook idea dumb. Actually we were not that kind to them but for good cause, they really were a bad idea. We didn’t just criticize, we told you why the very underlying concept was broken, they were truly “shiny things for the stupid“. Once released they flopped for all the reasons we said. Rather than taking common sense into account and fixing the underlying problems, Intel threw money at the wrong places to fix the wrong things. We laughed.

In the end Intel even admitted that Ultrabooks were a failure but only internally. When the money dried up, so did interest in the program which proved SemiAccurate’s view from day one. Now Intel is talking up 2-in-1s and convertible tablets throwing money down a differently named hole for the same wrong reasons. Expect similar success for similar reasons.

Time: N/A

XBox One and PS4:

SemiAccurate was the first one to correctly call the CPU and GPU of the ‘next gen’ consoles, now called XBox One and PS4. We exclusively told you when the XBox One taped out, when the chips were produced, and why it would lose to Sony by design.

We also were the first to tell you about the Sony PS4’s CPU and GPU, and technically why it would beat Microsoft’s offering like a drum. It did for all the reasons we said. Note: SemiAccurate did confuse some code names for a stillborn 360 shrink with the One CPU but our information was otherwise dead on for the One silicon.

Time: Various from months to years lead.

Intel and 22nm:

Intel’s 22nm process wasn’t exactly smooth sailing unless you compare it to the current 14nm woes. Unfortunately some analysts don’t live up to the name, they don’t seem to have a clue. SemiAccurate is not one to pull punches in the face of abject stupidity.

Time: N/A

Apple’s in-house GPU:

SemiAccurate has long known about Apple’s internal GPU designs, something we wrote up in a fair amount of detail here. The news was finally broken by the mainstream tech press May 29th, 2013, five months after SemiAccurate’s story but with less information. To date almost two years later, no one has anywhere near the details we do.

Time: January 2013 to May 2013 ~5 months lead time.

Apple fabbing at TSMC:

Apple fabbing SoCs at TSMC on the 20nm node is now old news but in 2012 no one had a clue as to what would happen. Every site out there had written up that Apple would fab at Samsung, TSMC, Intel, and others as fact, often contradicting themselves directly. Only SemiAccurate had one story, first, and stuck to it. Better yet we were right too.

Time: Varies but over a year.

Samsung uses Qualcomm

Samsung did the unexpected at the launch of the GalaxyS IV phone and used Qualcomm chips. Rumors abounded as to what they would and would not use, and some did have the correct answer. SemiAccurate does not know if these were just lucky guesses or someone had an actual source. One thing we do know is that we were the only ones to tell you who they would use and why, the why being the important part.

Time: Weeks lead.

Kepler wins:

Only SemiAccurate called the Nvidia Kepler vs AMD’s GCN right, and we did it months before anyone had a clue about the specs of either card. Then again we had direct data about performance of both cards when others didn’t even have the names right. Go back and look.

Time: Several months depending on who you consider to have news vs posting every possibility in various stories.

Nvidia EOLs 280:

Nvidia prematurely EOL’d the 280/Fermi line because of cost issues. Only SemiAccurate had the story and had it right. Nvidia went on an astroturfing campaign to counter our story, directly lying to journalists who inquired about the EOL. In the end they had to admit it, once again SemiAccurate was right, Nvidia’s traditional honesty surfaced again.

Time: Weeks but the eventual admission was forced by SemiAccurate’s story.

Sandforce imploding:

SemiAccurate was the first to tell you about the major shakeup in the flash SSD controller market caused by Sandforce’s implosion. That was in February of 2014, the next site to get a clue about the problems was over five months later in July 2014. VR-Zone got wind of our older Sandforce article. We were right and to date we have not seen anything like the analysis of the implications we brought you on this topic.

Time: February 2014 to July 2014 ~5 months lead.

Larrabee/Knights* and memory:

Thomas Ryan was the first to dig out that Intel’s Landing was going to use memory on package rather than a traditional memory on board configuration. This was nearly three years before Intel admitted it would be happening.

Time: November 2011 to July 2014 ~2 years 7 months lead.

Things we got wrong:


In January of 2014, SemiAccurate published a picture and specs of the new USB-C connector. Living up to our name/disclaimer, we got the picture wrong. That said we did get the exclusive story about what the connector would do, including the Displayport functionality and power right. When VESA briefed us over eight months later, the picture was very wrong.

In our defense, we will say that the port in question did support DP signalling and all the other stuff, we saw it working. In the end we will count this as wrong because the headline read picture and it wasn’t, sorry.

Nvidia’s Denver:

All our sources were pointing to Nvidia cancelling Project Denver aka their 64-bit custom ARM v8A core. We said it was really looking like it was dead in April 2014, but it wasn’t. At least our title was questioning the program rather than stating it like some others.

The End…. For now.S|A