
The Threadripper CPUs have been updated to the 3rd Generation and we're seeing how the 24 core, 48 thread TR 3960X performs.
AMD Threadripper 3960X Review
Last edited:
It's absurd how CPUs are now at the stage where many programs and workloads can't leverage the potential fully. Obviously games are a big one, but it seems other programs and tasks may need to catch up. Intel is so far behind. It's amazing how a company so much bigger can fall so far behind.
Intel are behind because they havn't really released a new product for years, not from complacency but because they were too aggressive with 10nm and their expectations of EUV development, and have been stuck waiting for physicists to catch up. If Intel had actually been complacent and not set such crazy targets for 10nm they wouldn't be in this mess strangely.
They didn't release a new product in years because they didn't have to. If AMD was there to compete at this level in the past decade Intel would be on 5nm chips today, not struggling to make 10nm. They just chose to bask in mountains of gold and countless virgins like Dragons because they could sell you tweaked Sandy Bridge processors year after year since 3rd January 2011 until 25th November 2019. Would you trade gold and virgins for fringe research in EUV that you don't even need??? Cmon...
By the way TTL,
Is there any way to present the numerical values on your graphs a little clearer? For me its not very easy on the eyes to extrapolate. Could it be embedded in the bars themselves perhaps?
We should really do compilation tests for these processors. I'd love to compare the times to compile UE4 from source with these multicore beasts.
Wendell and Steve GN mentioned this somewhere. It really depends on what you are compiling. Results vary violently from task to task. It would be nice if someone could do a chart so you can pick it if it fits your needs.
Sorry but no, that's just not how the world works, lets check some facts.
- Work on 10nm began in the mid-2000's while AMD was competitive with Intel, long before Sandy Bridge
- Intel spent the last 8 years spending billions more on CPU research and development than any other tech company in the world anyway, as they always did.
- Intel has been unequivocally losing the real CPU war for the last decade in mobile, they have been decimated in a market they threw billions at during the start of the decade, and 10nm was supposed to save them in this much larger fight; They never had a chance to rest on laurels.
- Intel has invested more time and money into their 10nm node than any company in history has ever spent on any other lithographic process.
- We have had half a decade of leaks detailing 10nm's problematic development and showing that Intel has been throwing as many engineers as they can at it.
- Had Intel been on target for 10nm(And not watered it down when management finally realised it wouldn't work a few years ago) they would currently have a node equivalent to TSMC's upcoming 5nm, and could have released an architecture with a >25% higher IPC than Skylake.
- Public companies are controlled by shareholders, if investors got any hint Intel wasn't pushing to be the best, the harm to Intel's financials would have been grave and far beyond any (non-existent) savings.
This is a case of near criminal mismanagement in a company where the engineers are clearly too detached from the financials and direction as a whole, a company or organisation growing too large and then collapsing under its own inefficiency is hardly a novel concept.
Iirc 14nm++ is the one where they took a step back and made the node larger again to improve thermals and yields? From the sounds of it before 14nm++ yields were grim, so I don't think it's too surprising they stuck to 4c/8t for so long while pushing node development so heavily, presumably it wouldn't have been very economical for them to attempt more than 4c on 14nm before then(Stuff like this is always possible you might just get shocking yields).
Coffee Lake had to have already been in the pipeline by the time Zen1 was first revealed though, 6c models had to have taped out long before Zen1 revealed with how close they were together, but Intel probably knew an attempted AMD overcut was likely on the cards when it came to core count and pre-empted it a tiny bit.
If that's true, a lot of what people (incl myself) are saying is not entirely fair. We know obviously that Intel can't just conjure up a 6c/12t or 8c/16t CPU out of thin air just because Zen is released and isn't Bulldozer all over again, but a lot of people don't realise (again, I'm incl myself here) that Intel might not have been able to produce anything more than what they were doing. The word everyone says is, Intel were deliberately holding things back. But if they NEEDED 10nm to rid them of their 14nm woes, and 10nm didn't work, they needed to wait until the 14nm process improved enough to actually be able to safely manufacture higher core count processors. Which means we should give Intel some slack. Obviously their plans for lithography haven't played out and the CEO got axed as a result, but it's less maniacal and more just... a mistake.