-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD THREADRIPPER VS INTEL SKYLAKE X
- Thread starter faceman123
- Start date
MS face their own competition now, on the desktop as well as cloud and communications networks. As you say collaboration in the gaming sphere certainly helped.
Deleted member 66701
Deleted member 66701
Lancaster uni are rebuilding thier dc with epyc cpus and HP are doing Doxford, Norcross and Wynyard over the next 12 months.
Its also a competitive thing.
If company A is buying 50,000 EPYC chips at $1000 each and company B 50,000 Xeon chips at $9000 each company A has the competitive edge as it has $400,000,000 in the expenses safe that it didn't need to spend on CPU's.
Other companies that might otherwise not have had the capital to enter the market now do because its costing them $400,000,000 less.
The idea that anyone will go Intel no matter what the price difference is very short sighted.
If company A is buying 50,000 EPYC chips at $1000 each and company B 50,000 Xeon chips at $9000 each company A has the competitive edge as it has $400,000,000 in the expenses safe that it didn't need to spend on CPU's.
Other companies that might otherwise not have had the capital to enter the market now do because its costing them $400,000,000 less.
The idea that anyone will go Intel no matter what the price difference is very short sighted.
They will probably resist that as long as possible.
It's not just denial, but Intel is just not used to working with smaller margins. Or rather, on smaller margins would they have been able to waste $billions on their failed ventures (Atom for mobile [$4 billion?], Larrabee [$1 billion*], retreat from x86 everywhere IoT [$?], buying MaAfee [$7.6 billion, maybe not all wasted] ttempts in the 2000s to dominate networking and comms [which as meant to have cost up to $3-4 billion at the time], etc.)?
For Intel to survive on smaller margins, these kind of blunder will have to stop. Guess the ARM vendors can breath a sign of relieve that things like Atom contra-revenue dumping are not coming back.
Anyway, all of the above makes AMD's biggest mistake (overpaying for ATI) look loose change.
*but Larrabee did sort of morpt into Xeon Phi
I meant the IPC increase from Skylake/Ryzen, it'll be interesting to see the 6c/12t vs AMD 8c/16t, Skylake-X might have the advantage with better motherboards and RAM compatibility.
What works for AMD must work for Intel, right?
Intel's cross core communications doesn't depend on System Memory performance, its a very different architecture, it is why you can't physically separate the cores and stitch them together on an interposer, which is why Intel's CPU's are a singular massive die and AMD are multiple small ones.
Its why AMD can make them so cheap.
AMD's bigger + now is power consumption, Intel could easily put out a CPU with two 8 core dies but the power requirement would be high. The new AMD chips have low power up to about 3.3 – 3.8Ghz. The laptop APU chips should give Intel real problems.
Wouldn't be that difficult for Intel to redesign their CPUs - they already have experience with multiple dies on an interposer and the Northbridge/SA is flexible enough to support the data path. (Its how they used to work and the multi-socket Xeons still communicate via the chipset).
It would be pretty much the same based on the data from the 7700K, 7900X, 6950X and 1800X - with possibly a very slight performance per core advantage to Intel though can't say for sure without an actual implementation.
Last edited:
AMD has a clause that allows them to use other manufacturers if GloFo can not keep up. I believe Samsung is the secondary supplier. they where also present at the event. There is the potential that working on them in the console space helped secure this deal.
I posted a link in the EPYC thread about an AMD job posting where they mentioned working for Amazon. Could you be talking about that?
If it was that easy to stitch two cores together Zen would have come out 3 years ago. Intel does have the advantage that AMD has paved the way, so while they can't use AMD tech they at least have a reference to work with.
That communication also results in a scaling loss, its not just chip-set interconnects, the die size and the low yields they are getting because of that has never been something Intel thought about or had any desire to improve? until now? so it'll be done by next week at the cost of 50p.
You don't think Intel have been working on trying to solve these problems for at least as long as AMD?
I have no doubt Intel will find a way to catchup with AMD but its not going to be as easy as you think it is, if it was they would have already done it, the fact that AMD achieved it first will give Intel the kick up the proverbial to put everything the have into it but its not as if Intel haven't already been trying. these things are not easy, not even for Intel or AMD.
Intel have more than that to worry about, AMD' performance and performance per watt is at least inline with Intel, much higher if you believe the slides in AMD's Data Centre reveal video, on top of that at a fraction of the cost.
Maybe come back to this question in a couple of years and see where Intel are at with it.
Don't forget Intel has been here before - the Q6600 was basically put together like Threadripper - they have all the R&D from multi-socket systems, etc. to draw on they don't need to solve any problems its simply about the desirability of doing it which before this point there wasn't much of a case for. They moved away from that model but the tech has all the capabilities of being repurposed to do it.
Intel have done mult-die chips before.
The Q6600 was simply an early attempt of a dual core by sticking 2 CPU's on a PCB, yes. (Note i didn't say interposer)
It was also a reaction to another AMD first, a mainstream X86 Dual Core, a proper one which the Q6600 was not. if i remember rightly it didn't actually work, the two cores didn't even communicate with each-other.
AFAIK the Q6600 worked fine (FSB core to core) - the Pentium D (netburst) before it also used 2 dies on an interposer and had some communication issues hence one of the reasons why that line flopped hard.
EDIT: Oooh interesting Intel has patented some of the techniques used to create Threadripper in 2010 and 2006 wonder if they'll play that card though not sure they could given how fundamental it is to the technology development (actually not sure how valid the 2006 stuff is).
EDI2: Both AMD's MCM Opterons and the Pentium D pre-date that so I guess they'd have a hard time getting anywhere with it.
Last edited:
From what I read on another forum, the chips you are thinking of communicated using the northbridge, which had a massive latency penalty. Compared to infinity fabric, which from my limited understanding the MCM communicate directly with each other, as well as being able to fetch data directly from the L3 cache. So yes they have done it before but not at the same scale that AMD. To do it at the same scale as AMD will take a lot of time. Assuming that intel hasn't already been doing such research, we may not see anything from them for another 3-5 years.
They could do what they have done before and use the notrthbridge, but IMO it would be a waste of money and would most likely be DOA.
Back in the day they interfaced via the chipset and that is still done for multi-socket, etc. setups - AFAIK with the System Agent design multiple cores including discrete packages can be strapped across the LLC (the integrated graphics package when a discrete module worked in that way).
Intel have papers going back to the Pentium D days on it i.e. http://ieeexplore.ieee.org/abstract/document/1645663/ so would be odd if they didn't have the R&D done.
Last edited:
Whats your question/point, its not clear to me?
Are you saying that using the chipset Intel can compete with AMD? If you are, I disagree because the latency will be too much of an issue. AMD is achieving near perfect scaling at the moment. I don't think communicating through a third party will allow intel to do this.
Q6600 inter-core latency at 4 cores (despite using FSB with dual dies) is pretty much the same as the 6950X, almost double that of the 7700K and Ryzen platform but less than the 7900X - I don't think Intel is going to struggle too much on the latency side - aslong as the link has enough bandwidth which could cause some considerations.
I think people are confusing the lack of desirability of doing it previously with their level of preparedness to do it.