[Col. Fishguts]

The transistor interconnects are done with thin metal layers. But AFAIK copper has been used for this since a long time (some processes also used aluminum sometimes IIRC).

[niffiwan]

BTW, would there be any benefits to using multiple CPUs? IIRC, there are "processor cards" on market, but they seem to be meant for professional applications such as data analysis and scientific calculations. I'd imagine they'd have the same problems as multicore processors, only worse.

IIRC multiple CPUs back when dual CPU mobos were a thing would only allow for ~30% more speed?  Of course this may have changed since the dual PIII days.

You've got to be talking about multiCPU's in desktops right? Because servers have had multiCPUs since... hell, probably before I was born! And multiCPU servers also massively predate multicore CPUs.

As for benefits, it always comes back to the application & the OS. With both setup correctly, you'll get nearly linear performance improvements as you add CPUs, applications written for IBM s/390 & successor systems come to mind. Without both setup correctly, you're generally limited by the single process/thread performance from a single CPU/core (ofc).

[Phantom Hoover]

My guess is that the added overhead from intercommunicating between two distinct CPUs means that they don't see any meaningful performance gains except on absurdly parallel workloads, i.e. not the kind of thing any desktop user is likely to want.

[niffiwan]

I agree & I disagree   Yes, it won't help the typical desktop workload, but no, you don't need an absurdly parallel workload to see meaningful performance improvements.  I'm not 100% sure, but I think the interconnect overheads aren't as high as you think for the appropriate hardware, e.g. with memory shared between separate CPUs (non-NUMA) & high speed interconnects (like... damnit I can't remember the name of the tech I'm thinking of... infiniband maybe for the modern example?)

[FrikgFeek]

AFAIK dual-CPU solution can't share cache, that alone slows them down considerably. Dual-CPUs are currently only ever available for server boards where different CPUs are assigned to different raid controllers. They usually have their own RAM too.

There's a good reason why not even the most hardcore enthusiast gaming mobos support dual-CPUs but will happily advertise 4-way SLI. It's just not worth the hassle to design a dual-CPU solution for desktops.

[jr2]

BTW, would there be any benefits to using multiple CPUs? IIRC, there are "processor cards" on market, but they seem to be meant for professional applications such as data analysis and scientific calculations. I'd imagine they'd have the same problems as multicore processors, only worse.

IIRC multiple CPUs back when dual CPU mobos were a thing would only allow for ~30% more speed?  Of course this may have changed since the dual PIII days.

You've got to be talking about multiCPU's in desktops right? Because servers have had multiCPUs since... hell, probably before I was born! And multiCPU servers also massively predate multicore CPUs.

As for benefits, it always comes back to the application & the OS. With both setup correctly, you'll get nearly linear performance improvements as you add CPUs, applications written for IBM s/390 & successor systems come to mind. Without both setup correctly, you're generally limited by the single process/thread performance from a single CPU/core (ofc).

Yes.  And why would I be talking about servers?  xD But if you wanted to talk about servers, my first experience with them would be Dual PA-RISC 120MHz servers with HP-UX 10.2 Yummm.. not.  Start-ups and shutdowns took like 5-10 minutes, backups took like an hour.  We were the last class to be taught using them. They upgraded those to ProLiant servers after they taught us the job using the old servers.  Makes sense, right?

My guess is that the added overhead from intercommunicating between two distinct CPUs means that they don't see any meaningful performance gains except on absurdly parallel workloads, i.e. not the kind of thing any desktop user is likely to want.

IIRC the issue with dual Desktop CPUs was that you had to switch between which CPU was running, and the non-active CPU could only complete it's workload and then wait until it had a turn again.  So it wasn't truly multithreaded, just like Windows 3.1 wasn't truly multitasking.  However you do see speed improvements as you can let one CPU crunch something and while it's doing that ask the other CPU to work on something else.  Someone more technically advanced than me can probably explain this more accurately and in an easier to understand manner.  I'm pretty sure that tech has matured a bit since those days, however, so if you had multiple (multi-core, of course) CPUs I'd think they would be able to be more efficient than that.

[jr2]

(click warning: auto-playing video ads, read below instead if this bothers you)

http://www.forbes.com/sites/antonyleather/2017/02/09/amd-ryzen-prices-revealed-massive-blow-to-intel/

AMD Ryzen Prices Revealed: Massive Blow To Intel
Antony Leather, contributor - I'm passionate about gadgets, PC hardware and computer modding.  Full Bio:

Hidden Text: Show

I’m a technology journalist and currently write for websites such as bit-tech.net and magazines such as Custom PC and Computer Active. I’ve been building and modifying PCs for 20 years and writing about them for over six years. I’ve reviewed countless pieces of hardware and peripherals and also have a passion for all things tech - from Microsoft Windows to Apple iOS, PC gaming to iPhone jailbreaking. I’ll be offering guides, tips and insights into current and future trends, PC tweaking, the weird and wonderful world of PC modding and plenty of gadget-orientated articles.
The author is a Forbes contributor. The opinions expressed are those of the writer.

With just a matter of weeks before AMD's highly-anticipated Ryzen CPUs are expected to hit shelves, it was only a matter of time before we got our first glimpse of the company's pricing structure. In short, Intel should be very, very worried.



Several websites have been pointing at online store www.shopblt.com, which has apparently leaked three AMD Ryzen SKUs all in the 7-series - the Ryzen 7 1800X, 1700X and 1700. Incredibly, the Ryzen 7 1700 is posted as costing just $317. This is hugely significant as according to a previous leak that supposedly revealed the entire AMD Ryzen range, this CPU will have 8 cores and 16 threads. This is the same as Intel's Core i7-6900K, which retails right now for $1,049.

Update: UK website Hexus has spotted UK pricing too, with £365.65, £283.31 and £235.35 listed on www.kikatek.co.uk, for the Ryzen R7 1800X, 1700X and 1700 respectively.

This means that the AMD chip is a whopping 70% cheaper. Even more interestingly, the Ryzen 7 1700 also has a similar clock speed to the Core i7-6900K, which points at the two CPUs offering similar performance in terms of instructions per cycle (IPC), given that AMD itself showed a similar Ryzen CPU competing on a level playing field with this exact Intel CPU earlier this year.



This is hugely significant - if the prices are true, which www.wccftech.com seems to think so given the same web shop leaked prices for older AMD CPUs, which it claims turned out to be spot on - as it means AMD is planning a complete rout of Intel across the board. Its Ryzen 5 CPUs will be even cheaper and could easily offer better value for money than Intel's mid-range 7000-series Core i5 and Core i7 CPUs, while the Ryzen 3 range of quad-cores will be competing with Intel's low-end dual-cores and quad-cores.

We still don't know exactly how the pricing and features of Ryzen CPUs will pan out - it's possible some of them will lack certain features, but for now all we know is that Ryzen could be far better value than Intel's current offerings and AMD could be set for some sizeable gains in market share in the next 12 months.

For more AMD Ryzen news, follow me on Twitter or Facebook.

[Grizzly]

Hardly surprising, considering that Intel's chips were horribly overpriced for what they offered.

But otherwise, I won't believe the hype untill I actually see some solid benchmarks from people who bought the things in retail.

[CP5670]

The Ryzen line looks impressive and should drive down the price of the i7s, which have been getting more and more expensive over time. However, the entire desktop x86 CPU market is kind of saturated at this point. If you have a decent Intel processor from the last 5 years or so (since roughly the 2600K), the improvement in anything new is pretty marginal, and the vast majority of games or mainstream programs don't benefit significantly from more than 4 cores. The only real reason to upgrade is for modern platform features like M.2 or USB 3.1. I don't expect to replace my 4790K for a long time.

[Dragon]

What's the single core performance of that thing? This actually matter more than the number of cores (as long as the latter is over 4, anyway). The highest scores I've seen were for 4.1Ghz Intels (the second best being a dual core i3, funnily enough), if that thing can measure up to them at a fraction of the price, it'd be a really great deal and could actually improve game performance somewhat. However, I see no use for 8 or 16 cores if their SCP is crap.

[MP-Ryan]

The question I always have for AMD when they substantially undercut Intel pricing is heat+durability.

I bought AMD processors a couple times, but I've gone back to Intel in recent builds because I like to achieve decent temperatures without fan speeds that sound like jet engines (even with aftermarket coolers).  If AMD processors are now actually performing at the same level as their Intel equivalents, I'm curious if the other features and factors line up as well.  That said, any cut into Intel's pricing is good news.

[Klaustrophobia]

Now can they catch back up in graphics also?

[Dragon]

The question I always have for AMD when they substantially undercut Intel pricing is heat+durability.

I bought AMD processors a couple times, but I've gone back to Intel in recent builds because I like to achieve decent temperatures without fan speeds that sound like jet engines (even with aftermarket coolers).  If AMD processors are now actually performing at the same level as their Intel equivalents, I'm curious if the other features and factors line up as well.  That said, any cut into Intel's pricing is good news.

If Ryzen is as cheap as it's supposed to be with the only problem being heat production, you'd be able to get a water cooling tower and still come out ahead. I heard those can be much quieter than fans.

[The E]

AMD is advertising TDPs between 65 and 95 Watts, which is pretty much identical to Intels Skylake chips.

[Klaustrophobia]

If this all lives up to the hype, it might be enough to draw me back.  Used to be an AMD fanboy.  Then I got my real job right about the same time Sandy Bridge happened and so was willing to pay Intel prices for noticeable better performance, and AMD never really caught back up.

[Dragon]

It does sound like something that could come back once the manufacturers saturate the physical limits, but seeing as most games don't even use multicore CPUs past 4 or so cores, it'd probably be an extreme high-end solution. Either that or physics cards will become a common component besides graphics (since CPUs primarily do physics in games).

[Mito [PL]]

I'm curious what would the performance be if someone actually made a fully multicore-ready game engine. Yeah, Unreal Engine can chew these frames per second with 4 cores, but that doesn't mean it actually does it optimally.

Photonics seem like the only way to progress in that case. Using light instead of electrons could allow higher speeds and even smaller sizes, but this field is relatively recent, so we don't know if it could actually be a practical solution for a desktop computer.

Well, actually... rather not. Electricity "moves" nearly with the speed of light (electrons do like a meter per hour, but they are "bumping" each other so the "wave" moves with c speed), and typical optic fiber wavelenghts are around 1,2 to 1,5 um. Not to mention that a diameter of an optical fiber is at least about 125 um - and that's the smaller, single-mode one.
And 10nm is on a boundary between UV and X-Ray waves, so... Well, with all other physical problems that someone would run into attempting to create an optical CPU of a similar transistor density comparable to at least a decade old processor... doubtful. At least in close future.

I'm rather interested in what could be made of typical electrical components (transistors), but with some more... unusual methods. Maybe processing with combinational circuits instead of sequential? Or some bipolar-MOSFET transistor hybrid?

[Mikes]

As far as Intel goes Moore's law has hit a brick wall for several years now. The only real improvements have been power consumption and heat for a while ...

... one might expect, now that if AMD is finally catching up on the performance front, they may then further optimize heat and performance just like intel.

(Unless some revolutionary new architecture can actually improve performance again, which is unlikely I guess.)

[Dragon]

Well, if you improve efficiency (reducing power consumption and heat for a given performance), then you should be able to bump the voltage while keeping the thermal output as it was. Hardcore overclockers manage 4GHz+ with elaborate cooling rigs, so if there's anything to be gained, it's there.

[The E]

I'm curious what would the performance be if someone actually made a fully multicore-ready game engine. Yeah, Unreal Engine can chew these frames per second with 4 cores, but that doesn't mean it actually does it optimally.

Define "optimally".

There isn't a lot you can do to parallelize certain tasks. You can't, for example, run gameplay logic in parallel threads easily; there's a certain this-before-that that has to stay intact if game designers want to make reasonable decisions about how the game flows.
Most game engines these days divide their thread pool such that high-level tasks like gameplay logic, physics, audio and rendering can be worked on in parallel, but there's not a lot you can do to further subdivide those tasks so that the engine spreads its load optimally across an arbitrary number of cores.

The lesson here is that multithreaded programming is hard.