You wouldnt abandon Intel for unstable and untrustworthy AMD

/thread

I like it how Intel's 10nm and 7nm roadmap got pushed back 2 years since 2015.

>>59021256

>7nm

Senpai Moore's Law

I might buy it just for the experience of having a semi-decent AMD cpu

>>59021299

You probably meant to say Moore Senpaitachi Law?

File: 1461062580684.png (145KB, 570x754px) Image search: [Google]

145KB, 570x754px

>>59021025
First of all, you miserable fucking faggot, don't you dare defame Onsa. She is better than you in every conceivable way.
Secondly everyone on this board could stand to learn a thing or two more about clock scaling. A given arch on a given process will have a unique curve of clocks vs power per core. The real inflection point on this curve, the overlap of peak frequency/power, is to a degree an adjustable point. The process itself can be tuned, individual Vts can be tuned, and the arch can be tuned for specific clock ranges.

Power per core is a metric everyone should familiarize themselves with. Total package power minus uncore divided by the number of cores. Power per core is vitally important for mobile, and its vitally important for enterprise to ensure maximum perf/watt is achieved. When you have dozens of nodes with 2 or 4 sockets each the difference of 30w per socket can really add up.

Looking at Kaby Lake is a great example, because its recent, and because it highlights clock scaling very well. Most here concern themselves with only desktop parts, and that leads most to ignore the mobile market, which is what Kaby Lake was really designed for. On the desktop we saw marginally increased clocks over Skylake, and more than one reviewer has pointed out that Kaby Lake appears to draw more power than Skylake when OC'd to the same 4.5ghz. This is because Kaby Lake is better tuned for clocks around 2.5-3ghz in their mobile SKUs.
About 3.6ghz is the top end for modest power per core figures, at that frequency a Kaby core is pulling 12.5w~. This is seen in the locked i7 7700.
Increasing clocks 600mhz as seen in the 4.2ghz i7 7700k shows power per core increasing to 21.5w~.

Kaby Lake really wasn't made to be a new generation of desktop parts. Its a bug fix, a new hardware video decoder(really only relevant for mobile) and its a process tune for more frequency in the range of mobile TDPs.

>>59021434
W e w

>>59021025
>we're more experienced in 14nm technology than them
You should be by now.

>>59021434

Weak pasta. Your kike Honda lover was a horrible girl and only Laimu was worse than her. Allow me to remind you that canonically she lets other bikers fuck her for money. You're a sad queer.

File: Onsa.jpg (96KB, 394x749px) Image search: [Google]

96KB, 394x749px

>>59021434
Relevant Kaby Lake figures:
http://www.tomshardware.com/reviews/intel-kaby-lake-core-i7-7700k-i7-7700-i5-7600k-i5-7600,4870.html

Toms has been doing a fantastic job of metering power consumption for the past couple years, and that provides great data points.
Now no Zen parts have been reviewed in depth yet, but we know their base frequencies, TDPs, and an insider dropped one tidbit of useful information that lets us gather similar info. That bit of information is the uncore power in Ryzen having a 15w target. Without violating TDP, that means the top binned Ryzen SKU with a 3.6ghz base clock is achieving its frequency at 10w per core. We know of course that Zen is a physically smaller core than Kaby/Skylake, but this could possibly spell good things for overclocking headroom. More so we have several other data points for lower tiered SKUs, and this can be used to extrapolate the scaling curve downward to see how a Zen core would clock in a many core enterprise SKU.
The gist of it is that a Zen core can hit right around 3ghz at 5w per core, absolutely north of 2.5ghz. This is going to give AMD quite an advantage in the enterprise market, certainly enough to make up for any lacking performance in specific ops.

Despite 14nm LPP and intel's 14nm Trigate FinFET processes being vastly different, each of these cases shows how an architecture scales on a given process. It doesn't compare process head to head because the vital role that architectural nuances play, but regardless it is something everyone should learn. Power per core is what engineers concern themselves with when designing a new arch, and it holds true from a $30 Atom to a $7000 Xeon.

>>59021504
It isn't pasta, sherlock. Onsa isn't a Honda fan, shes the token Yamaha rep.
Maybe spend less time having shit taste, and more time learning a thing or two about complex ICs.

So do i fuckin buy Intel or AMD

File: tzr 250.jpg (517KB, 1600x1200px) Image search: [Google]

517KB, 1600x1200px

>>59021813
Buy whatever offers you the best performance for your dollar.
Unless Ryzen has some glaring hardware bugs which have as of yet gone undiscovered it looks like the 6 and 8 core parts will shake up the market. Even if they're 10% behind Kaby Lake per core at equal clocks, the price point they're offered at makes them compelling. 12 or 16 threads of near Broadwell IPC at 4ghz isn't something to be disappointed by.

File: mme.jpg (76KB, 439x750px) Image search: [Google]

76KB, 439x750px

tfw just bought a 7700k

>abandon Intel
Happened long ago for me, after being jewed with socket 1156. Never again.

Zen is literally bulldozer but theyre marketing it as "not bulldozer"