News AMD downplays 29 percent Zen 2 IPC boost reports

What's going on here? Honest advertising?

 

Reading stuff properly is an underrated skill these days.

 

I know, it really is overrated...

 

We may well see that in one specific benchmark, in others perhaps 10-15% (numbers plucked from the air).

That's just synthetic testing for you.

Whatever the case it still looks like it'll be a good 'un.

 

Most places have a returns policy these day so you should be OK.

 

As a tech reviewer and enthusiast, I really appreciate their honesty.

 

Sure beats the hell out of Intel's current "benchmark against hopes and dreams" policy.

 

And their completely dishonest representation of their product that they already know is faster than AMD.

Intel are going to get an even more painful rear end when Zen 2 hits... I can see it already.

 

Depends... because there is a huge unknown: Will the switch from GloFail to TSMC affect pricing? At the rate TSMC is hoovering up contracts they have to hit production capacity issues at some point.

 

TSMC already hit capacity problems late last year with massive ASIC demand (not all of it from cryptocurrency minders, mind), and 7nm is going to be dedicated to servicing the Giant Fruit Corporation for some time. The 'Apple effect' on supply chain availability is well known and has been around for well over a decade (e.g. the iPod Nano pushing up global NAND prices), they have both the margins to outbid other buyers for wafer throughput and guaranteed large order sizes.

Intel's problem is that they have so much demand they are dedicating production to only parts with the heftiest of margins (to the degree that they've jut stopped bothering with LCC dies altogether and launched a lineup using just binned HCC dies). They will just have to wipe any tears from lost laptop and low-end consumer sales with their fat wads of cash. OEMs (and large corporations) also seem to prefer to wait for Intel availability rather than buy AMD parts to try and fill the same market segments, though that may change depending on how long Intel's demand issues last.

 

So, basically what you're saying means we will see older Intel CPUs (based on LCC) stay in the out hemisphere of pricing while the newer (binned HCC) Intel parts should first stabilise in price and mid- to long-term might even drop a little? That might just look like a brilliant strategy by Intel to get people to upgrade to their latest and greatest. If we didn't know about the production capacity issues, that is.

 

Retail prices may drop as orders are fulfilled and demand drops. Intel themselves are unlikely to lower tray prices, as those are big honking dies so the price floor is higher than with LCC dies.

 

ASUS CEO has made a prediction that Intel CPU shortage will be with us until the 2nd quarter of 2019. (sauce)

Either way, I was more concerned about the AMD side of things and that prices might go up there too due to the change to TSMC rather than what happens on the Intel side.

 

That was what I understood from your post as well. MSRP will not get touched by Intel.

I personally doubt it, but there is that risk, yes. Well... I still want to upgrade my CPU in Q1 or Q2 next year. Looks like it'll be AMD.

 

Depends on what design they go for for desktop Zen 2. If they use common chiplets with Rome, cost could go up: on top of the single 7nm 'core' chiplet, you need to package it with an entire additional 'northbridge' die. If they go for a larger monolithic 7nm chip as mostly a copy of Zen+ with some extraneous bits chopped out (e.g. remove the pretence of ECC support to simplify the memory controller) then there may be a slight cost increase from 7nm's increase cost/transistor, but not much. Or Zen 2 could be a monolithic 12nm die pushed more towards making up the single-thread performance gap than the perf/watt gap. Whether increased costs trickle down to the retail price depends on how thin AMDs margins currently are and whether they want to focus on price-parity with comparable Intel parts, or price-parity with their own previous lineup (could just kill off the x300 part like they did the x200 and shift every SKU up one bracket).

 

I dunno AMD was pretty quiet about the 3% IPC boost of Zen+ until launch and the 40% IPC claim before close to launch of Zen 1 ended up being 52%. So I think it's possible it is somewhat representative but AMD want to keep a handle on the hype train which has wrecked them before.

Also they don't want to give Intel a heads up until the last minute.

So while clearly it was a mistake, it doesn't mean it is not representative. Though my original predictions were for Zen 2 were a 5 - 7% IPC increase and a 13% peak, but that was a compete guess, right now I am more at a 10% IPC increase for Zen 2 which is still conservative given the rumour mill and especially against this.

 

Their CES slide for Zen+ was an "above 7-8%" increase.

 

That was performance not IPC.