AMD Zen CPUs delayed to early 2017, more details revealed

[Shawn Knight]

AMD's upcoming Zen architecture was originally expected to arrive – albeit in limited qualities – sometime before the end of this year. That’s no longer the case as AMD has officially delayed its launch until early 2017.

At a recent media event in San Francisco, AMD’s top brass including CEO Lisa Su and CTO Mark Papermaster revealed a bit of news that should make enthusiasts feel less bummed about the delay. According to AMD, it is focusing on high-performance products and it even provided a tiny bit of proof to back up its claim.

AMD matched its 8-core, 16-thread “Summit Ridge” Zen chip against Intel’s Core i7-6900K Broadwell-E chip. Using the Blender rendering benchmark with both chips clocked at 3GHz, AMD’s CPU was reportedly able to render the scene about half a second faster than Intel’s chip.

Adding to what we already know about Zen, AMD confirmed that desktop versions will utilize a new cache memory hierarchy with 8MB of L3 cache in addition to an improved pre-fetcher, separate low-latency L1 instruction and data caches as well as a larger unified L2 cache. Much of the reason AMD’s Bulldozer architecture struggled had to do with slow caches.

AMD also added that it has improved the instruction scheduler window by 1.75x and its execution resources by 1.5x. All things considered, Zen should provide around five times the amount of bandwidth to its core compared to previous designs.

That said, we still don’t know much about AMD’s target clock speeds, thermal output, pricing and so forth. AMD did say, however, that it plans to provide additional details at next week’s Hot Chips conference.

Lead image courtesy majestic b, Shutterstock

Permalink to story.

https://www.techspot.com/news/66021-amd-zen-cpus-delayed-2017-more-details-revealed.html

 

[wiyosaya]

So, perhaps production silicon might have a minor advantage in some benchmarks. If they price this at 1/2 the cost of the i7-6900K, then they might have a winner. As with the previous article, though, benchmarks on production silicon will be key.

 

[Jad Chaar]

Very exciting. May do my next build based on these chips.

 

[Evernessince]

So, perhaps production silicon might have a minor advantage in some benchmarks. If they price this at 1/2 the cost of the i7-6900K, then they might have a winner. As with the previous article, though, benchmarks on production silicon will be key.

Benchmarks have them beating top end broadwell-e intel process at the same frequency (do note this means the item processor is underclocked). If they price their top end consumer processor at half the cost of Intel's of course it's going to sell well.

 

[treetops]

I bought my I-5 4690k 2 years for 205$ on a black friday deal, they were normally 230$ or so. The price still hasn't dropped. However my video card can't say the same. CPU's haven't had any real gains in a while.

If AMD can beat intel in single threaded cpu performance at a lower price I'll be mighty impressed.

 

[cliffordcooley]

Judging by the one image stating "We are back, and just getting started!". Does this mean they are publicly announcing they are back competing with Intel? If so this is exactly what I thought they were doing. Going back to the drawing board, they just needed a break from the competition to do so.

 

[meric]

They dont need to beat intel, they only need to come close in performance and cost a bit less with an acceptable power consumption ratio. Come on AMD, give us something decent and take our money

 

[gazmatic]

As someone said before, they don't need to beat the 6900 in performance, just be close behind and half the price. I'll watch heat and wattage but the price is what would give thema winner.

 

[amstech]

'Projected' performance and specs look good so far.

 

[poohbear]

The situation is very similar to when AMD released the Athlon 64. Back then we had a same complacent Intel that didnt care to innovate as they were king. AMD caugut them with their pants down, is this what's shaping up to be a similar situation?

 

[Lionvibez]

I am looking forward to this chip.

However do not forget for single threaded performance Skylake is what intel offers not Broadwell.

I know the core counts line up with the E models but do not forgot about Skylake!

And by the time this hits market were going to be looking at Skylake's successor.

 

[Evernessince]

I am looking forward to this chip.

However do not forget for single threaded performance Skylake is what intel offers not Broadwell.

I know the core counts like up with the E models but do not forgot about Skylake!

And by the time this hits market were going to be looking at Skylake's successor.

The difference in IPC between Broadwell-e and skylake is near non-existent. You are talking about 0.1 FPS in most games and 1-2% in general computing.

 

[Lionvibez]

The difference in IPC between Broadwell-e and skylake is near non-existent. You are talking about 0.1 FPS in most games and 1-2% in general computing.

I'm aware of that but the difference still has to be pointed out.

Facts are Facts!

 

[namesrejected]

Hope they don't wait till Intel gets Cannon Lake out.

 

[HardReset]

Judging by the one image stating "We are back, and just getting started!". Does this mean they are publicly announcing they are back competing with Intel? If so this is exactly what I thought they were doing. Going back to the drawing board, they just needed a break from the competition to do so.

While Zen is generally designed from scratch, some parts of Zen (like branch prediction unit) are recycled (or recycled with slight modification) from Excavator design just to save time. Redesigning those parts also give more performance. So Zen is "getting back and started" while Zen+ is "real thing".

 

[fps4ever]

So I guess things are looking pretty good compared to the previous "made up benchmark" and subsequent "not looking so good" for AMD TS article. Hopefully lessons learned...

 

[wiyosaya]

So, perhaps production silicon might have a minor advantage in some benchmarks. If they price this at 1/2 the cost of the i7-6900K, then they might have a winner. As with the previous article, though, benchmarks on production silicon will be key.

Benchmarks have them beating top end broadwell-e intel process at the same frequency (do note this means the item processor is underclocked). If they price their top end consumer processor at half the cost of Intel's of course it's going to sell well.

Yes, I am definitely looking forward to it if it is appropriately priced, and not higher priced than Intel. Like I said, 1/2 price would almost certainly guarantee its success, and, perhaps, force $ntel to drop their outrageous prices for performance barely beyond my Ivy-E. Since I was tainted by the Bullcr4p fiasco, I do want to see benchmarks from production silicon first. Don't get me wrong, I only have one Intel build out out of five computers that I've built in the past few years. I've been an AMD fan for a long, long time, however, the last time that AMD was ahead of Intel, the AMD procs were priced as outrageously as Intel has their "flagships" currently priced.

 

[wiyosaya]

Interesting article on Zen over at Anand - http://www.anandtech.com/show/10578...lers-micro-op-cache-memory-hierarchy-revealed

 

[misor]

This move by AMD to delay 'intel cpu killer' is the same to the release of so-called iPhone/ipad killers...never came to light...

 

[HardReset]

This move by AMD to delay 'intel cpu killer' is the same to the release of so-called iPhone/ipad killers...never came to light...

Once again, it's not delayed. AMD said long time ago that some parts may come this year (and most probably some will) but big stock is next year.

To make it more precise, even 100 units counts to "some".

 

[misor]

Once again, it's not delayed. AMD said long time ago that some parts may come this year (and most probably some will) but big stock is next year.

To make it more precise, even 100 units counts to "some".

so a soft launch then?

 

[HardReset]

so a soft launch then?

I expect that AMD ships worst units (lower clock speed, perhaps cores disabled) to OEM's this year and keep better ones for next year launch. OEM's sell them renamed to "something cryptic" (like Phenom II B series) as OEM buyers usually have no clue what they are buying anyway. However OEM's may want to build stock before releasing and avoid releasing new products for after Christmas sale season so it would not be surprising if OEM machines equipped with Zen came out about same time as Zen's "retail" launch. CES on january is good guess for that time.

AMD could release small amounts for retail this year too but low availability would mean shitstorm so they probably want to release when have more to sell.

 

[Lionvibez]

This move by AMD to delay 'intel cpu killer' is the same to the release of so-called iPhone/ipad killers...never came to light...

Zen is not going to be some intel killer.

It will provide the necessary competition the market needs and probably be on Par with Broadwell E.

Some of you are getting way ahead of yourself.

This isn't going to be a athlon 64 vs Pentium 4. I expect it to be more like SB vs Ivy which is still good for everyone.

 

[lripplinger]

While the slides look impressive, since that is what corporations are good at, we don't know anything. We don't know what kind of setup they had for either of these machines for the claimed numbers in the slides. Until the actual product is out, I take this all with a grain of salt, like with anything else.

 

[HardReset]

While the slides look impressive, since that is what corporations are good at, we don't know anything. We don't know what kind of setup they had for either of these machines for the claimed numbers in the slides. Until the actual product is out, I take this all with a grain of salt, like with anything else.

We know that both CPU's were 8 cores and clocked to 3 GHz and since it makes no sense to use GPU on that test, we know everything that's important except Blender settings and compiler used.

So we can actually say that ctc Zen is faster than Broadwell-E on Blender, at least when used certain settings. Believe me, when real product comes out and if it turns out Broadwell-E is indeed faster on blender than Zen at any settings, AMD will get tons of negative PR. So AMD has no reasons to cheat there. So far AMD has IMO never caught cheating when publishing performance on PR slides or events.

What we can discuss is on how Blender result reflects on performance with other software.