AMD Financial Analyst Day (Official Zen & GCN News)

[TorqueS]

You can't forget about Excavator IPC gains.

Piledriver:  100Steamroller: 100   * 1.1  = 110Excavator:   110   * 1.05 = 115.5Zen:         115.5 * 1.4  = 161.7

That puts Zen at 61% higher IPC than Piledriver.

Okay...are we going to reach 100% by evening, and declare that Zen will be twice as powerful in single core as piledriver?

Then again, that would be plausible since we're moving from modules to cores..so in piledriver a core was kinda half a core...so when moving to Zen...core's a core...?

 

God dammit...AMD can you actually make something better than the 5960X? wtf?

[Trik'Stari]

What I think of when I hear "Financial Analyst day"

 

Not sure why. But that's what came to my mind lol

[Opcode]

If it's real we can put Zen at the same level as Haswell, considering GeekBench results are linear and fiable, but im sure i'm wrong and there are much better benchmarks(860K core at 4.4 = ~2500 points, G3258 core at 4.4 = ~4000 points)

From what I hear GeekBench favors Intel.

 

Okay...are we going to reach 100% by evening, and declare that Zen will be twice as powerful in single core as piledriver?

Then again, that would be plausible since we're moving from modules to cores..so in piledriver a core was kinda half a core...so when moving to Zen...core's a core...?

 

God dammit...AMD can you actually make something better than the 5960X? wtf?

In a module a core is still a core it just has to share resources with a brother core. The architecture itself was just poorly optimized hence why the drastic gains (Keller is a miracle worker regardless to what certain people say about him). With Zen they likely widened the pipeline to improve IPC basically going from two small cores to one big core (like Intel). He's also a FPU fanboy so I would also expect just as good floating point performance gains out of Zen.

[TorqueS]

In a module a core is still a core it just has to share resources with a brother core. The architecture itself was just poorly optimized hence why the drastic gains (Keller is a miracle worker regardless to what certain people say about him). With Zen they likely widened the pipeline to improve IPC basically going from two small cores to one big core (like Intel). He's also a FPU fanboy so I would also expect just as good floating point performance gains out of Zen.

Let's hope so, for the sake of competition. 

[spartaman64]

I absolutely agree ... People don't realize that if it was up to intel we will still be on a single core 32 bit chips.

I am exaggerating abit, but back in the days AMD were kicking ass and had many firsts.

and amd got really screwed over when intel lied about pentium performance

[Briggsy]

I believe Zen is still modular, where 4-core/8 thread modules are combined to scale up to whatever the need is. If you look at the block diagrams, Zen is definitely going back to a traditional core design. 

 

 

What's interesting is that modern Intel CPU cores use 6 integer pipelines (similar to ye olde Athlon core from above) so the really basic pre-release Zen core block diagram appears like AMD is copying Intel's current design (as suggested by some tech sites), but in reality AMD are going back to their own designs that they abandoned 8 years ago. If it ain't broke don't fix it.

[NateGSR117]

Holy buckets this is amazing!

[UnbendingNose]

40% IPC over excavator

Which means it's like 15% less IPC than intel's current offerings haha.

[TheTeflonSlackerG]

I hope they are really back at the game, I like AMD with their affordability in pricing. Intel's kinda monopolizing the market right now, so a competition would really help the enthusiasts. 

 

Here is hoping for AMD to have a good upcoming years  B)

[DaRk0]

A 61% increase in IPC over current FX offerings is quite a substantial jump. It will be miles faster than the 4770k in threaded workloads. It boils down to pricing of these products if AMD can deliver an FX-990 or w/e naming scheme they choose that packs 8 cores and 16 threads with an 61% IPC uplift for around $300 (doubt it) then Intel will have some stupid serious competition. A 40% increase in IPC over Excavator would put core performance on par with at least Haswell. Also keep in mind these performance gains are independent of manufacturing process.

I would wait for 3rd party benchmarks before making such claims, but I'm all for having a CPU with similar IPC to i7 4770k with 8C/16C.

[Eroda]

Which means it's like 15% less IPC than intel's current offerings haha.

 

can you link me where you got your excavator results from so we can extrapolate the theoretical 40% ontop to compare how far it is behind intels current offerings

[Phorlorn]

Welp, I may start thinking about upgrading my early-2013 rig from AM3+ to AM4 in 2016. DDR4 memory should become more affordable at this point.

 

[linuxfan66]

i have one single issue. The new microarchitecture should already be on the market piledriving intel in a deep hole, and bulldozing their office, and building a zen meditation mound with a bobcat.

[LAwLz]

Sounds really good when it comes straight from AMD's marketing department, but I will keep my expectations low until I see impartial tests done by a third party.

[Aytex]

hype triain still real

[Victorious Secret]

Sounds really good when it comes straight from AMD's marketing department, but I will keep my expectations low until I see impartial tests done by a third party.

 

All we can ask for. I keep seeing "40%, no wait 50%, no wait 60%". Sorry, marketing slides from a marketing post do not a scientific and repeatable test make. 

 

Let Anand or someone get actual chips, put them through actual tests, and then let us know if this is the kool aid that the internet wants us to think it is. 

[Aytex]

All we can ask for. I keep seeing "40%, no wait 50%, no wait 60%". Sorry, marketing slides from a marketing post do not a scientific and repeatable test make. 

 

Let Anand or someone get actual chips, put them through actual tests, and then let us know if this is the kool aid that the internet wants us to think it is. 

will they fix mantle??

[miagisan]

will they fix mantle??

Mantle is pretty much dead and has been incorporated into dx 12 and vulkan

[Aytex]

Mantle is pretty much dead and has been incorporated into dx 12 and vulkan

lol my friend said dx12 gave a huge preformance increase dont know if its true or not

[KeltonDSMer]

I don't think FX actually does support ECC. I think only Phenoms, Pro APUs, and Opterons do.

 

 

You must not have seen the links I provided for you 8+ months ago when the same topic arose (FX ECC system memory support). I thumbed through the AMD product data sheets and found this (snipped from my post in the "AMD readies new FX microprocessors" thread from August 2014):

 

 

"There are several wiki articles that show AMD's wide ECC support. Here's the actual ECC memory page (First paragraph under Implementations):  http://en.wikipedia....wiki/ECC_memory

Here is the wiki chart for all the Bulldozer/Piledriver fx chips showing ECC support: http://en.wikipedia....microprocessors

 

Of course AMD's official documentation posted on their own site is much better source.

 

I looked at many "Product Data Sheet" PDFs and couldn't find any AMD 64 bit chip without ECC support. Check out the integrated memory controller section in each one, they all specifically mention ECC support.

 

Here is the "family 15h FX" data sheet straight from AMD: http://support.amd.c...-Series_PDS.pdf

Here is the page where all the other sheets can be found: http://support.amd.c...=ecc#k=ecc#s=21 (Click on tech docs)

 

It really does come down to the mobo makers to "implement" ECC support. Most of the higher end boards are good to go. Here is the M5A99fx Pro R2.0 User Guide: http://dlcdnet.asus....99FX_PRO_R2.pdf (Pages 25, 89)"

 

 

Maybe I had misinterpreted the documentation, but at the time it sure looked to me like every 64bit AMD CPU architecture released has supported ECC memory. What do you think?

 

[Humbug]

At last the dark days of the bulldozer revisions are drawing to a close. For most of it's history AMD has not had the performance crown. Yet they had competitive CPUs. But bulldozer was different; it was poor enough to cause major problems in the company performance.

I don't care that this may not be quite as fast as whatever Intel releases in 2016. I don't care that it may not get back the outright performance crown like Athlon64 did. All I care about is that it enables AMD to once again start selling a full range of CPUs with decent margins across a wide variety of price points so that their APU/CPU business can start making money again. What happened with bulldozer and it's variants is that they fell so far behind Intel that AMD ended up having to sell the entire product range at budget prices with razor thin margins in small quantities which has really hurt the company.

Just how good Zen is we don't know yet. But at least we can be confident that AMD competitive position in 2016 CPU market is going to be better than what it is today. How much better remains to be seen...

[Sakkura]

All we can ask for. I keep seeing "40%, no wait 50%, no wait 60%". Sorry, marketing slides from a marketing post do not a scientific and repeatable test make. 

 

Let Anand or someone get actual chips, put them through actual tests, and then let us know if this is the kool aid that the internet wants us to think it is. 

 

The 40/50/60% thing isn't frivolous, it's just that the number will be different if you compare with different things. AMD claims a 40% increase over Excavator, which a lot of people are misunderstanding as a 40% increase over Piledriver or Bulldozer. So if you want to compare with Piledriver, you have to convert AMDs number to a higher number to account for the performance difference between Piledriver and Excavator.

 

Of course AMDs claims have to be put to the test. That goes without saying. I think all we can be reasonably sure of now is that Zen will have significantly higher IPC than any of AMDs current architectures. But that's a no-brainer, and doesn't tell us how Zen will stack up against Skylake or Cannonlake (its 2016 competition). We simply won't know until sometime next year.

[Opcode]

I would wait for 3rd party benchmarks before making such claims, but I'm all for having a CPU with similar IPC to i7 4770k with 8C/16C.

 

Sounds really good when it comes straight from AMD's marketing department, but I will keep my expectations low until I see impartial tests done by a third party.

Keller is disappoint! Burn them! I jk I jk.

 

All we can ask for. I keep seeing "40%, no wait 50%, no wait 60%". Sorry, marketing slides from a marketing post do not a scientific and repeatable test make. 

 

Let Anand or someone get actual chips, put them through actual tests, and then let us know if this is the kool aid that the internet wants us to think it is. 

The variants you're seeing throughout the thread are due to questions and statements regarding Zen compared to other Bulldozer revisions. Obviously the IPC gain from Bulldozer itself to Zen will be much more substantial than Excavator to Zen.

 

Well AMD hasn't failed us yet on their word ever since Bulldozer was launched. Every IPC gain they stated along with Bulldozer revisions have lived up to expectations. Certainly we cannot state that Zen will in fact see a 40% increase in IPC without question. Although given their most recent track record I don't think they will pull out another stop now especially with how critical it is for Zen to deliver performance.

[mr moose]

snip

 

snip

 

I think you missed the point, @Victorious Secret was implying that any claim about performance from the marketing division of a company is frivolous until independently verified/reviewed by someone like at anandtech. And I think the specific reference to waiting for independent reviews to verify this made that quite clear.

[Bouzoo]

lol my friend said dx12 gave a huge preformance increase dont know if its true or not

Theoretically... yes. In reality? We'll have to wait and see. Even though some tests show gazillion % performance increase I'll be happy with 10%. Don't wanna get my hopes up