AMD Zen

[Citadelen]

@TheMissxu On a more serious note if anything Zen will make great strides in single-core performance as opposed to multi, AMD's focus with this processor is single-core. I see no reason why AMD can't achieve Skylake levels of performance, 22nm on a whole provided poor gains for Intel, I suspect the majority of Intel's gains in Skylake were due to 14nm and see no reason why AMD can't achieve similar gains.

[TheMissxu]

@TheMissxu On a more serious note if anything Zen will make great strides in single-core performance as opposed to multi, AMD's focus with this processor is single-core. I see no reason why AMD can't achieve Skylake levels of performance, 22nm on a whole provided poor gains for Intel, I suspect the majority of Intel's gains in Skylake were due to 14nm and see no reason why AMD can't achieve similar gains.

I'm basing my assumptions off of Zen moving to 8 real cores, no hyper threading or modules, *and* +40% IPC (which is per core).

[Citadelen]

@TheMissxu The 40% increase is purely architectural, look at my other post to see why I think Zen will do well.

[Nena Trinity]

Zen can be anywhere from 40% better to 100% better on IPC but that is only AMDs claims so far but I doubt they would lie about minimum being 40%...

[looncraz]

Why oh why do people keep saying this? WHY?

 

 

Did you really just use a stock-clocked comparison to compare different generations?

 

These cores are all clocked differently and have different turbo configurations, they can't be compared as generational changes.

[looncraz]

indeed.

i am referring to things like this:

where overall gains seem to be to tapering off.

 

 

Now, remove the outliers and average the results.The only clear outlier is the Dolphin benchmark, with PovRay coming darn close to being one (but then that claims 3dPM and 7-zip, if my mental math is correct).

 

Without removing the outlier, the Haswell average advantage here is 16.45% over Sandy Bridge.  Without the outlier, it is only 12.39%.

 

My multibench mean, which includes nearly 30 benchmarks, puts Haswell at 13.3% ahead of Sandy Bridge at the same clocks.  I recently retested a selection of these benchmarks with a 4690k to verify my numbers, and they were all accurate.  I had to do that because I'm showing a few areas where IPC was lost between Sandy Bridge and Haswell.

[TechGod]

I think it's safe to say it's going to be very similar to Bulldozer — amazing multicore performance (If rumors are true, nearing Intel extreme chip performance), but still mediocre single-core performance (probably near an i7 3770k single core/IPC performance). Of course, by the time Zen comes out, there'll be a Kaby Lake, which is ~4 generations ahead of an i7 3770k (which is about +20% in IPC performance).

TL;DR, amazing multicore performance, ~20% less single core performance (which translates to poorer performance in games in general)

No. Just no. And no.

[TechGod]

I'm basing my assumptions off of Zen moving to 8 real cores, no hyper threading or modules, *and* +40% IPC (which is per core).

Zen will have "Hyperthreading" 

[TechGod]

The bias is strong with the insecure Intel fantards, "Oh fuck competitive!  Let's talk shit to fuel our needs!  Intel cares about...oh wait!"

Have you read the crap in this thread?? Its ridiculous with uninformed people making uninformed comments. 

[looncraz]

I think it's safe to say it's going to be very similar to Bulldozer — amazing multicore performance (If rumors are true, nearing Intel extreme chip performance), but still mediocre single-core performance (probably near an i7 3770k single core/IPC performance). Of course, by the time Zen comes out, there'll be a Kaby Lake, which is ~4 generations ahead of an i7 3770k (which  is about +20% in IPC performance).

 

TL;DR, amazing multicore performance, ~20% less single core performance (which translates to poorer performance in games in general)

 

It's very unlikely to have this dynamic going on at all.  The only reason Bulldozer et. al. APPEARED to have had this dynamic was because it had so many cores at its disposal and  SUCKED in single threaded performance that it was all anyone could speak about to try to redeem it.

 

In reality, though, Bulldozer/Piledriver is AWFUL for multithreaded scaling in addition to single threaded performance.

 

Phenom II X4 had a four-core scaling ratio of 3.65

FX-4300 has a four-core scaling ration of 3.13

 

That's abysmal performance, which is partly because it can't keep the same clocks due to turbo.

 

To make it worse, absolute performance is lower than phenom II X4 980 BE as well.  Often by a lot.

 

http://www.anandtech.com/bench/product/700?vs=362

 

Steamroller helped immensely for multithreaded scaling, and made up for most of the IPC deficit with the phenom II, however mostly in non-FPU tasks (Cinebench doesn't really move except for the multithreaded scaling ratio and clockrates).

[looncraz]

Why?

 

Do you expect AMD to challenge the 5960x ? or fall flat on their face (as they usually do)?

 

The 5960x should be the top target for Zen.  However, I suspect that Zen's SMT implementation will not be as good as Intel's, so I'd imagine that it will more or less tie in single threaded, but lose in multithreaded.

 

This could give us a nice middle ground for performance and price choices between the i5 and the i7, though.  A quad core Zen with SMT should perform better than i5 and be cheaper than an i7, priced right along the Intel price/performance curve.  That means a a nice $250 option with appropriate performance.

 

This time around, I really hope AMD differentiates their CPUs based on SMT being enabled or not.  I want to see a huge breadth of CPU offerings to give consumers a broad choice.

Top CPUs for each, w/ SMT, w/out SMT, assuming identical clocks,    assuming 15% SMT scaling, about half of Intel's.FX-8500 SMT - $850+FX-8500     - $750+FX-6500 SMT - $475FX-6500     - $400FX-4500 SMT - $225FX-4500     - $200Intel's 8-core CPU is $1k, AMD would be FOOLISH to slap a lowerprice on theirs than what I show here (unless performance is worse).

[TechGod]

 

The 5960x should be the top target for Zen.  However, I suspect that Zen's SMT implementation will not be as good as Intel's, so I'd imagine that it will more or less tie in single threaded, but lose in multithreaded.

 

This could give us a nice middle ground for performance and price choices between the i5 and the i7, though.  A quad core Zen with SMT should perform better than i5 and be cheaper than an i7, priced right along the Intel price/performance curve.  That means a a nice $250 option with appropriate performance.

 

This time around, I really hope AMD differentiates their CPUs based on SMT being enabled or not.  I want to see a huge breadth of CPU offerings to give consumers a broad choice.

Top CPUs for each, w/ SMT, w/out SMT, assuming identical clocks,    assuming 15% SMT scaling, about half of Intel's.FX-8500 SMT - $850+FX-8500     - $750+FX-6500 SMT - $475FX-6500     - $400FX-4500 SMT - $225FX-4500     - $200Intel's 8-core CPU is $1k, AMD would be FOOLISH to slap a lowerprice on theirs than what I show here (unless performance is worse).

Why sell CPUs without SMT?

[looncraz]

Is the reason people don't currently have a Haswell or faster processor purely about price? If it is and Zen is significantly cheaper than its going to change things a lot. However if AMD intends to price it competitively then its Haswell like performance in 1 years time and at best its just a value product competing at a point of performance Intel has long ago exceeded. I can already buy a faster processor before Zen even comes out, I will get a full year of utility out of it and Intel will replace those parts by the time Zen gets here. AMD getting to that point 36 months later isn't impressive, because the target is continuously moving. AMD aren't competing with Haswell, they are competing with the processor Kabylake and Skylake-E. They don't do very well in that comparison performance wise being several generations behind,  unless somehow they have massively underesimtated the progress they have made rather than overestimating as usual.

 

I and almost everyone else wanting that level of performance will already have it before Zen comes out. At the time Zen arrives we will have the choice of that or a faster Intel CPU and so its again competing in a more budget constrained area. That matters to the younger PC gamers but to old hats like me I care about top line performance and Zen isn't a compelling product. All it does is potentially lower the price of that level of performance, so therefore price must be the factor holding back everyone buying that level of performance. But I don't think that is true, I think its more about the minimal increases from their existing setups and no amount of purchasing of MB+CPU+RAM is worth it when its only +20%. Even if it was $100 it likely still would not entice many of those people. If it had come out 3 years ago it would have been big, but its too late and its just trying to barely keep up as its top line target, reducing the price of that performance level (presumably since pricing hasn't been announced yet AFAIK).

 

Well, I have a Sandy Bridge i7 2600k.  One of the reasons I haven't upgraded is because I don't yet have the need, nor a compelling option.  I didn't want to pay money to Intel for this, but I did it anyway.  I didn't want to have to recommend a Pentium G3258 to someone over an AMD A4 7300 (IIRC) alternative, but I had no choice - the Pentium has more performance AND potential at the same price... even with a 800-1000MHz clock rate difference!

 

This is likely to change with Zen (albeit not right away as early adopters usually pay more).  I just want to be able to have a choice of similar performance at a similar price, but we don't have that at all right now except in a very few specific cases so that I can actually recommend AMD again (other than their APUs, which often do make more sense at many price points and workloads).

 

But the real hidden factor is just how many of us will default to AMD all things being equal.  If performance is the same, price is the same, features are the same.  I go with AMD every time.  Not because I like AMD (I actually never liked them as a company, I much preferred Cyrix (Centaur), but VIA has no interests in competing in the big leagues), but because we desperately need a viable second source.  It's a big-picture move.

 

AMD should sell based on single threaded performance and core count exclusively for their current products.  Problem is they can't, Intel doesn't sell products that are slow enough in the single threaded department, so AMD averages them and prices accordingly.

[TechGod]

Well, I have a Sandy Bridge i7 2600k. One of the reasons I haven't upgraded is because I don't yet have the need, nor a compelling option. I didn't want to pay money to Intel for this, but I did it anyway. I didn't want to have to recommend a Pentium G3258 to someone over an AMD A4 7300 (IIRC) alternative, but I had no choice - the Pentium has more performance AND potential at the same price... even with a 800-1000MHz clock rate difference!

This is likely to change with Zen (albeit not right away as early adopters usually pay more). I just want to be able to have a choice of similar performance at a similar price, but we don't have that at all right now except in a very few specific cases so that I can actually recommend AMD again (other than their APUs, which often do make more sense at many price points and workloads).

But the real hidden factor is just how many of us will default to AMD all things being equal. If performance is the same, price is the same, features are the same. I go with AMD every time. Not because I like AMD (I actually never liked them as a company, I much preferred Cyrix (Centaur), but VIA has no interests in competing in the big leagues), but because we desperately need a viable second source. It's a big-picture move.

AMD should sell based on single threaded performance and core count exclusively for their current products. Problem is they can't, Intel doesn't sell products that are slow enough in the single threaded department, so AMD averages them and prices accordingly.

There's a lot of people that are saying they hope Zen will perform well and that they'll switch to AMD if that happens. If they pick up even a quarter of the enthusiast market for CPUS and GPUs then rhey would be well on the road to recovery, of course, that depends on if people truly mean they'll change to AMD when Zen comes out. I can't wait for it to come out...

[looncraz]

Why sell CPUs without SMT?

 

More price points you can hit and the more upselling you can do.

 

If you look at Intel's current product line you see some rather funky prices.  You see the Pentiums and Celerons running at $40~$100 with only a minor change in performance for each step across the entire spectrum (with something like a baker's dozen of choices).  Then, clock speed drops slightly for the i3, but Hyper-Threading is enabled, and the price begins at $115 and ends around $160 with 8 choices.  Then we move to the i5 at $185, which ends at $250 for Haswell, with the overpriced $300 Broadwell outlier.  The quad core i7 begins at $305 and ends at $340 (skt 1150, current NewEgg offerings and prices) with only five choices.  The only thing that earns that extra price?  Hyper-threading.

 

AMD needs to copy this method.  You start looking at a $40 CPU but notice that the $50 CPU has 15% higher clock speeds, then you notice that the $60 CPU has the 7.5% higher clock speeds, but also SMT enabled.  This moves right up the entire product line and is how people end up spending more on CPUs than they planned when still thinking rationally.

 

AMD, though, should have fewer options in each pricing tier.  A $20 difference in price will work just fine, except at the very low end which AMD should not abandon to Intel (Intel has so many options so close together in this price range exactly because they sell well - very well).

 

They could even make a single core version with SMT enabled to come in at a lower price point.  One thing I want AMD to do is completely stop including heatsinks (or at least increase sales of OEM CPUs).  They're not good at it, and I end up buying a cheap $12 cooler that is 3x better than anything AMD or Intel have made in the last few years.

 

I also think SMT won't scale as well on Zen as with Haswell.  I expect 15% would be a pretty good result for Zen, and I only give them that as they should have enough experience watching Intel's evolution and with their own experience with CMT.  And they already have most of the tech needed to extract much of the SMT scaling capabilities.  It took the Intel team years to get to that level from an already working implementation.

[looncraz]

There's a lot of people that are saying they hope Zen will perform well and that they'll switch to AMD if that happens. If they pick up even a quarter of the enthusiast market for CPUS and GPUs then rhey would be well on the road to recovery, of course, that depends on if people truly mean they'll change to AMD when Zen comes out. I can't wait for it to come out...

 

I will be buying a Zen CPU or two or three or more ;-)

 

It doesn't need to be better than Sandy Bridge for that to happen, I have tons of systems that I will be upgrading for myself and others in and around the middle of 2017. I can't suggest AMD for some of these builds as the budgets are too low to enter into the AMD quad core region.  Anything below that is Intel territory.  And anything above a $750 usually gets an Intel CPU as well (because they will have dedicated graphics and the APU options vanish).

 

If AMD offers some good competition on the low end, they will pick up much greater volume.  They will need a small dual core die, though, and it doesn't look like they have any intentions of creating one.  They may, however, have some harvested dies (I'm hoping so).

[Dr_Curry_MONSTER]

Still waiting for Zen before I buy my pc. If its a disappoint I'll be pissed asf. I've waited 7months + 

[TechGod]

Still waiting for Zen before I buy my pc. If its a disappoint I'll be pissed asf. I've waited 7months +

Same!

[Dr_Curry_MONSTER]

They're obvs gonna have consumer chips as there FX series was a big fail

 

Isn't Zen, so far, being advertised for servers?

[You Scrub....The Floor]

I will upgrade to Zen... For all we know it could be their last CPU series creation. (Not including rehashes like the r9 300 series)

[TechGod]

Well, AMD hasn't been clear on what's for consumers and what is for servers. Right now they're advertising the others for normal consumers, and the big zen ones as Opteron replacers. Not saying there won't be a strong consumer one, but that doesn't exactly make zen that.

No, Zen is consumer first. They changed priorities.

[looncraz]

Well, AMD hasn't been clear on what's for consumers and what is for servers.  Right now they're advertising the others for normal consumers, and the big zen ones as Opteron replacers.  Not saying there won't be a strong consumer one, but that doesn't exactly make zen that.

 

Zen for "performance" consumers is, supposedly, known as "Summit Ridge" CPUs, with up to eight cores, on AM4 (formerly known as FM3).

 

We actually know much less about the intentions with the server market than the consumer market.  We believe AMD will top out at an 8-core die simply because of what we know about Summit Ridge, but it is entirely possible that the build a larger die (12 or 16 cores) for the server market in addition to using MCM to use multiple dies per CPU.

[looncraz]

No, Zen is consumer first. They changed priorities.

Do you have a source for that?

 

I'd be surprised if AMD decided to go to the lower margin market first, unless they were intending on pushing the CPUs to market early (in which case less validations and certifications are required).

 

If this is the case, it could only be because 14nm LPP yields are higher than expected, which is actually what Global Foundries claimed some months back with LPE (they're already at 80% yields, which is amazing).

[TechGod]

Do you have a source for that?

I'd be surprised if AMD decided to go to the lower margin market first, unless they were intending on pushing the CPUs to market early (in which case less validations and certifications are required).

If this is the case, it could only be because 14nm LPP yields are higher than expected, which is actually what Global Foundries claimed some months back with LPE (they're already at 80% yields, which is amazing).

Desktop users will be happy to know that the first Zen processor out the door will be AMD’s high-end desktop CPU (AMD was very deliberate in this, it’s not an APU). AMD will be aiming high and then cascading Zen down into APUs and lower-end products.

http://www.anandtech.com/show/9231/amds-20162017-x86-roadmap-zen-is-in

[looncraz]

http://www.anandtech.com/show/9231/amds-20162017-x86-roadmap-zen-is-in

 

"Desktop users will be happy to know that the first Zen processor out the door will be AMD’s high-end desktop CPU (AMD was very deliberate in this, it’s not an APU)."

 

Fantastic.

 

I also like "Meanwhile AMD has confirmed that Zen will be shipping in 2016"