AMD Zen Performance Preview

I only went with the i5 I have now as it was in a really well priced desktop and my 8350 setup burned up. Eagerly awaiting actual numbers from Zen purchasers or reviewers. Hoping AMD is coming back, they have been left behind for quite some time now.
 
Pretty interesting comments by someone who saw the fps of the BF1 comparison.



Is BF1 really that CPU Intensive?

I don't play Call of Modern Battlefield type games, bit I haven't run into a game in forever that couldn't maintain 60fps even on relatively modest CPU's.

Last one was probably Red Orchestra 2 on large 32 player maps back before it was patched to address hight CPU use, and that was in 2012.
 
The source is a random forum poster. Made up. So if it actually hit those prices it would be amazing.

If AMD sells a CPU for $500 that competes with Intel's $1000 CPU, then Intel will drop prices to match.
 
If AMD sells a CPU for $500 that competes with Intel's $1000 CPU, then Intel will drop prices to match.

And there would be no winner. If AMD actually got a 1000$ CPU performance they are going to charge at least 800$. Else the CEOs head will be served on a platter before the next stock holder meeting.

But that reddit comment doesn't exactly seem to dictate a 1000$ performance.

Seems Zen will start from around 210$.

amd-zen-cpu-pricing.jpg
 
And there would be no winner. If AMD actually got a 1000$ CPU performance they are going to charge at least 800$. Else the CEOs head will be served on a platter before the next stock holder meeting.

But that reddit comment doesn't exactly seem to dictate a 1000$ performance.

Seems Zen will start from around 210$.

Well, that's my point. If they have a 6900K - level CPU, they're going to sell it for less than Intel, but not 50% less. They could sell them for $700-800 like you said, and they'd fly off the shelves. Well, "fly" relative to Intel's 8C/16T CPUs anyway. Now that might spur Intel to drop prices, so either way the consumer wins.

What's not known is how much it actually costs AMD to make those high end 8C/16T CPUs. Complex silicon, so probably not the greatest yields, and they have R&D costs to recoup.

Then again, I could be totally wrong. AMD might crank out these chips for <$100 each in cost and sell them for $500 in volume. This would totally destroy Intel's market, and AMD might be the new top dog for years to come.
 
New Horizons for me was more to view the product AMD has and learn a bit more about the new technology. I was not really anticipating the results and live demos, so my curiosity is piqued, I was expecting more liberal improvements so maybe will be surprised in a good way.
 
Well, that's my point. If they have a 6900K - level CPU, they're going to sell it for less than Intel, but not 50% less. They could sell them for $700-800 like you said, and they'd fly off the shelves. Well, "fly" relative to Intel's 8C/16T CPUs anyway. Now that might spur Intel to drop prices, so either way the consumer wins.

What's not known is how much it actually costs AMD to make those high end 8C/16T CPUs. Complex silicon, so probably not the greatest yields, and they have R&D costs to recoup.

Then again, I could be totally wrong. AMD might crank out these chips for <$100 each in cost and sell them for $500 in volume. This would totally destroy Intel's market, and AMD might be the new top dog for years to come.

I think a 6700K cost 15$ to make. Its not the manufactoring cost that's the thing. Its all the R&D. And Intel doesn't have to pay a 3rd party profit (GloFo).

So a price war AMD will always lose. That's one of the problems fighting an IDM.

Zen is going to pay for all AMDs future R&D. Hopefully the completed death starved GPUs as well.
 
Yea, I think everyone wants a crack at it. We're like two months out from release. I can hardly remember the last time we got a concrete metric this far out with such a new comer.
From AMD? Never. They rarely allow gpu's to be reviewed nowadays.

There was a time when AMD backed up its hype train with tidbits of real information... that came from reliable independent sources.

But that was long ago. Now they just spend money on fog machines and polish up mirrors.
 
And there would be no winner. If AMD actually got a 1000$ CPU performance they are going to charge at least 800$. Else the CEOs head will be served on a platter before the next stock holder meeting.

But that reddit comment doesn't exactly seem to dictate a 1000$ performance.

Seems Zen will start from around 210$.

amd-zen-cpu-pricing.jpg


IN all reality, Intel will likely drop pricing in response to a competetive AMD product, but not all the way to match, probably to about 20% above AMD because they have the better brand recognition and market presence, so they won't need to go all the way to parity. If AMD's challenge keeps up - however - and their reputation improves in the market long term, that's when we'll likely see Intel enter the price war for real.

AMD on the other hand, if they have a part that is truly competitive with Intel's $1000 CPU's they will of course not just pump it out at $400. They are not in the charity business. They also will not be able to charge quite as much as Intel due to the deficit in brand recognition and reputation.

There are going to be marketing and finance people at AMD trying to establish a launch price that uses price modeling to try to predict how much Intel will drop their prices in response and try to predict a good price for their product that falls just under Intel, so that they are the cheaper alternative, but not by more than 10-20%

If anything they will likely err on the high price side in their estimates, because it is much easier to correct it if they are wrong and set it too high after the initial launch by lowering the price or offering discounts, than it is to raise the price after the initial launch. Customers tend to revolt against arbitrary price hikes in the same generation of product.
 
I think a 6700K cost 15$ to make. Its not the manufactoring cost that's the thing. Its all the R&D. And Intel doesn't have to pay a 3rd party profit (GloFo).

So a price war AMD will always lose. That's one of the problems fighting an IDM.

Zen is going to pay for all AMDs future R&D. Hopefully the completed death starved GPUs as well.

2016 was a good year for AMD, the company has put back solid cashflow and that is important. The success of Zen and Vega even if on the greater picture we say that AMD takes 10% back from Intel Corp and 10% from NVidia, that corrolates into billions for AMD to distribute to R&D and that in turn puts out better products.

While Polaris took a while to settle, against the 900 series that it was designed to go against, it does well and the RX470 is probably the best affordable value for money card. Where I am from the 1050ti is the same price as the 470 Nitro 4GB OC and the 1050ti is not in the same league at all.

Not to digress, I am holding out that Zen looks more promising than I expected and hopefully AMD keeps heading in the right direction.
 
I think a 6700K cost 15$ to make. Its not the manufactoring cost that's the thing. Its all the R&D. And Intel doesn't have to pay a 3rd party profit (GloFo).

So a price war AMD will always lose. That's one of the problems fighting an IDM.

Zen is going to pay for all AMDs future R&D. Hopefully the completed death starved GPUs as well.

Well, keep in mind that AMD is also a much much leaner company. They don't have the overhead in armies of engineers salaries like Intel has.

That being said, Intel also has almost $18 billion in cash on hand, so they can afford to run at a loss short term in order to try to break AMD, and make sure they don't have the money to properly fund the development of Zen+
 
2016 was a good year for AMD, the company has put back solid cashflow and that is important. The success of Zen and Vega even if on the greater picture we say that AMD takes 10% back from Intel Corp and 10% from NVidia, that corrolates into billions for AMD to distribute to R&D and that in turn puts out better products.

While Polaris took a while to settle, against the 900 series that it was designed to go against, it does well and the RX470 is probably the best affordable value for money card. Where I am from the 1050ti is the same price as the 470 Nitro 4GB OC and the 1050ti is not in the same league at all.

Not to digress, I am holding out that Zen looks more promising than I expected and hopefully AMD keeps heading in the right direction.

2016 cash flow is still in the red ;), at least so far. And to get that kind of marketshare uptake, is not easy to do unless you have products that are above and beyond their competitors, and they don't have them yet.
 
Well, keep in mind that AMD is also a much much leaner company. They don't have the overhead in armies of engineers salaries like Intel has.

That being said, Intel also has almost $18 billion in cash on hand, so they can afford to run at a loss short term in order to try to break AMD, and make sure they don't have the money to properly fund the development of Zen+

Well nobody likes to work for free, including AMD employees. Something some people tend to forget. Lisa Su also publicly said that she didn't want AMD to be the discount company. Aka she wants higher prices and margins.

The price of Zen will pretty much match the performance. If it actually delivers 6900K performance it wont be 500$. It may not even be 800$ but rather 900$. If its 6900K performance AND beating it in perf/watt it will be 1000$ or more. AMD also wanted 849$ for Fiji before the 980ti launch. not to mention the known 800$ FX. Or how Polaris prices went up before launch.
 
Well nobody likes to work for free, including AMD employees. Something some people tend to forget. Lisa Su also publicly said that she didn't want AMD to be the discount company. Aka she wants higher prices and margins.


They need higher margins, 30% for any company in any industry is too low, 40% good, 30% bad lol. The only reason a company would go that low is because they are getting killed in the marketplace. And this is what we saw the past 10 years. To get higher margins the node will help but not all of it, they need higher prices.
 
They need higher margins, 30% for any company in any industry is too low, 40% good, 30% bad lol. The only reason a company would go that low is because they are getting killed in the marketplace. And this is what we saw the past 10 years. To get higher margins the node will help but not all of it, they need higher prices.

Ye, just look at AMDs R&D budget. It should as an absolute minimum be 3x higher than it is now. That's 500M$ more per quarter they need for that alone.
 
2016 cash flow is still in the red ;), at least so far.

Existing debts reflect that, but AMD made money this year, as a year it was green.

The old enemy of your enemy is your friend works for AMD here, right now Nvidia and Intel have emerged as the giant players in the same space, neither is capable to take on Samsung, Apple in the mobility market where the major money is at but both occupy major share in the PC market. Nvidia are still reliant on AMD and Intel X86 while Intel has now become dependant on relations with AMD in Graphics IP which will be profitable for AMD longterm given that Intel and Nvidia have burnt bridges a long time ago makes AMD, Intel's sole out from being completely locked out the GPU stakes. The impending head to head between the two giants lets AMD somewhat fly under radar, while helping themselves to market share.

I just like that AMD under Lisa Su has direction again, two years ago it looked like they were dead certain to liquidate, but she has done very well on limited resources to make sure AMD diverted its resources to investor friendly markets, what they need and we as the end user is a strong product to leverage back that much needed profits, as above AMD taking 20% of the combined share will not hurt Intel or Nvidia, but it would corrolate to billions for AMD.
 
I think a 6700K cost 15$ to make. Its not the manufactoring cost that's the thing. Its all the R&D. And Intel doesn't have to pay a 3rd party profit (GloFo).

So a price war AMD will always lose. That's one of the problems fighting an IDM.

Zen is going to pay for all AMDs future R&D. Hopefully the completed death starved GPUs as well.

Any large CPU/GPU is only going to be a few bucks in materials, true. The manufacturing cost is much, much higher though, it's probably on the order of at least $200 per CPU. If you have your own fab, you have to buy equipment, operate the clean room, pay workers, etc. If you don't have your own fab, you pay someone else. Then you need to QA those chips and deal with flaws. You can drop that cost when you start making them in large volumes of course, but you still have to deal with those costs.
 
Existing debts reflect that, but AMD made money this year, as a year it was green.

The old enemy of your enemy is your friend works for AMD here, right now Nvidia and Intel have emerged as the giant players in the same space, neither is capable to take on Samsung, Apple in the mobility market where the major money is at but both occupy major share in the PC market. Nvidia are still reliant on AMD and Intel X86 while Intel has now become dependant on relations with AMD in Graphics IP which will be profitable for AMD longterm given that Intel and Nvidia have burnt bridges a long time ago makes AMD, Intel's sole out from being completely locked out the GPU stakes. The impending head to head between the two giants lets AMD somewhat fly under radar, while helping themselves to market share.

I just like that AMD under Lisa Su has direction again, two years ago it looked like they were dead certain to liquidate, but she has done very well on limited resources to make sure AMD diverted its resources to investor friendly markets, what they need and we as the end user is a strong product to leverage back that much needed profits, as above AMD taking 20% of the combined share will not hurt Intel or Nvidia, but it would corrolate to billions for AMD.


Well no existing debt doesn't play into that, margins (gross or net) and gross profits (still in the red for the year) don't play into debt, debt is a separate line item. now if you are talking about net profits, still in the red, and that will include interest paid on debt.

yeah the whole Intel and AMD thing, just go back in my posts, people are making a mountain out of a molehill, the business ramifications and anti trust laws, won't allow such an "arrangement" that would push or even touch the only other player in the graphics industry.

The rest of the stuff you typed, yeah I will agree with that, but this is a very long and hard road for them, not something that is going to happen in a few Q's, its years and in those years, we have no idea how the dynamics can change because of the other two companies.

It took YEARS for AMD to be in the position its in right now, so to expect it back and healthy as quickly as you seem to think it will, nope. Is it healthier than before sure, but to sustain what they have done till this point, Intel and nV have to be stagnant. Which they will not stay that way in the face of any possible competition. Always fall faster then coming back.

And if you don't think Intel and nV won't fight even for that 10% per market, think again, they fight AMD and they will fight hard. Didn't we see that with the 1060? We don't know what Intel plans yet, but don't expect them to just lay on their backside if Zen is going to push them on market share.

Just because nV has their release schedules a bit after AMD's doesn't mean they don't care, certain things have to be done at certain times based on their internal schedule, speeding them up can only be done a certain amount, can't move mountains to get a product that is so complex out faster, there is only a certain amount of latitude they have. Something goes for Intel, in the meantime AMD might gain some things, if they have a product to do it, otherwise things will stay flat as always.
 
Last edited:
Oh you mean Intel has something cooking that is great but decided to dump out Kaby Lake at a less then 2% gain in performance. Only thing good out Kaby Lake was for mobile.
 
Any large CPU/GPU is only going to be a few bucks in materials, true. The manufacturing cost is much, much higher though, it's probably on the order of at least $200 per CPU. If you have your own fab, you have to buy equipment, operate the clean room, pay workers, etc. If you don't have your own fab, you pay someone else. Then you need to QA those chips and deal with flaws. You can drop that cost when you start making them in large volumes of course, but you still have to deal with those costs.

No, its quite cheap.

As someone once said. The first chip cost a billion, the next a penny. R&D is all the cost. You can also look at AMDs payments to Glofo, they are quite small. Last quarter they paid 284M$ and that includes a juicy margin to GloFo. And they got what in return for that? 20 million chips of various things?
 
Ultimately how long a company stays behind is based on its product. What I know know is AMD have enticing technology and bringing it together seems to be the focus of Zen/Vega, AMD also still appeals to growing market space.

In the absence of any imperical data, zen needed to check the boxes:

1) Cache subsystem needed to be considerably faster
2) IMC had to be considerably faster
3) IPC needs to yield substantial growth.

I believe the first two can be responsible on their own for up to 40% IPC gains that AMD targeted, if we assume that no company lays down bare minimum targets to exceed that AMD would need to re-do the entire arch which the did, they dropped CMT shared FPU design with deep pipelines for a SMT design and a more effient design. The part does not seem relient on clockspeed to produce results. My hope is for around Haswell/Devils Canyon type IPC with a much better scalablitity and SMT.

In short I feel a lot more comfortable with the first hurdle, given that Zambezi didn't come close to a mainstream i7 and barely got over an i5. CMT kind of saved them in high integer loads but that was overshadowed by extremely bad scaling
 
Oh you mean Intel has something cooking that is great but decided to dump out Kaby Lake at a less then 2% gain in performance. Only thing good out Kaby Lake was for mobile.


It all depends on schedules man, projects like these don't happen over night, Intel's release plan is 1.5 years per gen. Just to give you an idea of "small updates" 80% of the microcode has to be redone for just a Tick to Tock changes.

So if Intel started Kaby Lake 1.5 years ago, where was Zen at that point on Intel's radar? Not even a blip. But as Kaby Lake is coming out now and more R&D shifts over to Cannon Lake, which is on 10nm, will their response be just lower node size? Will that be enough? Looks to be enough right? More than enough to maintain an advantage. So the real response to Zen probably won't come out till Coffee Lake, if they need to make any major changes to maintain an advantage.
 
Ultimately how long a company stays behind is based on its product. What I know know is AMD have enticing technology and bringing it together seems to be the focus of Zen/Vega, AMD also still appeals to growing market space.

In the absence of any imperical data, zen needed to check the boxes:

1) Cache subsystem needed to be considerably faster
2) IMC had to be considerably faster
3) IPC needs to yield substantial growth.

I believe the first two can be responsible on their own for up to 40% IPC gains that AMD targeted, if we assume that no company lays down bare minimum targets to exceed that AMD would need to re-do the entire arch which the did, they dropped CMT shared FPU design with deep pipelines for a SMT design and a more effient design. The part does not seem relient on clockspeed to produce results. My hope is for around Haswell/Devils Canyon type IPC with a much better scalablitity and SMT.

In short I feel a lot more comfortable with the first hurdle, given that Zambezi didn't come close to a mainstream i7 and barely got over an i5. CMT kind of saved them in high integer loads but that was overshadowed by extremely bad scaling

No the main reason for the IPC gain, is FP unit changes, no longer hampering 50% of the ALU's, 50% of the ALU's were going unused in many circumstances.

Once that bottleneck was removed, then the other changes help it get more. I wouldn't be surprised if 50% of the IPC increase was right from there.
 
Is BF1 really that CPU Intensive?

I don't play Call of Modern Battlefield type games, bit I haven't run into a game in forever that couldn't maintain 60fps even on relatively modest CPU's.

Last one was probably Red Orchestra 2 on large 32 player maps back before it was patched to address hight CPU use, and that was in 2012.

Frostbite engines are intense, I had a G4400 ITX setup and it absolutely destroyed the G4400. Another game that surprisingly wrecked a G4400 was the Forest despite recommending a Core 2 , there is no way in hell a Core 2 runs the Forest.
 
No the main reason for the IPC gain, is FP unit changes, no longer hampering 50% of the ALU's, 50% of the ALU's were going unused in many circumstances.

Once that bottleneck was removed, then the other changes help it get more. I wouldn't be surprised if 50% of the IPC increase was right from there.

That and Zambezi vs Sandy showed Zambezi's cache and IMC to be a third of a Sandy
 
Frostbite engines are intense, I had a G4400 ITX setup and it absolutely destroyed the G4400. Another game that surprisingly wrecked a G4400 was the Forest despite recommending a Core 2 , there is no way in hell a Core 2 runs the Forest.

Not that intensive for an 8 core CPU.
 
That and Zambezi vs Sandy showed Zambezi's cache and IMC to be a third of a Sandy


Cache and IMC alone can't cover the performance difference between those two chips, Its impossible when the bottleneck is else where, which is a much harder bottleneck then the cache and IMC
 
No, its quite cheap.

As someone once said. The first chip cost a billion, the next a penny. R&D is all the cost. You can also look at AMDs payments to Glofo, they are quite small. Last quarter they paid 284M$ and that includes a juicy margin to GloFo. And they got what in return for that? 20 million chips of various things?
That's true for "popcorn" chips, i.e., the smaller designs. Large CPUs and GPUs, especially as they're going to smaller and smaller feature sizes run into other problems and keep the costs up. Fewer total dies per wafer, just due to the design size. Fewer GOOD dies per wafer due to the feature size. More failures during packaging due to the higher pin counts.

I'm not arguing that R&D is the lion's share of cost, but large designs aren't as cheap to make as things like memory chips and such.
 
I think I have a solid example here that replicates what AMD showed us. This is a 6950X, 8C/16T run at all cores clocked and locked at 3.5GHz, with dual channel 2400MHz RAM. (Ram being Quad or Dual, 8GB or 16GB really made no meaningful impact.) This is using the NEW Blender file download posted on its Horizon site using the 150 sample size. FYI, the file uses the same exact file name as the previous, so be careful if you still have both versions.

3500-2400Dual-Blender150.png


So all that said, Zen seems to be instep with Broadwell-E in terms of IPC, at least in terms of Blender.
 
Last edited:
As an AMD fan I am somewhat hesittent to outright committ to IPC as Blender tests SMT more than IPC, however I am far more comfortable with the possible level of IPC than I was. Do I think AMD wins, no only a fools hope at best, do I think AMD have a very good product yes and I would absolutely love to put a 4C/8T CPU into a ITX build. *holds thumbs it is a $150 part*

AMD doesn't have to outright beat Intel in performance to start winning again (would be a plus but very unlikely), it is more important for them improve their margins again and get back into the black to fund future R&D.
 
Were you able to reproduce their results on your 6950x with two cores disabled at 150 samples?

Because I re-ran the benchmark with 150 samples, and I still have some serious doubts.

Well, from what i see on screen it looks like within a second from live demo.

I am damn near dead on. The results from my timer using the live stream are ~35.75 seconds.
 
Glad to hear you replicated their result Kyle, makes it far less likely that anything was being hidden or distorted. Now the question is will this result carry over into other tasks.
 
Huh. Then my results I posted in the other thread make absolutely no sense. (45.04s with the 150 sample file on my 3930k @ 4.8)

Blender must have some sort of funky core scaling issues going on for the results to wind up as they do.
Well, you are trying to match results from a different CPU architecture with less cores. I am not sure what you are trying to accomplish with that hardware in relation to what AMD showed us this week.
 
Back
Top