AMD Engineer: Developers Given Total Freedom with 'Zen'

[Clyne]

I wish I had some better cooling to see what I could do with my 8320, running 4.4 Ghz at 1.284 volts right now with a be quiet! dark rock pro 2 and I get about 62-63 degrees with Prime95.

Oh nice, to bad I'm just using stock cooler so I can't get far with oc if I want acceptable temps.

[patrickjp93]

can also be affected by instruction set

IPC is Instruction-Set agnostic. Now, as per what that IPC can turn into performance, that is instruction set dependent.

[limepanda]

I believe that the given estimates for Zen by AMD in regards to IPC were roughly 10% slower than Haswell or Ivy Bridge. This means they will be at least 2 generations behind Intel and then on a larger process than them as well. I sure hope Zen has some serious core count, energy efficiency, and base clock overhead to warrant a $250 price tag. These are not going to be cheap processors, and they may suck pretty bad compared to Intel's next step.

What I've gathered from the most recent information is that zen is going to match skylake clock for clock, will have a considerably higher base clock than skylake, is going to match at least haswell for efficiency, will have 8 full cores and 16 threads, and will be priced around $200. If that's not enough to make Intel shit themselves then I don't know what would.

[patrickjp93]

What I've gathered from the most recent information is that zen is going to match skylake clock for clock, will have a considerably higher base clock than skylake, is going to match at least haswell for efficiency, will have 8 full cores and 16 threads, and will be priced around $200. If that's not enough to make Intel shit themselves then I don't know what would.

Except none of this is confirmed. The theoretical best IPC AMD is going to get is +/- 3% of Haswell. Clock speeds are an unknown at this point, but 8 cores in 95W at 14nm FF are not going to be busting much past 3.6GHz at turbo (5960X is 140W at 3.5GHz, though their TDP measurement algorithms are very different). As for the price... HAHAHAHAHAHA! With AMD so desperate for money, that's one doozy of a joke.

[MageTank]

Except none of this is confirmed. The theoretical best IPC AMD is going to get is +/- 3% of Haswell. Clock speeds are an unknown at this point, but 8 cores in 95W at 14nm FF are not going to be busting much past 3.6GHz at turbo (5960X is 140W at 3.5GHz, though their TDP measurement algorithms are very different). As for the price... HAHAHAHAHAHA! With AMD so desperate for money, that's one doozy of a joke.

We still don't have 100% confirmations on the 8 cores, 16 threads yet either. They keep saying "8 core Zen" but i have this nagging feeling that it might be 4 cores, 8 threads. 

 

I expect similar clock speeds to current FM2+ chips, with it ranging from lower end chips having around 3.2ghz, with high end chips reaching 3.7-3.9ghz on turbo. No guess on how OCing will be, as it would be a complete shot in the dark at this point. Though, as i have said plenty of times, if it matches Haswell in IPC, and performs on par with it, it will be a success. Matching Haswells performance (and assuming the 8 core, 16 thread thing isn't made up) then you have a CPU that hits both Z97 and X99 platforms, and is capable of AVX3. Sign me up.

[patrickjp93]

We still don't have 100% confirmations on the 8 cores, 16 threads yet either. They keep saying "8 core Zen" but i have this nagging feeling that it might be 4 cores, 8 threads.

I expect similar clock speeds to current FM2+ chips, with it ranging from lower end chips having around 3.2ghz, with high end chips reaching 3.7-3.9ghz on turbo. No guess on how OCing will be, as it would be a complete shot in the dark at this point. Though, as i have said plenty of times, if it matches Haswell in IPC, and performs on par with it, it will be a success. Matching Haswells performance (and assuming the 8 core, 16 thread thing isn't made up) then you have a CPU that hits both Z97 and X99 platforms, and is capable of AVX3. Sign me up.

Based on the core diagram, I'm doubting AVX 512 support (3.2 btw, not 3).

[MageTank]

Based on the core diagram, I'm doubting AVX 512 support (3.2 btw, not 3).

I thought the block diagram showed 2x 256 FMAC. Can't they combine both to be able to handle AVX3(.2)? 

 

EDIT: Found this: 

[Prysin]

I thought the block diagram showed 2x 256 FMAC. Can't they combine both to be able to handle AVX3(.2)? 

 

EDIT: Found this: 

sorry for not bothering to read up on this myself. but what is AVX3 and what does it do for performance?

[MageTank]

sorry for not bothering to read up on this myself. but what is AVX3 and what does it do for performance?

 

This will summarize it:

 

 

 

Intel® Advanced Vector Extensions 512 (Intel® AVX-512)

In the future, some new products will feature a significant leap to 512-bit SIMD support. Programs can pack eight double precision and sixteen single precision floating numbers within the 512-bit vectors, as well as eight 64-bit and sixteen 32-bit integers. This enables processing of twice the number of data elements that IntelAVX/AVX2 can process with a single instruction and four times the capabilities of Intel SSE.

 

Intel AVX-512 instructions are important because they open up higher performance capabilities for the most demanding computational tasks. Intel AVX-512 instructions offer the highest degree of compiler support by including an unprecedented level of richness in the design of the instruction capabilities.

 

Intel AVX-512 features include 32 vector registers each 512-bit wide and eight dedicated mask registers. Intel AVX-512 is a flexible instruction set that includes support for broadcast, embedded masking to enable predication, embedded floating point rounding control, embedded floating-point fault suppression, scatter instructions, high speed math instructions, and compact representation of large displacement values.

 

 

 

 

[HalGameGuru]

We still don't have 100% confirmations on the 8 cores, 16 threads yet either. They keep saying "8 core Zen" but i have this nagging feeling that it might be 4 cores, 8 threads. 

 

I expect similar clock speeds to current FM2+ chips, with it ranging from lower end chips having around 3.2ghz, with high end chips reaching 3.7-3.9ghz on turbo. No guess on how OCing will be, as it would be a complete shot in the dark at this point. Though, as i have said plenty of times, if it matches Haswell in IPC, and performs on par with it, it will be a success. Matching Haswells performance (and assuming the 8 core, 16 thread thing isn't made up) then you have a CPU that hits both Z97 and X99 platforms, and is capable of AVX3. Sign me up.

 

They had questionable justifiability on calling a core a "module" and the CMT redundant hardware "cores", on an SMT design I don't think they could get away with calling SMT threads cores. If they claim 8 core, it had better be 8 hardware cores and 16 threads if its SMT enabled.

[limepanda]

Except none of this is confirmed. The theoretical best IPC AMD is going to get is +/- 3% of Haswell. Clock speeds are an unknown at this point, but 8 cores in 95W at 14nm FF are not going to be busting much past 3.6GHz at turbo (5960X is 140W at 3.5GHz, though their TDP measurement algorithms are very different). As for the price... HAHAHAHAHAHA! With AMD so desperate for money, that's one doozy of a joke.

 

We'll see, Skylake was only around 5% up on Haswell, AMD could do better than +/-3% between now and release. AMD have made great gains in efficiency recently and I think Zen will be very competitive as far as that's concerned (also rumors state that Zen will be 16nm FF through TSMC). As for price, AMD is desperate to gain back market share and a low price could be how they can achieve that. Better to have a million people buy a $250 CPU than 250 thousand people buy a $1000 CPU.

[MageTank]

They had questionable justifiability on calling a core a "module" and the CMT redundant hardware "cores", on an SMT design I don't think they could get away with calling SMT threads cores. If they claim 8 core, it had better be 8 hardware cores and 16 threads if its SMT enabled.

I totally agree. If it really is 8 cores, 16 threads, i will just have to give the Skylake build i am planning on doing to my father, or younger brother. 

[Clyne]

I totally agree. If it really is 8 cores, 16 threads, i will just have to give the Skylake build i am planning on doing to my father, or younger brother. 

I'm also interested into how many PCIE lanes you get....it's not like Intel where you have to go X to have 3 way SLI or higher.

[MageTank]

We'll see, Skylake was only around 5% up on Haswell, AMD could do better than +/-3% between now and release. AMD have made great gains in efficiency recently and I think Zen will be very competitive as far as that's concerned (also rumors state that Zen will be 16nm FF through TSMC). As for price, AMD is desperate to gain back market share and a low price could be how they can achieve that. Better to have a million people buy a $250 CPU than 250 thousand people buy a $1000 CPU.

Sorry in advance for what will probably be a double post, but i had to do it. AMD will not be selling the high end Zen CPU's for $200. Not on our best days. I imagine they will put it either on par with the Z97 i7's, or X99 i7's. You can probably bet its going to be roughly $300-$350 for the high end Zen's, and the rest will trickle down to compete with the rest of Intel's offerings. AMD has said before that they do not want to be known as the "cheap brand". Besides, if Zen competes with Haswell head to head, and offers what they promise it will, they are well within their right to charge i7 prices for them, and it will still sell. 

 

BTW: Millions of people is kind of a large stretch. Not that many people actually build their own computers, and the ones that do most definitely do not aim for enthusiast level hardware. Most people buy pre-builts. 

 

Aside from that, i do agree that Zen will be competitive with Skylake. It might not be exactly as fast as Skylake, but it won't have to be if it offers better features. I am 100% certain that IF AMD offers 512bit  AVX3.2, it will be a home run.

 

 

I'm also interested into how many PCIE lanes you get....it's not like Intel where you have to go X to have 3 way SLI or higher.

It's hyper transport. Plenty enough to support 4 way SLI and Crossfire.

 

EDIT: I should add that it depends mostly on the board manufacturers (Remember, its north + south bridge for AMD, partly why ITX board selections end up being terrible). The northbridge will have X amount of lanes, while the south bridge has its own set of lanes, and HyperTransport acts as a point to point bus. Hope this makes sense.

 

EDIT 2: I'll double clarify with an example. 990FX chipset had 38 PCIE 2.0 lanes, and the north bridge had 4 PCIE 2.0 lanes dedicated to A-Link Express 3. 

[Swndlr]

October 2016? Intel will have it beat before it comes out. (if they don't have it beat already)

This scares me, knowing that Intel probably already has something way better that will just crush Zen on release

[MageTank]

This scares me, knowing that Intel probably already has something way better that will just crush Zen on release

"Way better" is going a bit far out. Intel is reaching the point where any improvement made is less substantial because its already highly refined. That is why these last few architectural changes, have not been that substantial. Broadwell from DC was a sidegrade, being slightly faster. Skylake from DC was only 5% faster on its best days (with most of the area's they improved in, coming from the increased memory bandwidth obtained by DDR4, and Skylakes superior IMC). I think Kaby Lake will end up as another Broadwell-esque upgrade (or side grade), with the next big push coming from Cannonlake. 

 

Zen's #1 fear should be Skylake-E. We won't be seeing that for a little while, so as long as AMD pushes to get Zen out some time within 2016 (and not 2017), they should be in the clear for the most part.

[You_are_a_cunt]

Well, if Zen proves to be as good as they claim, I might just start setting some money aside for an upgrade.

If it's a flop, I can always use the money to switch (back) to Intel.

[limepanda]

Sorry in advance for what will probably be a double post, but i had to do it. AMD will not be selling the high end Zen CPU's for $200. Not on our best days. I imagine they will put it either on par with the Z97 i7's, or X99 i7's. You can probably bet its going to be roughly $300-$350 for the high end Zen's, and the rest will trickle down to compete with the rest of Intel's offerings. AMD has said before that they do not want to be known as the "cheap brand". Besides, if Zen competes with Haswell head to head, and offers what they promise it will, they are well within their right to charge i7 prices for them, and it will still sell. 

 

BTW: Millions of people is kind of a large stretch. Not that many people actually build their own computers, and the ones that do most definitely do not aim for enthusiast level hardware. Most people buy pre-builts. 

 

Aside from that, i do agree that Zen will be competitive with Skylake. It might not be exactly as fast as Skylake, but it won't have to be if it offers better features. I am 100% certain that IF AMD offers 512bit  AVX3.2, it will be a home run.

 

 

 

It doesn't matter that they no longer want to be known as the cheap option, they need to gain back market share asap.

 

$200 is a bit of a stretch but I think sub $300 is definitely possible. 

 

I said a million, which I think is perfectly reasonable. The idea behind what I said is that if they offer a low price in order to get more people switching to AMD, it becomes easier to keep them with AMD which is beneficial in the long run for them.

 

I believe that Zen could be what saves AMD if they play their cards right.

[MageTank]

It doesn't matter that they no longer want to be known as the cheap option, they need to gain back market share asap.

 

$200 is a bit of a stretch but I think sub $300 is definitely possible. 

 

I said a million, which I think is perfectly reasonable. The idea behind what I said is that if they offer a low price in order to get more people switching to AMD, it becomes easier to keep them with AMD which is beneficial in the long run for them.

 

I believe that Zen could be what saves AMD if they play their cards right.

I still do not see one million people in this world building their own computers though. As large of a group as we might think we are, PC building is kind of a niche market in and of itself. Most of the hardware ends up in pre-built machines, or purchased from boutique builders. That being said, i am not saying you are wrong. I do agree with you on the market share thing. However, they don't need to price that low in order to get it back. 

 

If Zen is 8 cores, 16 threads and supports AVX3.2, it is an automatic win for video editors/workstation setups. The fact that it has the potential to match Haswell's IPC also means it will be just as viable to use for gaming, making it a great multi purpose CPU. This gives people that have a difficult time choosing between Z97 and X99, an easier, two in one solution. Not to mention the fact that their chipsets can support quad SLI without the need of the ever expensive PEX 8747 bridges, means people that want to run 4 way SLI without spending $300-$500 on a board, can do so. That saves the consumer money on that front too.

 

I think if they price their 8 core, 16 thread chip at $300, it would sell like hot cakes.

[Notional]

We'll see, Skylake was only around 5% up on Haswell, AMD could do better than +/-3% between now and release. AMD have made great gains in efficiency recently and I think Zen will be very competitive as far as that's concerned (also rumors state that Zen will be 16nm FF through TSMC). As for price, AMD is desperate to gain back market share and a low price could be how they can achieve that. Better to have a million people buy a $250 CPU than 250 thousand people buy a $1000 CPU.

 

The high end ZEN's will not be cheap but i7-e pricing I would presume. We should see mid-end to high end workstation chips in the 3-400$ mark.

 

How many they sell to what price is basic supply and demand along with variable costs. If the cost of the chips means that those million chips results in lower profits than selling 250k, obviously they would go for the 250k. Basic micro economy here.

[limepanda]

I still do not see one million people in this world building their own computers though. As large of a group as we might think we are, PC building is kind of a niche market in and of itself. Most of the hardware ends up in pre-built machines, or purchased from boutique builders. That being said, i am not saying you are wrong. I do agree with you on the market share thing. However, they don't need to price that low in order to get it back. 

 

If Zen is 8 cores, 16 threads and supports AVX3.2, it is an automatic win for video editors/workstation setups. The fact that it has the potential to match Haswell's IPC also means it will be just as viable to use for gaming, making it a great multi purpose CPU. This gives people that have a difficult time choosing between Z97 and X99, an easier, two in one solution. Not to mention the fact that their chipsets can support quad SLI without the need of the ever expensive PEX 8747 bridges, means people that want to run 4 way SLI without spending $300-$500 on a board, can do so. That saves the consumer money on that front too.

 

I think if they price their 8 core, 16 thread chip at $300, it would sell like hot cakes.

 

I don't know, going by this AMD sell around 35 million desktop CPU's per year. I don't think it's unreasonable to think that they'll sell a million to system builders over the lifespan of the chip.

 

I agree, as long as Zen lives up to what's being said about it now it'll provide some very serious competition for Intel.

 

 

The high end ZEN's will not be cheap but i7-e pricing I would presume. We should see mid-end to high end workstation chips in the 3-400$ mark.

 

How many they sell to what price is basic supply and demand along with variable costs. If the cost of the chips means that those million chips results in lower profits than selling 250k, obviously they would go for the 250k. Basic micro economy here.

 

I doubt they'll price them that high, like I said in another reply, AMD are desperate to gain market share asap. This is do or die for them now. The lower they price their cpu's the more they will sell thus making up the profit difference between low and high price products, this is also basic economics.

[Notional]

I don't know, going by this AMD sell around 35 million desktop CPU's per year. I don't think it's unreasonable to think that they'll sell a million to system builders over the lifespan of the chip.

 

I agree, as long as Zen lives up to what's being said about it now it'll provide some very serious competition for Intel.

 

I doubt they'll price them that high, like I said in another reply, AMD are desperate to gain market share asap. This is do or die for them now. The lower they price their cpu's the more they will sell thus making up the profit difference between low and high price products, this is also basic economics.

 

Ehm no? Market share is never a objective in itself. What AMD needs are profits. If they get higher profits selling 250k instead of 1 million, then that's what they are going to do because that makes them more money. Higher sales/revenue is not a good thing if cumulative profits goes down. Market share is nice, but only if it makes you money.

 

Right now those 35 million CPU's makes them little money. Better to sell 10 million Zen's at a higher price margin and thus higher profits, if the overall profits outmatch the 35 million current sales. In reality AMD will of course try to get more than 35 million sales at different price classes, but the point is that selling more for the sake of selling more makes little sense long term. There are a lot of variables to take into account here (subsidising profits to gain marketshare could be a strategy for instance, but only makes sense short term and might hurt profits in the long run). 

[limepanda]

Ehm no? Market share is never a objective in itself. What AMD needs are profits. If they get higher profits selling 250k instead of 1 million, then that's what they are going to do because that makes them more money. Higher sales/revenue is not a good thing if cumulative profits goes down. Market share is nice, but only if it makes you money.

 

Right now those 35 million CPU's makes them little money. Better to sell 10 million Zen's at a higher price margin and thus higher profits, if the overall profits outmatch the 35 million current sales. In reality AMD will of course try to get more than 35 million sales at different price classes, but the point is that selling more for the sake of selling more makes little sense long term. There are a lot of variables to take into account here (subsidising profits to gain marketshare could be a strategy for instance, but only makes sense short term and might hurt profits in the long run). 

 

Of course it is! If you gain market share you are likely to hold on to it. You can then start ramping up prices to get your profits. Think of market share as a long term investment, once you get people to invest in your technology they are likely to upgrade with your hardware. Here's an example: If someone decides to buy Zen because it offers similar performance to Intel for a better price then they're going to buy an AMD compatible motherboard. When the time comes to upgrade their cpu that person is more likely to buy another AMD cpu even if it's a bit more expensive than the last generation because they already own an AMD motherboard and have invested in that ecosystem and a lot of the time feel emotionally attached to "their brand". This makes market share a very important objective. Just look at how it's helped Nvidia, AMD has had a better price to performance ratio than Nvidia for 3 generations now yet Nvidia still dominates. Why? The long lasting effects of market share.

 

Your argument makes no sense? If you price the product higher, sure, you'll make more profit per sale, but your sales will be low and you'll lose market share. If you price a product lower you'll make less profit per sale, but your overall sales will be high and you'll gain market share. The profit side of the equation will work out about the same, however by using a low price strategy AMD can gain back market share over Intel which is critical at this point and you couldn't do with a high pricing strategy. 

[patrickjp93]

I thought the block diagram showed 2x 256 FMAC. Can't they combine both to be able to handle AVX3(.2)?

EDIT: Found this:

In theory, but the use of mantissa extension would make AVX 3.2 run much slower than needed. 2 different threads each running AVX 2 instructions using different FMACs would likely be faster.

[patrickjp93]

We'll see, Skylake was only around 5% up on Haswell, AMD could do better than +/-3% between now and release. AMD have made great gains in efficiency recently and I think Zen will be very competitive as far as that's concerned (also rumors state that Zen will be 16nm FF through TSMC). As for price, AMD is desperate to gain back market share and a low price could be how they can achieve that. Better to have a million people buy a $250 CPU than 250 thousand people buy a $1000 CPU.

Other way around where production costs are concerned, and investors.