ZEN ES Benchmark from french hardware Magazine

[R0H1T]

No, they won't. iGPU is a godsend for these types of systems because it's good enough to do what people want to do with them -- surf the web, play videos, play Facebook/Flash games, etc.

If you want to play real 3D games, then you buy a dGPU. That's pretty much the only reason to have one these days, unless you are trying to upgrade the media functionality of an ancient system (but most people don't upgrade systems, they buy new ones).

Wasn't the argument awhile back something to the tune of "an (AMD) APU must be better than a lowly (750Ti?) dGPU for it to replace an i3/i5/i7 plus the said dGPU" so now the buyers of these i3/i5/i7 systems surf web using their OCed IGP?

The SR7 is clearly targeted at the 5820/5830k buyers, 5960x if we're being generous. Anyone who wants/needs an IGP can always wait for the Zen APU, it's not like AMD's IGP will be less than anything the competition has, in fact Zen APU will have class leading graphics & you can quote me on that.

The history of civilization is full of unsound decisions. Aside from buying something future proof (perf/platform features) it could be a mix of succesful salespeople, ego driven desires (e.g. bragging), single/two threaded browsers (Firefox w/ plugin container process not long ago), etc. No need to wrap a technical fact based discussion around it.

There's no cure for greed, envy, lust, gluttony, pride et al. you know the seven deadly sins. The world we live in is living proof of that

 

[railroadmaster]

No, they won't. iGPU is a godsend for these types of systems because it's good enough to do what people want to do with them -- surf the web, play videos, play Facebook/Flash games, etc.

If you want to play real 3D games, then you buy a dGPU. That's pretty much the only reason to have one these days, unless you are trying to upgrade the media functionality of an ancient system (but most people don't upgrade systems, they buy new ones).

A dedicated GPU isn't just for games it can also be useful GPU computing on workstations.

 

[itsmydamnation]

The history of civilization is full of unsound decisions. Aside from buying something future proof (perf/platform features) it could be a mix of succesful salespeople, ego driven desires (e.g. bragging), single/two threaded browsers (Firefox w/ plugin container process not long ago), etc. No need to wrap a technical fact based discussion around it.

So they want the CPU epeen but not that GPU epeen?

 

[Arachnotronic]

So who exactly is buying an i5/i7 level desktop system (4core 8 thread) to "surf the web, play videos, play Facebook/Flash games, etc"? Sounds like the I3/Pentium range to me.

Somebody who buys an all-in-one PC from an OEM who will not sell you a system with a nice industrial design, nice screen, and other premium features without also selling you the nice processor.

It's in the OEMs' best interests to find ways to get consumers to buy more expensive computers. Basic PC biz stuff

Seriously, the population of people who buy tower desktops is on the decline (I'd imagine it's mostly businesses and gamers at this point), and the subset of that population that builds their own systems and gives a hoot about the CPU inside? Much, much, much smaller.

Also, a lot of people in the enthusiast community bemoan that a 5 year old SNB still works just fine. Most people don't *want* to have to spend money on new PCs (they are not cheap), so they figure that they should buy the nice system with the fastest processor that they can afford so that it lasts longer. The idea of a system lasting for 5 years+ and still being awesome and doing everything they want it to do is a pretty compelling feature.

Unfortunately for the PC industry, the very dynamics that drive people to buy better/more expensive computers also likely contribute to the lengthening upgrade cycles.

 

[snuffles]

General question: I'm not a hardware engineer, but wouldn't Zen's granular clock increments coupled with its fine meshed sensor suite allow the chip itself to be pretty much variable speed? In other words, the 3.4 GHz is just a "guaranteed baseline" for stock cooling so the general public isn't freaked out by the fact that it's completely variable. In other words, if a PC has a really good cooling solution, then its "base clock" can increase to, say, 4GHz, as that figure is primarily (beyond transistor switching speeds and whatnot) determined by heat density and the rate at which heat can be dissipated.

So instead of finding worst case scenario and manually keeping base/boost at a certain level ("traditional" CPUs with burn-in and whatnot to determine max safe frequency), the base frequency of Zen is completely variable depending on operating conditions (with some lower bound).

 

[bjt2]

General question: I'm not a hardware engineer, but wouldn't Zen's granular clock increments coupled with its fine meshed sensor suite allow the chip itself to be pretty much variable speed? In other words, the 3.4 GHz is just a "guaranteed baseline" for stock cooling so the general public isn't freaked out by the fact that it's completely variable. In other words, if a PC has a really good cooling solution, then its "base clock" can increase to, say, 4GHz, as that figure is primarily (beyond transistor switching speeds and whatnot) determined by heat density and the rate at which heat can be dissipated.

So instead of finding worst case scenario and manually keeping base/boost at a certain level ("traditional" CPUs with burn-in and whatnot to determine max safe frequency), the base frequency of Zen is completely variable depending on operating conditions (with some lower bound).

These are the disadvantages. If you want predictable performances and timings, you must use real-time OS and fixed frequency hardware. Even with turboless CPUs, modern OSes can't guarantee certain or tight timings: even on beefed-up machines a video playback can stutter sometimes...

 

[DrMrLordX]

General question: I'm not a hardware engineer, but wouldn't Zen's granular clock increments coupled with its fine meshed sensor suite allow the chip itself to be pretty much variable speed? In other words, the 3.4 GHz is just a "guaranteed baseline" for stock cooling so the general public isn't freaked out by the fact that it's completely variable. In other words, if a PC has a really good cooling solution, then its "base clock" can increase to, say, 4GHz, as that figure is primarily (beyond transistor switching speeds and whatnot) determined by heat density and the rate at which heat can be dissipated.

So instead of finding worst case scenario and manually keeping base/boost at a certain level ("traditional" CPUs with burn-in and whatnot to determine max safe frequency), the base frequency of Zen is completely variable depending on operating conditions (with some lower bound).

I agree with your assumption. I think that Summit Ridge will have flexible clockspeeds so long as XFR is active. Disable that and you can run at a fixed clockspeed if you like.

 

[itsmydamnation]

Somebody who buys an all-in-one PC from an OEM who will not sell you a system with a nice industrial design, nice screen, and other premium features without also selling you the nice processor.

A quick look at dell and HP shows about 1 in 4 desktop systems (atleast in AU) come by default with a dGPU AMD only needs a few models from each vendor and RR is going to be out before EOY 17 anyway.

It's in the OEMs' best interests to find ways to get consumers to buy more expensive computers. Basic PC biz stuff

Upsell is biz 101, that doesn't mean you have to start with an IGP

Seriously, the population of people who buy tower desktops is on the decline (I'd imagine it's mostly businesses and gamers at this point), and the subset of that population that builds their own systems and gives a hoot about the CPU inside? Much, much, much smaller.

you should inform Nvidia of that...

Also, a lot of people in the enthusiast community bemoan that a 5 year old SNB still works just fine. Most people don't *want* to have to spend money on new PCs (they are not cheap), so they figure that they should buy the nice system with the fastest processor that they can afford so that it lasts longer. The idea of a system lasting for 5 years+ and still being awesome and doing everything they want it to do is a pretty compelling feature.

Or maybe its the insainly poor value from Intel since about SB-E onwards especially when factoring in the need to change sockets. Give the enthusiast something of the equivalent kind of jump as you get on the GPU side and see how we go. the 3.4~3.6ghz lowest SKU 8 core @ around 400USD, 3.6~3.8ghz lowest SKU 6 core @ around 300 USD with 10G-BASE-T onboard should make a pretty good start.

im sure your hoping for a terrible OC from Zeppelin, im sure you'll ignore the German Overclockers claiming it beats BWE (but not enough to justify replacing an X99 system) and search for other rumors, you know you could team up with Juanrga ........

AMD would still make lots of revenue and GP from the above prices with Zepplin being only around 200mm sq. Remember a 315mm 32nm SOI BD was from 115- 245 USD at launch. $200USD for the lowest SKU 4core/8T to 450-500 for the highest 8core SKU AMD will do just fine.......

Unfortunately for the PC industry, the very dynamics that drive people to buy better/more expensive computers also likely contribute to the lengthening upgrade cycles.

AMD is not the one that needs to fill 9 fabs they can in effect ignore market dynamics because their install base is so low at this point.

 

[ksec]

AMD decide to attack the higher end simply because that is the only way to enter the market.

While it is useless without an IGP for the mass market, those market segment aren't making lots of money. And the only thing AMD needs now is profits for survival.

The margins is small in the low end, you need lots of volume to make up for it. And The agreement with GF means AMD has to be wise with their resources. Volume isn't something AMD should compete on, and will very likely to lose if they enter.

Unless AMD make a name out of Zen/Ryzen, I doubt any notebook with that APU is going to sell well irrespective of price.

Not to mention AMD is competing Intel at an inferior node. Intel has mastered the 14nm low power node. And it is working extremely well for 4W to 25W segment. I doubt AMD Zen APU would do equally well at those W range. Scaling down is hard. ( Although the much better GPU will likely make up for it, but consumer will have to trust AMD made a decent APU first )

Competing in the higher end means they could undercut Intel's heavy margin while still earning a profit. For Intel they are unlikely to react much afterall the desktop high end market is a niche.

If I were AMD CEO I will do whatever it takes to get an AMD Ryzen CPU inside the next update of iMac. That is literally the best marketing play guarantee to generate noise.

It will be interesting on the Server side, where Intel is charging their top bin E7 CPU at $7K / CPU. I am pretty sure AMD could do it for 4-5K.

 

[bjt2]

Not to mention AMD is competing Intel at an inferior node. Intel has mastered the 14nm low power node. And it is working extremely well for 4W to 25W segment. I doubt AMD Zen APU would do equally well at those W range. Scaling down is hard. ( Although the much better GPU will likely make up for it, but consumer will have to trust AMD made a decent APU first )

They manage to compete in the 15W range with old architecture, 28nm APU (see image in link in my signature), why they can't with new architecture, 14nm APU?

 

[beginner99]

General question: I'm not a hardware engineer, but wouldn't Zen's granular clock increments coupled with its fine meshed sensor suite allow the chip itself to be pretty much variable speed? In other words, the 3.4 GHz is just a "guaranteed baseline" for stock cooling so the general public isn't freaked out by the fact that it's completely variable. In other words, if a PC has a really good cooling solution, then its "base clock" can increase to, say, 4GHz, as that figure is primarily (beyond transistor switching speeds and whatnot) determined by heat density and the rate at which heat can be dissipated.

So instead of finding worst case scenario and manually keeping base/boost at a certain level ("traditional" CPUs with burn-in and whatnot to determine max safe frequency), the base frequency of Zen is completely variable depending on operating conditions (with some lower bound).

Heat isn't the only limit. As we have seen with last couple generations of intel CPUs delidding the CPU greatly helps with temps but you usually don't get much out of it, maybe 200 mhz higher OC at best. The limit it defined by the uArch and process. This means you can't ramp up clock speed as much as you like without also increasing voltage. I don't see how the chips can know it's own stability limits hence there must be some pretty large amount of margin built in. Meaning manual OC will be better for sure and the actual benefit of the tech rather small probably like 200 mhz at most.

 

[Abwx]

AMD decide to attack the higher end simply because that is the only way to enter the market.

.

Because there no possibility of contra revenue in this segment..?.

I doubt AMD Zen APU would do equally well at those W range. Scaling down is hard. ( Although the much better GPU will likely make up for it, but consumer will have to trust AMD made a decent APU first )

At low power any advantage Intel could have will vanish, it s at the upper frequencies that a node could have a superiority, at low and mid frequency GF s process is even possibly better than Intel s.

 

[KTE]

The history of civilization is full of unsound decisions. Aside from buying something future proof (perf/platform features) it could be a mix of succesful salespeople, ego driven desires (e.g. bragging), single/two threaded browsers (Firefox w/ plugin container process not long ago), etc. No need to wrap a technical fact based discussion around it.

8C\16T would be suitable for us Docker users tho

And BTW... Although server domain, it is HUGE in the enterprise world right now.

4C will become 'de rigeur' and 2C relegated, very soon. I'd expect that mobile power constraints is what is delaying this change, for now.

AMD has already started this movement with Excavator.

Sent from HTC 10
(Opinions are own)

 

[KTE]

Yes, on semi...
Moreover more info from twitter:

Discussion between Dresdenboy, Instlatx64 and Canard PC:

https://twitter.com/Dresdenboy/status/812053513031352326

1) AIDA test too good to be true, so they didn't publish yet. Double throughput of Skylake on some intructions, they think it's a bug in AIDA. Instlatx64 answers that skylake/kabylake have some trouble on port 5 and thus even Bristol Ridge have higher throughput for some instructions (e.g.: VEXTRACTI128)

2) Canard PC sample has problems with SMT and uop cache and then, coupled with semiaccurate infos i would not draw conclusions on Canard PC tests being worse than new horizon event...

Other tweet:

https://twitter.com/InstLatX64/status/813539775944785921

Finally decent cache latencies...

Those would be very good, agreed.

And yes, the CanardPC sample is early and still having bugs.

Sent from HTC 10
(Opinions are own)

 

[krumme]

I am sure amd will be willing to bundle zen with the upcomming small p12 for those oem that start talking about gpu at i5 or i7 levels there is so high margin ofloading some gf capacity and gain a discrete dgpu as sales parameter seems like a win win for all imo.
Yes sire. You can have this 8c zen with discrete gpu with 4GB ram. (No need to mention its ddr3 on 64 bit bus lol)

 

[inf64]

FX cpus already do that for i3 price . I don't see anyone advertising them, though

FX CPUs can only compete with i3 and i5 in MT benchmarks, it is unbalanced mainstream chip. Zen on the other hand is basically AMD's Haswell on 14nm node, competing with intels finest cores across all almost workloads ( AVX2 heavy workloads might be the only case where it will fall behind but this has to be verified like everything else).

Depends on actual performance. I know enough of loads where 8T are as good as slightly faster 4T and enough where they are even worse.

It doesn't matter. If 8T Zen OCs well it will be better than i5 no matter the workload. SMT will just give it huge edge where more threads will be usable (mutithreading, multitasking) while it will perform no worse in fewer threaded workloads. It is better than i5 the same way the i7 is better than i5 , there are no downsides to 8T i7 Vs 4T i5.

 

[Abwx]

FX CPUs can only compete with i3 and i5 in MT benchmarks,.

i3 are well below FXs in MT, i5s can sometimes compete with the help of some instructions.

http://www.hardware.fr/articles/940-19/indices-performance-cpu.html

 

[krumme]

Yes, on semi...
Moreover more info from twitter:

Discussion between Dresdenboy, Instlatx64 and Canard PC:

https://twitter.com/Dresdenboy/status/812053513031352326

1) AIDA test too good to be true, so they didn't publish yet. Double throughput of Skylake on some intructions, they think it's a bug in AIDA. Instlatx64 answers that skylake/kabylake have some trouble on port 5 and thus even Bristol Ridge have higher throughput for some instructions (e.g.: VEXTRACTI128)

2) Canard PC sample has problems with SMT and uop cache and then, coupled with semiaccurate infos i would not draw conclusions on Canard PC tests being worse than new horizon event...

Other tweet:

https://twitter.com/InstLatX64/status/813539775944785921

Finally decent cache latencies...

Wow !!!! L2 latency excactly at skl level with twice the size. Guys remember that. It is at twice the size l2. Damn ! Truly a historical moment.

Amd is actually beating Intel here hands down. Caling that "decent" is the understatement of the decade.

 

[itsmydamnation]

Not really if it twice the number of ways then its really just equal, AMD always had fast arrays and the L2's in Deneb, thurban etc were fine, its just bulldozer.................

 

[krumme]

Not really if it twice the number of ways then its really just equal, AMD always had fast arrays and the L2's in Deneb, thurban etc were fine, its just bulldozer.................

Naa ok thanx.

 

[rvborgh]

They finally improved over K10 on the L2

https://twitter.com/InstLatX64/status/813539775944785921

for comparison:

 

[lolfail9001]

It doesn't matter. If 8T Zen OCs well it will be better than i5 no matter the workload. SMT will just give it huge edge where more threads will be usable (mutithreading, multitasking) while it will perform no worse in fewer threaded workloads. It is better than i5 the same way the i7 is better than i5 , there are no downsides to 8T i7 Vs 4T i5.

First it needs to OC well, you know. Also, i7 with disabled HT is faster than i5 per clock, just saying.

They finally improved over K10 on the L2

I am a little conflicted. So, you guys are telling me Sandy Bridge has far better L3 cache latency than any of Intel's uArches afterwards?

 

[bjt2]

Wow !!!! L2 latency excactly at skl level with twice the size. Guys remember that. It is at twice the size l2. Damn ! Truly a historical moment.

Amd is actually beating Intel here hands down. Caling that "decent" is the understatement of the decade.

INTEL plans to shift to greater L2 size in SKL-X (1MB?), so maybe they increased latency to accomdate a bigger cache without major redesign? Haswell/Broadwell cache is faster. But also can clock lower...

EDIT: decent is an euphemism...

 

[witeken]

Maybe you should reconnect with witeken from the past

Waht do you mean?

 

[bjt2]

First it needs to OC well, you know. Also, i7 with disabled HT is faster than i5 per clock, just saying.

I am a little conflicted. So, you guys are telling me Sandy Bridge has far better L3 cache latency than any of Intel's uArches afterwards?

I don't remember, but last intel archs have L3 clock separate from core clock. Maybe sandy bridge has l3 at core clock? or maybe sandy bridge does not have ring bus?