In ST the power budget is in the 30-50% of the MT power budget, chips like the 8370E are close to the upper
percentage because they have low base frequency and quite high turbo, besides it consume only 65W in MT, so it cant be used as comparison with Zen, that being said this doesnt invalidate your point about Zen ST frequency, if it does 3.4 in MT there s no way that it would be limited to 3.7 in ST.
Lets have a little bit of a theoretical discussion:...that being said this doesnt invalidate your point about Zen ST frequency, if it does 3.4 in MT there s no way that it would be limited to 3.7 in ST.
Lets have a little bit of a theoretical discussion:
Voltage and clock scaling are not always linear, and eventually, more voltage does not help clock scaling, only generates more heat and power consumption. Every x86 architecture to date seems to hit a frequency wall not helped by higher voltage, even under LN2 cooling. I had an 8 GHz FX-8120 that would not go even 5 MHz more with an extra 0.1v on LN2.
Scenario 1: Zen does not overclock well but the 3.4 GHz base clock is achievable with low voltage.
Maybe Zen needs 1.1v to achieve 3.4 GHz base clock at 95w TDP, but needs a seemingly exponential increase in voltage to reach higher frequencies.
AMD finds that a minimum of 1.475v is necessary to reach 4 GHz, which is (theoretically, in this situation) deemed outside of safe operating voltage. AMD then cannot allow a 4 GHz turbo, even for one core, part of the time.
Scenario 2: Zen is an overclocking monster with very good voltage / clock ramping until 4.4 GHz.
AMD finds that a minimum 1.35v is necessary to reach 4.4 GHz, which is within their safe operating voltage for turbo states. Zen then receives very high turbo frequency.
We have no idea what the maximum clock speeds are that the architecture is capable of. Intel ran into this problem with Kaby Lake, the 7700K only has a 300 MHz turbo because going higher is unfeasible. To be honest, you see this with Piledriver as well. Why does the quad-core FX-4350 only have a 4.2 GHz base clock, +100 MHz turbo and 125w TDP?
We will see what happens when more details are released in coming months.
Just comparing clocks doesn't say much about the design or the process.Ok, but we are talking of three nodes away... Sandy bridge topped at 3.9GHz turbo, if i remember well, on 32nm and on 14nm kabylake tops at 4.5GHz. 8370E tops at 4.3Ghz on 32nm and in three nodes AMD manages to be worse than a process of many years ago? Higher IPC does not automatically mean higher FO4.
The Stilt said:The 32nm SHP SOI ... started as a complete lemon with some huge issues with the leakage and the general inconsistency. In it's current form the 32nm SHP SOI is a completely different animal. ... The silicon consistency is immaculate and the average static leakage on full Vishera dies is up to 45% lower than it was at the time of launch. The only trouble is that it took around three years for it to reach its peak.
Looking at the variation of the most recent 32nm SHP silicon version manufactured in July 2014 (80 samples), the absolute maximum variation in SIDD is 33.1%. When the couple rare extremities of both ends are removed, the average variation is just 10.84%.
Also, there's the power draw of the integrated components:The Stilt said:Zen is by no means a speed demon design. On the contrary actually. It is closer to Hounds than Bulldozer in terms of latencies.
Design wise I would recon the L2 caches on Zen are the most limiting factors for Fmax. If AMD would modify Piledriver to have similar L2 characteristics as Zen does, the resulting part would have Fmax of ~2.8GHz, instead of the usual ~4.7GHz
And, to top it off, the AM4 spec is more demanding for motherboard power delivery/component quality, making keeping down the TDP more important than with AM3+:The Stilt said:95W TDP Zeppelin parts should draw around ~ 80W from the primary (VDDC, CPU core) power plane. The rest (~15W) is for the northbridge, IO, FCH, etc and will be drawn from the secondary VDD_SoC plane and other minor, low-current planes.
The Stilt said:When AMD moved from 32nm SHP SOI to 28nm BULK the voltage stability became extremely important. Despite the platforms using parts made with different processes (e.g AM3+ and FM2+) had exactly the same load-line specification (1.3mOhms) in reality the smaller and othewise inferior 28nm process was significantly more sensitive to voltage variations / fluctuations. Achieving a stable voltage supply through proper (load dependent) load-line calibration can result in hundreds of MHz additional headroom when close to Fmax, even on the more recent 28nm (Godavari) chips.
For Zen the load-line appears to be (based on the existing VRM designs) significantly tighter than it was with previous AMD designs and much tighter than the Intel VR12 spec (which is already strict) specifies...
The Stilt said:AM4 requires much beefier VRMs (all CPU / APU planes), higher quality PCB (due higher currents, faster signaling), etc. The design guidelines for all Zen based platforms are quite demanding, so you cannot get away with the same garbage quality as one could with "AM1" or FM2+. Unless you want to break the platform cross compatibility of course. All FM2+ boards < 60$ (CSP) are generally garbage and in most cases cannot operate properly with all the APUs available for the platform.
The Stilt said:The load-line specification for both AM3+ and FM2+ is extremely loose (1.3 mOhm & 2.1mOhm) and because of that the voltage droop by the specification is very large. To ensure that the operation parameters remain within a spec, an AM3+ part which draws 100A of current and requires 1.3000V to be fully stable must have at least 1.4300V default voltage (130mV droop @ 100A). Some of the motherboards are built to have lower Rll (< 1.3mOhm / 2.1mOhm) than the specification dictates, or the end-user might adjust it to be lower than the default value. If the droop is lower or non-existing for either reason, it appears that the parts are extremely overvolted from the factory. The truth of course being that they are configured for perfectly right default voltage, which complies with the specifications (droop) and contains some standard safety margins.
On AM4 the situation is significantly better, since the load-line spec. is less than half of what it was on AM3+.
The Stilt said:The Fmax vs. Vdd curve on 14nm LPP is vastly steeper than on 28nm HPP, close and beyond the optimal Fmax range
Not necessarily, for the reasons I listed. However, in current desktop tech it's usually the case.At the end of the day it is the energy you consume at the wall that you are going to pay and not only the CPU
So one system used less power to finish the same work at less time and this system has the higher perf/watt.
Regarding the voltage droop thing: Since Steamroller AMD has some tech to live with it. With BR they demonstrated the capability to reduce required voltage margins. I doubt that they left out anything of that in Zen. This means, since BR (AM4) the chips might require lower voltages than one was used to with AM3+/FM2+ and lower.Just comparing clocks doesn't say much about the design or the process.
Also, there's the power draw of the integrated components:
And, to top it off, the AM4 spec is more demanding for motherboard power delivery/component quality, making keeping down the TDP more important than with AM3+:
8370E was on 32nm. Three nodes away from Zen. And it seems that zen will have FO4 similar to BD. Anyway lower than core, because zen have 19 stage integer pipeline, versus 14-17 (not disclosed, these are the estimations) of core architecture. And Excavator with 28nm bulk HDL managed to have 4.3GHz turbo too. With two node and similar FO4 they did not managed to increase max clock turbo? With GPUs that gained 20% max clock? Surely I don't expect 20% gains at so high clocks, but certainly i don't expect a regression from 4.3GHz...
Just comparing clocks doesn't say much about the design or the process.
Also, there's the power draw of the integrated components:
And, to top it off, the AM4 spec is more demanding for motherboard power delivery/component quality, making keeping down the TDP more important than with AM3+:
Sent from HTC 10
(Opinions are own)
Which means absolutely nothing for Zen launch.This is your opinion. INTEL managed to go from 3.9GHz of SB (3820 32nm) to 4.5 of KBL (7700 14nm).
I posted a GF graph of quite linear scaling on LVT transistors up to 4-5GHz...
About the hilarious TDP discussion.
Zen idle: 93W
Intel idle: 106W
Zen load: 187W
Intel load: 191W
Zen delta: 94W
Intel delta: 85W
So tell me, who has the highest performance per watt?
Which means absolutely nothing for Zen launch.
Intels uarch was +40% ahead of AMD in IPC ALREADY...
You really don't get it.
More work done/IPC takes Power!
IPC/Doubling resources/SMT is not free!
Jeeeeez you are on auto-repeat llike a parrot.
Try a power analyzer... Run lower IPC code vs higher IPC code.
See the power difference.
Sent from HTC 10
(Opinions are own)
There are some slight rumors rumbling that should be taken with a grain of salt that prices will be as follows:
$499 Top binned fully unlocked 8c/16t Zen
$349 8c/16t Zen
$249 6c/12t Zen
$149 4c/8t Zen
As I've said before, if AMD can release any 8c/16t part with roughly the IPC shown during their demo at $400 or less then that is a GIANT win for consumers everywhere. Seems like there are tons of us out there with 2500k/3570ks that have been able to make due for the past 5 years but have held off wanting both better IPC and more cores. I've been dabbling in some fun video editing projects and I definitely want more cores than I currently have.
All this back and forth about power numbers and we are talking about less than 1-2% from the numbers shown... Looks like they more or less consume the same which is a huge step for AMD.
There are some slight rumors rumbling that should be taken with a grain of salt that prices will be as follows:
$499 Top binned fully unlocked 8c/16t Zen
$349 8c/16t Zen
$249 6c/12t Zen
$149 4c/8t Zen
As I've said before, if AMD can release any 8c/16t part with roughly the IPC shown during their demo at $400 or less then that is a GIANT win for consumers everywhere. Seems like there are tons of us out there with 2500k/3570ks that have been able to make due for the past 5 years but have held off wanting both better IPC and more cores. I've been dabbling in some fun video editing projects and I definitely want more cores than I currently have.
Same energy per core/cycle means nothing on its own. You've already been told this many times, or to prove the other unknown factors and variables around it. You have no evidence for them, but you even dodge logical reasoning for them and carry on repeating the same thing.There are three nodes of difference. More than 50% less power/transistors.
Indeed AMD declared same energy/cycle per core than excavator.
That means that two zen cores draw same power an excavator module (2 cores) at same frequency.
A12 9800 is a 65W apu with 4.0/4.2GHz clock and 2 XV modules (plus a GPU).
Usually roughly 50% of power budget is allocated to the GPU, so 4 zen core draw about at most 40W at 4.0GHz or 8 core 80W at 4Ghz. So we are in budget with 8c at 4GHz.
If you don't trust even official AMD statements, but keep your opinions, it's useless continue to discuss... We will see when final silicon is out...
Now it's you that are talking like a parrot, negating at libitum AMD official statement. I was only explaining you why your statements are in contrast with AMD statements.
But ok. You are right and AMD is lying, coz IPC, power and who knows what else and AMD statements are shit...
All this back and forth about power numbers and we are talking about less than 1-2% from the numbers shown... Looks like they more or less consume the same which is a huge step for AMD.
There are some slight rumors rumbling that should be taken with a grain of salt that prices will be as follows:
$499 Top binned fully unlocked 8c/16t Zen
$349 8c/16t Zen
$249 6c/12t Zen
$149 4c/8t Zen
As I've said before, if AMD can release any 8c/16t part with roughly the IPC shown during their demo at $400 or less then that is a GIANT win for consumers everywhere. Seems like there are tons of us out there with 2500k/3570ks that have been able to make due for the past 5 years but have held off wanting both better IPC and more cores. I've been dabbling in some fun video editing projects and I definitely want more cores than I currently have.
Same energy per core/cycle means nothing on its own. You've already been told this many times, or to prove the other unknown factors and variables around it. You have no evidence for them, but you even dodge logical reasoning for them and carry on repeating the same thing.
Mate, these PR slides are not science and absolutes cannot be derived from them, like you keep doing. Crusading wildly based on them dents your own credibility.
I don't doubt AMDs tenacious ability for continuous improvement though, and they have done very well if the ST performance is good. Personally, I'm now expecting >40% IPC increase, which is revolutionary to say the least.
But I doubt your exaggerations have any base in reality, as is gradually being shown.
Yes I do think AMD might have better efficiency with their SMT implementation. They've had higher mispredicts/stalls/cache latencies and CPU resources free than Intel for a long time now.
Also, as for Power... Turbo boost has the capability to exceed max power bands on Intel and AMD for short boosts, while still keeping under the TjMax/Tctl. That's exactly its beauty. That if critical temps allow, to boost the Core frequencies as much as possible, until the temps start becoming too high. Its a pretty complex loop with multiple factors.
Multiple AMD and Intel chips have been far lower power than the TDP rating, so this kind of feedback related boost is the right evolutionary step. More performance, within the provided cooling capacity. Why not?
Sent from HTC 10
(Opinions are own)
All this back and forth about power numbers and we are talking about less than 1-2% from the numbers shown... Looks like they more or less consume the same which is a huge step for AMD.
There are some slight rumors rumbling that should be taken with a grain of salt that prices will be as follows:
$499 Top binned fully unlocked 8c/16t Zen
$349 8c/16t Zen
$249 6c/12t Zen
$149 4c/8t Zen
As I've said before, if AMD can release any 8c/16t part with roughly the IPC shown during their demo at $400 or less then that is a GIANT win for consumers everywhere. Seems like there are tons of us out there with 2500k/3570ks that have been able to make due for the past 5 years but have held off wanting both better IPC and more cores. I've been dabbling in some fun video editing projects and I definitely want more cores than I currently have.
Same energy per core/cycle means nothing on its own.