I presume that TDP is only for the base clock and even here I would be a bit suspicious - 8700K with six cores at 3.7 GHz have the same 95W.
https://ark.intel.com/products/126684/Intel-Core-i7-8700K-Processor-12M-Cache-up-to-4_70-GHz
Guys how its possible to push 9900k 4,6Ghz all cores with TDP of 95W ?
I presume that TDP is only for the base clock and even here I would be a bit suspicious - 8700K with six cores at 3.7 GHz have the same 95W.
https://ark.intel.com/products/126684/Intel-Core-i7-8700K-Processor-12M-Cache-up-to-4_70-GHz
Good points, but I don't think cinebench represents the best case scenario for HT/SMT scaling.
For example, in Blender a 8700K is 47% faster than a 8600K. Even a 7700K beats the 8600K and that is with a 50% core deficit. Clock for clock the 8600K edges a 7700K but only barely. A 8700K only has 33% less cores than a 9700K.
https://www.gamersnexus.net/hwreviews/3157-intel-i5-8600k-review-overclocking-vs-8700k-8400/page-3
In Blender at least, based on this a 8700K would beat a 9700K if it is indeed an 8/8 chip. Other apps and benchmarks will be a mixed bag probably, like your CB example
I cursed at Intel for not having an 8/8 model the other day. Glad I was wrong.
About 25% boost in both cases.
Right now coffee lake has about 20% clockspeed advantage and some ipc advantage. So if the eight core can hold close to that clockspeed, CL would have about a 25% advantage over Ryzen. So Zen 2 will have to have a combination of 25% clock speed plus ipc gain to even equal CL. They may achieve that, but maybe not, and gaining more than that even with a new architecture and process seems like a tall order to me. So I dont think Zen 2 will necessarily be the CL killer some are assuming it to be. Not saying it wont be, but I dont think it is a forgone conclusion like many are assuming it will beValid point ... Call me naive, but I think they have had so long to work on this (4-5 years+ with all the delays) that they could well have gone back to the drawing board and bumped up the uarchitecture, maybe increased the caches or something. Especially once they realised probably by 2016 that 10nm was a shot process and even 10nm+ would not achieve the same clocks as 14nm++ (nevermind 14+++).
I don't know whether the spectre fixes and Ryzen's unexpected competitiveness could have influenced this or whether these events were too late in the game?.
Either way by the time Kabylake was launched as a stopgap they must have had a tough decision to make, that's where 14nm++ & 14nm+++ 6 & 8 core skylake dice came in to hold the fort whilst they Sellotaped 10nm+ together with a rejig of icelake microarchitecture...
Perhaps they had the foresight (alongside some handy information from poached ex AMD employee's) to scrub the IGPU and put on some more cores to the MESH..like 12 for instance.
Just speculation on my part, but I can't fathom how intel is going to compete with zen 2 with an 8/16 icelake, that has only 5-10% IPC and lower clocks than coffeelake can you?
The "leak" could very well be fake, but the TDP is defined for the base frequency, which is also suspect. The turbo frequency is not constrained by TDP at all, but only by the VR specification (ICCmax) and circuit timing. So we'll see "95 W" chips that run at 300-500 W out of box, similar to Skylake-X.The issue is that on Intel the TDP is an actual power limit (PL1).
So either the claimed clocks or the TDP are inaccurate, or the process variant is significantly more efficient than its predecessor.
Most likely either of the claimed specs is inaccurate IMO.
Your right, it is not a foregone conclusion that Ryzen 3 can best the worst case version of icelake I specified, let alone the one I think they will launch.Right now coffee lake has about 20% clockspeed advantage and some ipc advantage. So if the eight core can hold close to that clockspeed, CL would have about a 25% advantage over Ryzen. So Zen 2 will have to have a combination of 25% clock speed plus ipc gain to even equal CL. They may achieve that, but maybe not, and gaining more than that even with a new architecture and process seems like a tall order to me. So I dont think Zen 2 will necessarily be the CL killer some are assuming it to be. Not saying it wont be, but I dont think it is a forgone conclusion like many are assuming it will be
The "leak" could very well be fake, but the TDP is defined for the base frequency, which is also suspect. The turbo frequency is not constrained by TDP at all, but only by the VR specification (ICCmax) and circuit timing. So we'll see "95 W" chips that run at 300-500 W out of box, similar to Skylake-X.
TDP is always PL1 (sustained power limit) on Intel.
Base frequency doesn't exist unless Turbo is disabled.
https://newsroom.intel.com/wp-conte...04/8th-gen-intel-core-h-series-press-deck.pdf
Page 34.
Of course the base frequency exists. TDP is the worst-case non-AVX power at the base frequency, which directly implies that the base frequency is the minimum frequency the processor runs at when limited to TDP by PL1. If you run a power virus, I'm sure you'll see the frequency approach base, especially on those parts where Intel makes an actual effort* into validating the base frequency (mobile and server).TDP is always PL1 (sustained power limit) on Intel.
Base frequency doesn't exist unless Turbo is disabled.
https://newsroom.intel.com/wp-conte...04/8th-gen-intel-core-h-series-press-deck.pdf
Page 34.
This is false. Show us a coffee lake chip drawing 200+ watts at stock, please.Of course the base frequency exists. TDP is the worst-case non-AVX power at the base frequency, which directly implies that the base frequency is the minimum frequency the processor runs at when limited to TDP by PL1. If you run a power virus, I'm sure you'll see the frequency approach base, especially on those parts where Intel makes an actual effort* into validating the base frequency (mobile and server).
* For desktop, Intel clearly doesn't give a fuck about base frequency, as can be seen from Skylake-X and Coffee Lake. They pick some lowball figure to make the TDP "work" and ship them to customers drawing 200+ W.
This is false. Show us a coffee lake chip drawing 200+ watts at stock, please.
Of course the base frequency exists. TDP is the worst-case non-AVX power at the base frequency, which directly implies that the base frequency is the minimum frequency the processor runs at when limited to TDP by PL1. If you run a power virus, I'm sure you'll see the frequency approach base, especially on those parts where Intel makes an actual effort* into validating the base frequency (mobile and server).
* For desktop, Intel clearly doesn't give a fuck about base frequency, as can be seen from Skylake-X and Coffee Lake. They pick some lowball figure to make the TDP "work" and ship them to customers drawing 200+ W.
Typically the lowest an 8700K will go at stock is 3.95GHz during Prime95 FMA3.
If the power limits are configured as they should (95W PL1, 119W PL2), it is impossible for the CPU to consume more than that in any workload.
PL2 is oddly close to the power consumption that a "good sample" of 8700K draws during a similar torture test with MCE settings (47x cores, 44x cache)
PL2 = configurable Power Limit which mobo makers / system integrator can set according to their needsPL2 (which seems to be equivalent to MCE activated)
MCE draws almost exactly 120.6W/1.210V=100Amps average during torture testing, I've seen this aswell on a "bad sample" of 8700K and it drawed high 120s to low 130W while needing 1.280Vcore minimum to be stable. Mind you both CPUs had the toothpaste replaced with liquid metal, so it should be comparable to a soldered one.
Settings were not stock to start with, since MCE increases 6c multi from 43x to 47x.The only nitpick i have here is that Your "settings" are not stock for memory, VCCSA and VCCIO. Memory that fast and 20-30% higher voltages are bound to generate extra load compared to stock 2666 dram.
Settings were not stock to start with, since MCE increases 6c multi from 43x to 47x.
As I noted, Intel makes no effort to provide tight bounds on the power consumption of "desktop" (overclocked laptop) parts. If you try this on mobile or server, you will see it approach the base frequency, with maybe 2-3 bins of margin on most parts. Actually, if you look at the reviews for the controversial "i9" MacBook, it runs at 3.1 GHz (base 2.9) when editing videos at 45 W.Typically the lowest an 8700K will go at stock is 3.95GHz during Prime95 FMA3.
If the power limits are configured as they should (95W PL1, 119W PL2), it is impossible for the CPU to consume more than that in any workload.
Improper configuration of the power limits is not something which Intel does, but something that the motherboard ODMs do (and which poor reviewers do not counter).
The International Rectifier series of VRMs that everyone and their mother uses has 94% efficiency at optimum current, and 92+% across a broad spectrum. Some newer Infineon chips that are starting to trickle down to consumer have 96% efficiency. Inefficient VRMs are a myth.Also here is Hardware.fr's measurements on 8700K during Prime95 28.10.
The VRM efficiency is typically < 83% on modern boards.
113.8 * 0.83 = < 94.454W.
The International Rectifier series of VRMs that everyone and their mother uses has 94% efficiency at optimum current, and 92+% across a broad spectrum. Some newer Infineon chips that are starting to trickle down to consumer have 96% efficiency. Inefficient VRMs are a myth.