AMD Ryzen Chips 10% Smaller When Compared to Intel Skylake Dies
Click here to post a comment for AMD Ryzen Chips 10% Smaller When Compared to Intel Skylake Dies on our message forum
Denial
It's also on a larger process(slightly).. interesting.
Alessio1989
igpu...
schmidtbag
Yes, a little discouraging. This could be another situation where some tests perform great while others are very lacking, which is reminiscent of Bulldozer. But, at least Ryzen has already had proof of outperforming Intel in real-world everyday applications. Meanwhile Bulldozer was really only theoretically faster than Nehalem or Sandy Bridge or whatever was available at it's release time.
There are many things to consider, though. For example, transistor count isn't everything; realistically, the only way for AMD to outperform Intel in just about any test while still retaining full instruction set compatibility is to make the design as efficient as possible. In other words, involve the fewest amount of transistors as possible without crippling the pipeline.
Loophole35
Likely this.
Denial
There is no way an Intel 4 core CPU with an iGPU is only 49mm2. They definitely subtracted the iGPU out for the comparison.
Edit: Anandtech estimated that 6700K is 122.4 mm2 - so yeah, definitely not factoring the iGPU here.
poornaprakash
Nope
Nothing to do with iGPU :wanker: It simply means Ryzen on its GloFo 14nm process is slightly more denser than Intel's 14nm process despite Intel 14nm is slightly superior to GloFo 14nm process.
PrMinisterGR
That sounds wrong. There is no way that GloFo or anyone else has a better 14nm than Intel.
PrMinisterGR
No way it's 49mm2 with an iGPU on it. These are numbers per CPU core.
fantaskarsef
I wonder if this has a negative effect on heat dissipation, I wonder what TPD's we're talking about in those two models.
Also, if Intel's chip uses less size for a core than AMD's does that somewhat work in conjunction with AMD's lower IPC, because smaller chip with bigger cores = less cores = less IPC?
Denial
No it's not - Intel's process is more dense, it shows the specifications of it in the graph in the article. AMD's architecture is clearly more "compact" requiring less transistors for (hopefully) the same performance. This probably also factors into the power being lower.
I think the bigger effect on heat dissipation is lack of iGPU - but big Zen is also 8/16 so those chips will probably be similar in size to Intel's 4/8 + iGPU (since Zen doesn't have an iGPU). This is only a 4mm2 difference and less transistors should mean lower power anyway. Either way I don't think it's going to be an issue - as long as AMD doesn't glue their headspreader on with a 30 meter gap like Intel does.
As far as the second paragraph - Intel uses more size for a core, it's unknown whether that core is faster or not though. Also both companies seem to have no problem fitting up to 16+ cores and whatnot, so I don't think the difference is going to factor much into the total core count.
Warrax
It's CPU cores only, not even I/O and stuff.
Alessio1989
'cause having a GPU on the same die of the CPU it's not like adding more Lego blocks.
Memory controller and cache BUS, north-bride... all thisinfluence ALL the CPU parts disposition on the die. Skylake CPUs have to deal with all this crap. It's NOT Lego.
But most important: who cares? Execution time (ie the only objective important performance factor) and power consumption are the only things that matter.
poornaprakash
Intel 14nm process is slightly denser which I agreed. But if we look at the 8 MB L3 cache die area for example Ryzen is more dense with an area or 16mm2 in comparison to Intel 19.1mm2 die area. Since we do not know the transistor count of Ryzen we cannot confirm anything.
wavetrex
Jim Keller confirmed alien from outer space.
Too bad he has to work with this inferior human technology... 14nm planar... ugh oh.
I'm guessing his spaceship uses three-dimensional 1nm carbon structures and is million times more power efficient than everything we have right now.
But he'll get there ... eventually, and build the parts to repair his ship so he can go back to his home planet 😉
poornaprakash
Ryzen uses no 14nm planar but a 14nm FinFET process 😛uke2:
DARKSF
Transistor count is not a measure for IPC for example T-bred B had 37.6M transistors and had higher IPC than any P4 which varied from 42M up to the Prescott 125M transistors with HT with which it already had parity in IPC or was even a bit slower than it in IPC.
Anyway good news if they are able to pull the same IPC from less silicon.The corner stone for IPS has always being the Branch Predictor and i remember reading somewhere that the improved branch predictor is guilty for up-to 30% of the claimed 40% IPC increase in the Zen architecture.
Same is for Intel the Sandy Bridge generation jump in IPC is again resting for most of it's part on the branch prediction optimisation more than anything else.And this is why you won't see any big performance jump from Intel while they use this architecture cause it is already optimised as much as it can be now the only way up is CPU clock or a radically new architecture.
Any way. IN Jim Keller i trust for the last almost 2 decades 🙂 Hey anyone wondering why iPhones are with dual core CPUs while the rest are using 8 now even 10 probably is not only the highly optimised iOS only responsible but the begining the A5 CPU ? :P
Disputes
working out the values for these chips show
intel @ 99.3138459632mm^2 + gpu and misc = 122.4mm^2
and
amd @ 125.546804021mm^2 + misc?? = maybe 130mm^2
anandtech.com/show/9505/skylake-cpu-package-analysis
says the chip is 122.4mm^2
if thats correct..... intel are in for alot of trouble.
very impressive amd
Math
AMD
44^2 x 8 + 1.5^2 x 8 + 16^2 = 15762
root(15762) = 125.546804021mm^2
Intel
49^2 x 4 + 0.9^2 x 4 + 19.1^2 = 9972.05
root(9863.24) = 99.860152213mm^2 (122.4mm^2 with gpu) seems about right for gpu takes up 1/3 of the chip
(i did do this another way before but i think it was wrong. i got 120mm^2 and 144mm^2 for intel and amd which makes no sense as intel dont have a 2mm^2 gpu)
also would amds fab process be cheaper as it has less metal layers?
Yogi
As a wise man once said. It's all Lego.
Alessio1989
Today, transistors are cheap, wires are expansive. You do not place transistors to minimize the die any-more, you place them to minimize the wires. Save some space to obtain a couple of dies of a wafer slice is no more a big issue compared to the wire cost.
But again, who cares which is smaller/bigger?
xvince1
I think it's simply the "dead or alive" request for AMD. If their dies are smaller, they can be competitive again vs Intel's wafer
