They are sure as heck not at an IPC disadvantage....

how do u know ipc when we dont even know clocks yet on intels (bar the smallest)?

Also, cooling on air seems out for intel, which is no small thing for consumers/hedt.

Because we know AMD IPC is several percent behind Intel. We also know SkylakeX isnt going to go backwards in IPC...we also know threadripper is the same as any existing zen cpu just with more ccx. No arch changes/updates so no IPC changes.

You are aware of what IPC is, right?

Edit: lol, aircooling isnt out for intel...o_O. And since its hedt, you'd think people can afford an aio anyway...

notbI don't know what would be the profit for AMD in a $850 16-core CPU (it could be similar to Ryzen 7).
Let's assume it is $200. Why not sell for $1400 and get another $550? Intel's 16C/32T is expected at $1700.
Few weeks back AMD was commenting on the mediocre Q1 results and they said that high-profit products will get a higher priority. What happened to that?

Do you have any idea how profitable the 1700, 1700x, and 1800x are? AMD is making only ONE chip and this is it with 8 cores and it costs them only $60 to fab! Where do they get the extra pci lanes? Why they simply connect the wires I am sure are already available internally on the ONE chip they are making!

systemBuilderDo you have any idea how profitable the 1700, 1700x, and 1800x are?

No, exactly. I'm waiting for the financial statement.

systemBuilderAMD is making only ONE chip and this is it with 8 cores and it costs them only $60 to fab!

Where did you find this information?
What about unallocated costs? Are they included in this figure?

Also... $60 is a lot....

systemBuilderWhere do they get the extra pci lanes? Why they simply connect the wires I am sure are already available internally on the ONE chip they are making!

Are you saying that Ryzen 7 is crippled?

Estimated costs for Threadripper production are $120 and so i simply cut that price in half, assuming the packaging for threadripper is twice what it costs for Ryzen - we know that threadripper is just 2 Ryzen cpus interconnected by the infinity fabric!

$60 on a $300 part is 400% markup, i bet you wish you had a product with that kind of markup. The markup is even better on 1700x ($349) and 1800x ($449).

In VLSI fabrication you ALWAYS build the ultimate CPU and the stuff that fails test with JTAG is lasered out. Some of the Ryzen 5 cores probably have 8 working cores but if the customer pays only enough for 4 cores, they disable the other 4 cores. I think it's pretty clear (because there are so few SKUs and just about every chip overclocks at maximum GHz) that AMD took the time to optimize yields and process margin before releasing Ryzen. While $60 to make a $300 part is not as profitable as i486 (which was $25 on a $400 part - more profitable than printing money!), it is still a great profit margin. Ryzen chips are 22% smaller than equivalent Intel chips (part of this probably the lack of integrated graphics).

I don't know why AMD motherboards only have 24 PCI lanes it could be a limitation of AM4 - not the chips themselves.

systemBuilderEstimated costs for Threadripper production are $120 and so i simply cut that price in half, assuming the packaging for threadripper is twice what it costs for Ryzen - we know that threadripper is just 2 Ryzen cpus interconnected by the infinity fabric!

Again, where does this figure come from and what does it include?
Regarding "cost of manufacturing" - basically the only thing you can scale with number of cores is the cost of waffle and that's just a few % of the 60/120 USD you talk about.

systemBuilder$60 on a $300 part is 400% markup, i bet you wish you had a product with that kind of markup. The markup is even better on 1700x ($349) and 1800x ($449).

First of all: why do you even use markup? It's a measure used by people with 1-man businesses (e.g. house painting) that can't calculate costs effectively.
The values you suggest are nothing special, anyway.
I work in insurance. Every day I see products with infinite markups...

systemBuilderIn VLSI fabrication you ALWAYS build the ultimate CPU and the stuff that fails test with JTAG is lasered out.

WOW, that was obvious. But I see you have a big fondness for technical abbreviations.

systemBuilderWhile $60 to make a $300 part is not as profitable as i486 (which was $25 on a $400 part - more profitable than printing money!), it is still a great profit margin.

Markup is not profit margin.

systemBuilderI don't know why AMD motherboards only have 24 PCI lanes it could be a limitation of AM4 - not the chips themselves.

So Ryzen 7 is crippled not internally but by socket design? This is your theory? :-)

notbI work in insurance.

I can tell.

systemBuilderI can tell.

?

Honestly, you'll give a source for that $120 or not? Maybe you missed a zero and it's actually $1200? :-)

notbHonestly, you'll give a source for that $120 or not? Maybe you missed a zero and it's actually $1200? :)

You should probably stick with insurance.

AquinusYou should probably stick with insurance.

What's your problem? You have something against me? Against insurance business in general?
Someone posted information about rumored Threadripper manufacturing cost. I'd simply like to see the source.

EarthDogCaptn... its a crapshoot for their 8 thread to reach 4 ghz on all threads (and blow its tdp out of the water in the process)... I mean, I WANT to see it, but just am not sure, with the info we have already, that's going to be possible with all threads. Zen2...?

The Mac Daddy boost to 3.9 ( 4ghz with xfr) boost is two cores not 1 if I remember correctly. ( The cores have pairs, can someone with Ryzen confirm this please?) 3.8-4ghz should be possible because of the mcm design assuming you can deliver enough power as we'll probably be at the 250w+ trying to power all 16 cores at 4ghz. Epic motherboard VRM will be required but there's nothing stopping the silicon it's self from hitting Ryzen speeds.

AquinusIsn't Ryzen 5 1400 a single CCX CPU?
Edit: I see that they just disabled two cores and half of the L3 on each CCX in 1400. At that point, why didn't they just do a full CCX? :kookoo:

Because they are making 8 core chips as a base not quad cores.

Having multiple different cpu designs costs a lot more money.

AMD have gone for the Jack of all trades type cpu design and it works pretty good.

systemBuilder

notbI work in insurance.

I can tell.

AquinusYou should probably stick with insurance.

Keep it impersonal please.
systemBuilder: If you can't provide a source that's fine, but don't resort to baiting.

pantherx12The Mac Daddy boost to 3.9 ( 4ghz with xfr) boost is two cores not 1 if I remember correctly. (The cores have pairs, can someone with Ryzen confirm this please?

2 core boost confirmed on R7 :)

notbSomeone posted information about rumored Threadripper manufacturing cost. I'd simply like to see the source.

I don't need a source to know that the following statement is absurd.

notbMaybe you missed a zero and it's actually $1200? :)

If manufacturing costs were that high, not only would there be no room for profit because your pricing would have to start at at least $1,500 and since a lot of dies are cut from the same wafer and as @systemBuilder said:

systemBuilderIn VLSI fabrication you ALWAYS build the ultimate CPU and the stuff that fails test with JTAG is lasered out. Some of the Ryzen 5 cores probably have 8 working cores but if the customer pays only enough for 4 cores, they disable the other 4 cores. I think it's pretty clear (because there are so few SKUs and just about every chip overclocks at maximum GHz) that AMD took the time to optimize yields and process margin before releasing Ryzen.

Plus, it's not like you're making profit immediately after building it if you factor in the cost of development, the cost of your employees, the cost of marketing, etc. along with the cost of manufacturing. There are a lot of expenses that need to be taken into account when talking about the production of ICs and the actual manufacturing cost is probably a lot less than you think it is but, you probably don't realize how much money is going to the other things that got the product there. Insurance and manufacturing are two very different industries that operate differently. I would even argue that building software is more like building ICs than like insurance because the big cost isn't the manufacturing of a product or deployment of software, it's in all of the other things that lead up to it and @systemBuilder was talking about the manufacturing costs not total costs.

infraredIf you can't provide a source that's fine, but don't resort to baiting.

AMD isn't going to release what their margins are on a CPU, that's unwise but, it's not unrealistic to call out a nutty claim like this:

notbMaybe you missed a zero and it's actually $1200? :)

...because if manufacturing costs were that high, entry level thread-ripper would probably be north of $1.5k and possibly even 2k to actually break even because manufacturing costs are one part of the equation, not the entire one. While I understand how systemBuilder's claim can't necessarily be confirmed due to information that no one other than someone in the industry would be aware of, I was calling out the actual erroneous claim that manufacturing costs would be as high as 1.2k because that definitely is nuts regardless if you have a source or not. Anyone who can do simple math can find out real quick that a number that high doesn't quite add up.

...but if being in the insurance industry gives @notb a unique perspective on computer hardware, just image what being a software engineer adds to my perspective. :kookoo:

AquinusI don't need a source to know that the following statement is absurd.
[cut]

So much text!
You could simply say that you don't get irony.

notbSo much text!
You could simply say that you don't get irony.

If that's the best you got, then I don't think it's me who's not getting the irony. :love:

infraredKeep it impersonal please.
systemBuilder: If you can't provide a source that's fine, but don't resort to baiting.


2 core boost confirmed on R7 :)

Thank you Kindly, I doubt they'll do it that way but it means thread ripper maybe 4 core boost. That would be fantastic since 4 cores are the sweet spot for an awful lot of stuff.

Ahh, so XFR is one core? I appreciate the clarity too..

SKylake X is also doing the same thing (full boost 3.0 to 2 cores instead of one).

EarthDogAhh, so XFR is one core? I appreciate the clarity too..

SKylake X is also doing the same thing (full boost 3.0 to 2 cores instead of one).

Xfr is just 100mhz on top of the boost assuming temperatures are fine and tdp is within specs. So it should still be 2 cores that are boosted. ( Basically it's marketing wank as I'm sure you picked up on it being the clever chap you are, they boost to 4.1 but AMD couldn't garantee 4.1 on every chip so came up with a word so they can advertised 4.1ghz without actually committing to it. )

I am definitely not 100% on this though I've been following all of this pretty closely but it's sometimes hard to remember what I've read in a forum or on an actual article.

a comment i saw, puts what i posted b4 much better:

"However at high core counts intel's monolithic die design concentrates all the thermal energy in one area which leads to an exponential build up, they have to clock the cores much lower to maintain stability, the spaced out design with cores being connected by the infinity fabric avoids this issue and allows AMD to clock Threadripper at Ryzen 1800X (or more apparently with the 12 core) and EPYC at much higher clocks than their Intel counterparts. With the sheer number of cores being used in these chips and with each chip being considerably faster due higher clocks, Xeon is going to be in for one hell of a pounding."

at worst, it seems fair to say that the gap in per core performance has become much narrower, and the amd modular fabric approach has been a big help.

pantherx12Xfr is just 100mhz on top of the boost assuming temperatures are fine and tdp is within specs. So it should still be 2 cores that are boosted. ( Basically it's marketing wank as I'm sure you picked up on it being the clever chap you are, they boost to 4.1 but AMD couldn't garantee 4.1 on every chip so came up with a word so they can advertised 4.1ghz without actually committing to it. )

I am definitely not 100% on this though I've been following all of this pretty closely but it's sometimes hard to remember what I've read in a forum or on an actual article.

Or it could just be the beta version, of the natty concept of intelligently auto overclocking in small increments, if the appropriate sensors among the zillions (~700?) amd have embedded in zen, say its ok.

the better u cool, the better it auto OCs. Simple. I like it.

It does sound a bit of a yawn now tho.

You need to figure out how to edit and add to posts instead of double+ posting throughout the thread.... :)

EarthDogYou need to figure out how to edit and add to posts instead of double+ posting throughout the thread.... :)

Mod This! :eek:

notb?

Honestly, you'll give a source for that $120 or not? Maybe you missed a zero and it's actually $1200? :)

The source is exactly the same guy ("Bits n Chips") who tweeted that Threadripper @ 16 cores would cost $849. At the same time he tweeted that the manufacturing costs for that chip are $110-120! Therefore, since the chip is little more than 2x the pins and 2x the cores, wired together, It stands to reason that a single 8-core CPU would cost about $60 to manufacture.

systemBuilderThe source is exactly the same guy ("Bits n Chips") who tweeted that Threadripper @ 16 cores would cost $849. At the same time he tweeted that the manufacturing costs for that chip are $110-120! Therefore, since the chip is little more than 2x the pins and 2x the cores, wired together, It stands to reason that a single 8-core CPU would cost about $60 to manufacture.

But should we believe him? How do we know he's a reliable source?
As for the "2x" idea - it's, sadly, not that simple. Production cost includes packaging, some allocated cost etc. This would make a difference.
Keep in mind Ryzen CPUs are often bundled with a cooler. :-)