• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

INTEL BRING THE BIG GUNS!

Silent_Scone said:
No, they don't.
Click to expand...

Whether they're disabled dies or just chunks of silicon doesn't really matter. The physical space is there. Yes AMD seem to suggest they're non functional and only there to support the IHS more uniformly but this seems a very odd thing to do from a cost point of view. Why go to the expense of connecting what looks very much like a silicon die, to the PCB? Why not just use something cheaper like a shim?

I've not seen anything conclusive about whether the electrical connections exist to put a working core into that physical space or even whether this would be possible. There is the potential though.
 
Last edited:
Minstadave said:
Seen mixed ideas on what the extra two "dies" are, AMD seem to suggest they're non functional and only their to support the IHS more uniformly. This seems a very odd thing to do from a cost point of view.

I've not seen anything conclusive about whether the electrical connections exist to put a working core in the physics space or whether this would be possible.
Click to expand...

It's not odd at all, really. It's necessary given the physical characteristics and CCX design similarities. Also the layout likely helps with thermals.

The downside is of course that it's rather large.
 
Silent_Scone said:
It's not odd at all, really. It's necessary given the physical characteristics and CCX design similarities. Also the layout likely helps with thermals.

The downside is of course that it's rather large.
Click to expand...

Has it ever been done before? I've not seen a chip with chunks of unused silicon on. Dead dies yes as part of binning but not chunks of silicon purely for load bearing.
 
Silent_Scone said:
Not to my knowledge, but then removing them completely would be interesting when applying pressure to the IHS with a cooler.
Click to expand...

So when you say it's not odd, it's never been done before and it also seems a very expensive way of doing it.

I suspect that there is a little more to the story than just it being there to support the IHS better.
 
There is something that bothers me about this claim that two of the Threadripper dies are not dies at all but just chunks of metal to support the over sized Heat Spreader.

Why have such a huge Heat Spreader where you need two dummy dies to support it structurally? that makes no sense, if Threadripper is designed like its only ever going to be a two die CPU then you wouldn't make it the size of a 4 die CPU and have a very big expensive 4 die socket.
Surly if your intention was just to have a 2 die CPU then you would make a much smaller much cheaper 2 die CPU.

Add to that are we supposed to believe that socket just happens to be the same size as the 32 core EPYC socket with the same number of pins?
I don't believe that, Threadripper is the same size with the same number of socket pins as EPYC because its the same CPU, those EPYC CPU's that didn't make it as 24 core CPU's become 12 and 16 core Threadrippers, which means yes, if AMD wanted to be really disruptive they could release a 20+ core Threadripper.

Threadripper is the same platform as EPYC, currently in 16 and 12 core form they may only have 2 real and 2 dummy dies but that's 2 dummy dies on a real interposer, its probably done because they don't need to put real dies on them for Threadripper, they could if they wanted to.

I think AMD just don't want to show their hand to Intel. "oh no no... its not a 4 die CPU, the other two are fake" it doesn't wash, not for me anyway.
 
Last edited:
humbug said:
There is something that bothers me about this claim that two of the Threadripper dies are not dies at all but just chunks of metal to support the over sized Heat Spreader.

Why have such a huge Heat Spreader where you need two dummy dies to support it structurally? that makes no sense, if Threadripper is designed like its only ever going to be a two die CPU then you wouldn't make it the size of a 4 die CPU and have a very big expensive 4 die socket.
Surly if your intention was just to have a 2 die CPU then you would make a much smaller much cheaper 2 die CPU.

Add to that are we supposed to believe that socket just happens to be the same size as the 32 core EPYC socket with the same number of pins?
I don't believe that, Threadripper is the same size with the same number of socket pins as EPYC because its the same CPU, those EPYC CPU's that didn't make it as 24 core CPU's become 12 and 16 core Threadrippers, which means yes, if AMD wanted to be really disruptive they could release a 20+ core Threadripper.

Threadripper is the same platform as EPYC, currently in 16 and 12 core form they may only have 2 real and 2 dummy dies but that's 2 dummy dies on a real interposer, its probably done because they don't need to put real dies on them for Threadripper, they could if they wanted to.

I think AMD just don't want to show their hand to Intel. "oh no no... its not a 4 die CPU, the other two are fake" it doesn't wash, not for me anyway.
Click to expand...

Well, there's only one way to find out ;)
 
Silent_Scone said:
Well, there's only one way to find out ;)
Click to expand...

Yes :) but you don't actually need to pull them off to look see... think about it, the dies are connected via an interposer in the CPU's PCB, that Interposer is not going to be shaped like a figure of 8 to support only the 2 real dies, is it?
its the same square interposer in the same PCB on the same socket as EPYC, they are EPYC CPU's, 2 of the dies may well be fake but as i said they are fake dies (because real ones are not needed) on a real interposer ready for real dies.
 
Last edited:
humbug said:
Yes :) but you don't actually need to pull them off to look see... think about it, the dies are connected via an interposer in the CPU's PCB, that Interposer is not going to be shaped like a figure of 8 to support only the 2 real dies, is it?
its the same square interposer in the same PCB on the same socket as EPYC, they are EPYC CPU's, 2 of the dies may well be fake but as i said they are fake dies (because real ones are not needed) on a real interposer ready for real dies.
Click to expand...

You're overthinking this. The substrate likely is the same, yes. The difference is they're not disabled or defect, they're simply not real (allegedly)
 
Silent_Scone said:
You're overthinking this. The substrate likely is the same, yes. The difference is they're not disabled or defect, they're simply not real (allegedly)
Click to expand...

That is what i said...

Yes :) but you don't actually need to pull them off to look see... think about it, the dies are connected via an interposer in the CPU's PCB, that Interposer is not going to be shaped like a figure of 8 to support only the 2 real dies, is it?
its the same square interposer in the same PCB on the same socket as EPYC, they are EPYC CPU's, 2 of the dies may well be fake but as i said they are fake dies (because real ones are not needed) on a real interposer ready for real dies.
Click to expand...
 
humbug said:
Yes :) but you don't actually need to pull them off to look see... think about it, the dies are connected via an interposer in the CPU's PCB, that Interposer is not going to be shaped like a figure of 8 to support only the 2 real dies, is it?
its the same square interposer in the same PCB on the same socket as EPYC, they are EPYC CPU's, 2 of the dies may well be fake but as i said they are fake dies (because real ones are not needed) on a real interposer ready for real dies.
Click to expand...

If you remember a little while ago when der8auer posted his Threadripper delid video, he was asked by AMD to take it down. I said at the time the only real long term reason for the 2 dummy dies was to make the insertion of one or 2 real dies an easy option for AMD whenever they wanted to. As Threadripper is an enthusiasts platform AMD know we will clock it "all cores" as high as possible. At the moment, getting 32 cores to 4ghz is not going to happen. I would bet though that about 18 months or so before Threadripper is EOL, that 4ghz all 32 core clock will be game on. At that point those 2 dummy places will cost AMD next to nothing to implement and Intel with absolutely no answer to.
 
humbug said:
There is something that bothers me about this claim that two of the Threadripper dies are not dies at all but just chunks of metal to support the over sized Heat Spreader.

Why have such a huge Heat Spreader where you need two dummy dies to support it structurally? that makes no sense, if Threadripper is designed like its only ever going to be a two die CPU then you wouldn't make it the size of a 4 die CPU and have a very big expensive 4 die socket.
Click to expand...

Main reason - 4096 pins. If the cpu package was smaller, the pins would have to be smaller as well, and they are already fragile enough. That's what comes with 64 pcie lanes - lots of pins needed.
 
amigafan2003 said:
Main reason - 4096 pins. If the cpu package was smaller, the pins would have to be smaller as well, and they are already fragile enough. That's what comes with 64 pcie lanes - lots of pins needed.
Click to expand...

Nah, that don't wash. Epyc is 128 PCI lanes and is the same size. Although current Threadripper mobo's have 64 PCI lanes, any future releases could easily cope with 128.
 
Its going to be cheaper to do this and get good yields than build 16 or 32 core dies that will sell in limited quantities, its a clever, cost effective solution. Through die testing Ryzen they get to cherry pick the parts and sell the duffers as 3s/5s/7s, no waste, get a duff 16/32 core chip and you loose all 16/32 or make too many and you have to sel 16/32 core chip as something lesser anyway so the cost of the silicon is still incurred.

The package cost will be cheaper too as the can sell the same to enterprise and retail upping the quantity, lowering the cost.

On the back end it will reduce the number of test vehicles required too, they can use same prober etc, cutting down time, cost and complexity, so on the one hand it seems crazy but actually, fair play to them.
 
You must log in or register to reply here.
Back
Top Bottom