No Dummies please

Pieter3dnow said:
post count ;) I mean he says something then proves what he says does not matter then replies anyway as if it mattered.
Click to expand...
actually looks more like not making a DIRECT statement as to quell any real argument or debate, so when he LOOKS to be proven wrong or bias he can then claim something else. Obfuscate, obfuscate...
 
JustReason said:
actually looks more like not making a DIRECT statement as to quell any real argument or debate, so when he LOOKS to be proven wrong or bias he can then claim something else. Obfuscate, obfuscate...
Click to expand...

Well there is something wrong just not with Threadripper :) .
 
_mockingbird said:
They are probably not functioning dies.
Click to expand...

At this point I think it's likely that the worst chips on amd ryzen wafers are being used as the supports for ThreadRipper. Though I can't comment on how much worse they are xompared to 4 or 6 core variants.
 
mnewxcv said:
At this point I think it's likely that the worst chips on amd ryzen wafers are being used as the supports for ThreadRipper. Though I can't comment on how much worse they are xompared to 4 or 6 core variants.
Click to expand...

They probably have non-functioning memory controllers or something else that makes them completely unusable.
 
JustReason said:
then, pray tell, what is your point?
Click to expand...

Why are you bothering with him, even Dan told him no one gives a shit and it's splitting hairs on a terminology. You know he will spin everything to make it sound bad for AMD, no matter where the actual truth may be.
 
juanrga said:
The point is clearly stated on my former posts in this thread.
Click to expand...
No it isn't otherwise I and others wouldn't have to ask.

You make it sound like a problem. It obviously isn't as the other 2 dies, whatever their makeup, change nothing as far as performance.

You said AMD lied and no where in any report of statements MADE by AMD do they lie. They state dummy dies and in all respects of the meaning they are in fact dummies as they have no way to impact performance being no traces or connections if you will. Because you didn't understand or your inability to accept reality in contradiction to your narrative, none of what AMD stated changes nor does it constitute a lie any measure.

So again, What exactly is your point?
 
Gideon said:
Why are you bothering with him, even Dan told him no one gives a shit and it's splitting hairs on a terminology. You know he will spin everything to make it sound bad for AMD, no matter where the actual truth may be.
Click to expand...
And that is the point, as you stated.
 
Is it possible for regular (very talented obviously) human beings to hack a die off one substrate onto another? They're placed on a ball grid array, aren't they? And can we reverse engineer which traces go where?
 
michalrz said:
Is it possible for regular (very talented obviously) human beings to hack a die off one substrate onto another? They're placed on a ball grid array, aren't they? And can we reverse engineer which traces go where?
Click to expand...

Why bother? There is no way to get it to work if you even had the means. Threadripper and x399 do not have the electrical connections nor was this socket ever made to have 4 dies active. In other words you'd have to connect them to the Infinity Fabric, and um who has the ability to do that? If anyone could enable 4 dies on this socket, it is going to be by AMD.

I just don't get the fascination with this.
 
thesmokingman said:
Why bother? There is no way to get it to work if you even had the means. Threadripper and x399 do not have the electrical connections nor was this socket ever made to have 4 dies active. In other words you'd have to connect them to the Infinity Fabric, and um who has the ability to do that? If anyone could enable 4 dies on this socket, it is going to be by AMD.

I just don't get the fascination with this.
Click to expand...

My motives are purely academic. I'm not that fascinated, I realize this is not the Athlon II anymore.

Just wanted to know if one can change the dies around the board to check if they're broken. I know nothing about the infinity fabric but I'll google :)
 
michalrz said:
My motives are purely academic. I'm not that fascinated, I realize this is not the Athlon II anymore.
Just wanted to know if one can change the dies around the board to check if they're broken. I know nothing about the infinity fabric but I'll google :)
Click to expand...

It is what is used/needed to communicate between the dies, if that is not the case no beuno.
 
JustReason said:
No it isn't otherwise I and others wouldn't have to ask.

You make it sound like a problem. It obviously isn't as the other 2 dies, whatever their makeup, change nothing as far as performance.
Click to expand...

I know that this doesn't change performance. I stated in more than one post that "this discussion about dies is irrelevant for final users," If you could read what I wrote...

JustReason said:
You said AMD lied and no where in any report of statements MADE by AMD do they lie.
Click to expand...

Precisely der8hauer purchased a retail chip and dismantled to check if AMD was lying or not. He found that AMD lied. Relevant quotes have also been given in this thread. I will repeat some of the quotes published in the press for you:

Update: AMD has clarified that two of the four silicon die are 'dummy' blanks rather than disabled Ryzen chips. They are included for structural stability.
Click to expand...

AMD responded that two of the die on shipping Threadripper models are not the same as those Der8auer found in the engineering sample - the additional two die are "dummy" die that the company uses to provide structural stability for the package.
Click to expand...

How far would you go to prove a point? Well, the overclocker and CPU delidding expert der8auer has decided to put AMD's "Dummy Dies" claim to the test, to find out once and for all whether or not AMD is using failed/defective Ryzen dies or silicon spacers.
Click to expand...

He found retail ThreadRipper has four Ryzen dies, two enabled and two disabled, just as the engineering sample. AMD lied. Period.
 
juanrga said:
He found retail ThreadRipper has four Ryzen dies, two enabled and two disabled, just as the engineering sample. AMD lied. Period.
Click to expand...

So where is the proof of all this lying ? Anyone made it work on 2 dummy dies or not?
 
This is just one big argument between the pot and kettle calling each other black.
 
juanrga said:
I know that this doesn't change performance. I stated in more than one post that "this discussion about dies is irrelevant for final users," If you could read what I wrote...



Precisely der8hauer purchased a retail chip and dismantled to check if AMD was lying or not. He found that AMD lied. Relevant quotes have also been given in this thread. I will repeat some of the quotes published in the press for you:







He found retail ThreadRipper has four Ryzen dies, two enabled and two disabled, just as the engineering sample. AMD lied. Period.
Click to expand...
I am just gonna chock this one up to you're being terribad at language interpretation. Der8hauer is a MORON then, apparently he too has comprehension skills. At no point in AMDs statement do they speak to the chemical/physical makeup of the dummy dies just that they are structural support and therefore lack the necessary connections to be "turned on". He inferred they meant not dies of any sort and therefore wasted his money proving nothing in contrast to what he knew before and thusly wrangled a few of you gullible types in.

Now if he wanted to further investigate, then I would suggest looking at his original TR and looking for traces as it may actually have traces and those dies are "turned off". Hence AMDs clarification and the subsequent press releases. And the only lying here is the narrative you are posturing, emperors new clothes and all that.
 
_mockingbird said:
They probably have non-functioning memory controllers or something else that makes them completely unusable.
Click to expand...

According to AMD, they have no traces beyond the dies themselves. They do not and cannot be connected to anything?

Gideon said:
You know he will spin everything to make it sound bad for AMD, no matter where the actual truth may be.
Click to expand...

What gives you that impression? I have no preference between either manufacturer. I've often had more Intel CPUs because of AMD's half-assed platforms and Intel having better performance over most of the last two decades I've been involved in the computing industry. Price / performance ratio's aside, these are facts which can be proven. AMD's platforms have often performed worse in some ways, were less flexible, or were more problematic. I've pointed out the reasons for this many times. AMD's CPU's over the last 10 years were lacking compared to their Intel counterparts. Again, this isn't bias or fluff, it's fact. How does this make me biased?

I've also been a fan of Ryzen / Threadripper and I've said as much in many threads. Yes, the platform annoys me with memory compatibility issues and a few other problems, but I'm far more pissed off at Intel right now for it's VROC and recent high end CPU pricing fuckery. I like and have liked products produced by both companies. I think both can be bone headed at times and call each of them out on things I do not like. It just happens to be that AMD has had a worse track record over the last couple of decades. That's not bias, that's just how it is.

michalrz said:
Is it possible for regular (very talented obviously) human beings to hack a die off one substrate onto another? They're placed on a ball grid array, aren't they? And can we reverse engineer which traces go where?
Click to expand...

No.

thesmokingman said:
Why bother? There is no way to get it to work if you even had the means. Threadripper and x399 do not have the electrical connections nor was this socket ever made to have 4 dies active. In other words you'd have to connect them to the Infinity Fabric, and um who has the ability to do that? If anyone could enable 4 dies on this socket, it is going to be by AMD.

I just don't get the fascination with this.
Click to expand...

Agreed. I'll elaborate and spell this out one more time. People fascinated with this idea are wanting an Epyc CPU at a discount or wanting to relive the glory days of unlocking a fourth core, or pencil modding previous CPUs. That's not going to happen here and frankly, you don't want it to. Two dies are active in Threadripper. Assuming those are basically identical to the two functional dies, you would end up with double the TDP of a stock Threadripper. Threadripper struggles to achieve 4.0GHz / 4.1GHz clock speeds through overclocking and even the beefiest AIO's can't handle it. You need a good custom loop to achieve these results without thermal throttling. Remember, AMD is using these extra dies as jack stands for the heat spreader. It's unlikely they work, or work well. Threadripper is supposed to represent the top 5% of all Ryzen silicon. It's probably safe to say that the non-functional dies didn't make the cut given they are being used like cinder blocks.

The next thing you have to understand, is that AMD has a history of cutting it close with it's VRM implementations. Intel tends over specs it's VRD specifications and plans them out well in advance for refreshes and future products. AMD has a poor history of doing that. While I haven't delved into the white papers deep enough, or looked at the memory design guidelines for socket TR4 motherboards, I am almost certain that X399's current VRM design is in adequate for a four die Threadripper CPU.

Let's suppose that it could accommodate a lower voltage chip with lower clocks. That seems reasonable right? Something like 2.0GHz with a 2.5GHz boost clock could be useful with 32 cores on a socket TR4 motherboard. Well, there are still a couple of issues with that idea. Let's assume the TDP's can be lowered enough and TR4's existing motherboard designs could handle the power draw. Let's also assume that current cooling designs would be enough for such a CPU. After all those issues are overcome, we still have more problems. We've only got 4 memory channels, not the eight that a four die Threadripper would need. So what does that mean? It means that a four die Threadripper needs to be built to allow the extra two dies to access memory remotely, and to share NUMA nodes with another processor die. It is highly unlikely that Threadripper was built to do this, nor was X399. Even if Threadripper could have two cores share a memory bus, and access memory remotely across NUMA nodes, this wouldn't be an ideal situation for desktop CPUs, and rarely a good idea for server applications. There would be massive latency penalties incurred from such a design.

X399 and Threadripper, by AMD's own admission was an afterthought. It was spurred by Ryzen's exceeding AMD's internal expectations, enough to make it compelling as an HEDT alternative to Intel's X99 and then upcoming X299 platforms.

michalrz said:
My motives are purely academic. I'm not that fascinated, I realize this is not the Athlon II anymore.

Just wanted to know if one can change the dies around the board to check if they're broken. I know nothing about the infinity fabric but I'll google :)
Click to expand...

You can't just "move" dies around yourself. CPUs can rarely be modified in any such way. They have to be built the way they are. The dies are not in sockets and can't be switched around by the end user.
 
Dan_D said:
According to AMD, they have no traces beyond the dies themselves. They do not and cannot be connected to anything?

Gideon said: ↑

Why are you bothering with him, even Dan told him no one gives a shit and it's splitting hairs on a terminology. You know he will spin everything to make it sound bad for AMD, no matter where the actual truth may be.


What gives you that impression? I have no preference between either manufacturer. I've often had more Intel CPUs because of AMD's half-assed platforms and Intel having better performance over most of the last two decades I've been involved in the computing industry. Price / performance ratio's aside, these are facts which can be proven. AMD's platforms have often performed worse in some ways, were less flexible, or were more problematic. I've pointed out the reasons for this many times. AMD's CPU's over the last 10 years were lacking compared to their Intel counterparts. Again, this isn't bias or fluff, it's fact. How does this make me biased?

I've also been a fan of Ryzen / Threadripper and I've said as much in many threads. Yes, the platform annoys me with memory compatibility issues and a few other problems, but I'm far more pissed off at Intel right now for it's VROC and recent high end CPU pricing fuckery. I like and have liked products produced by both companies. I think both can be bone headed at times and call each of them out on things I do not like. It just happens to be that AMD has had a worse track record over the last couple of decades. That's not bias, that's just how it is.
Click to expand...

I was not talking to you Dan, I was talking about Juanrga. I agreed with your assertion that AMD lied an it's some kind of big deal is stupid. You and I may not always agree on things but my comment was not about you. Most times I mention him by name I get a warning so I tend not to use Juanrga name, just a FYI.
 
Removing the dies will most likely destroy them. The first time Der8hauer delidded his TR the chip was non functional. I don't think he tested that this time.
 
Dan_D said:
According to AMD, they have no traces beyond the dies themselves. They do not and cannot be connected to anything?
.
Click to expand...

We all know this. He and I were speculating how dies are selected to be the support dies.
 
Gideon said:
I was not talking to you Dan, I was talking about Juanrga. I agreed with your assertion that AMD lied an it's some kind of big deal is stupid. You and I may not always agree on things but my comment was not about you. Most times I mention him by name I get a warning so I tend not to use Juanrga name, just a FYI.
Click to expand...

My fault, I misunderstood who the comment was directed at because my name was mentioned in the post.
 
juanrga said:
I know that this doesn't change performance. I stated in more than one post that "this discussion about dies is irrelevant for final users," If you could read what I wrote...



Precisely der8hauer purchased a retail chip and dismantled to check if AMD was lying or not. He found that AMD lied. Relevant quotes have also been given in this thread. I will repeat some of the quotes published in the press for you:







He found retail ThreadRipper has four Ryzen dies, two enabled and two disabled, just as the engineering sample. AMD lied. Period.
Click to expand...

der8hauer found 1 cpu had 4 ryzen dies, that does not automatically mean that all TR cpu's have 4 ryzen dies. Was it a retail stepping ES that was remarked? Are some shipping with blanks and some shipping with defective CCX's?

Since they can't be activated in any way the whole thread is pointless bollocks. AMD hater's will use this as evidence that AMD lied about the origins of the cpu, AMD fanboys are looking for a cheap pencil mod for extra cores.

When TR gets more cores it will be on 2 dies using Zen+/Zen2 or whatever its called this week, using the 7nm process
 
mnewxcv said:
We all know this. He and I were speculating how dies are selected to be the support dies.
Click to expand...

I doubt we will ever know for sure, but I suspect these are either totally defective dies that wouldn't have been functional under any circumstances or they are dies that are out of spec for their desired purpose.
 
Nathan_P said:
der8hauer found 1 cpu had 4 ryzen dies, that does not automatically mean that all TR cpu's have 4 ryzen dies. Was it a retail stepping ES that was remarked? Are some shipping with blanks and some shipping with defective CCX's?
Click to expand...

It's true that not all Threadripper CPUs may have four actual dies in them. I suspect that it may be the case in the short term, and as manufacturing yields improve, these may be replaced with a silicon blank or something else entirely. As for engineering samples being remarked retail, that's highly unlikely if not bordering on impossible. Most, if not all the engineering samples for a given product cycle will be marked and set aside before the retail run begins. They may in fact be final retail silicon, and often are, but their microcode is different. Engineering samples are marked as such in CPU-Z. I don't see processor manufacturers remarking these CPU's and updating their microcode to match a retail CPU after they are already made and then boxing it up.

Nathan_P said:
Since they can't be activated in any way the whole thread is pointless bollocks. AMD hater's will use this as evidence that AMD lied about the origins of the cpu, AMD fanboys are looking for a cheap pencil mod for extra cores.

When TR gets more cores it will be on 2 dies using Zen+/Zen2 or whatever its called this week, using the 7nm process
Click to expand...

It is pointless and I don't understand why anyone would bitch about this. What difference does it make if the CPU is a remarked Epyc CPU or something purpose built with non-functioning blanks or failed dies that don't connect to anything? I agree with you on the last part of your statement. I can't imagine we'll see Threadripper CPU's with more than 16 cores until Ryzen+ or Ryzen2, or whatever it's called comes out. AMD has to get the TDP of Threadripper while maintaining reasonably high clocks before they can consider adding more cores / dies to it. Remember, Threadripper is an HEDT / Workstation CPU. It is not a server CPU. Therefore, dropping the clock speed isn't what AMD would want to do in order to provide us more cores.
 
Nathan_P said:
der8hauer found 1 cpu had 4 ryzen dies, that does not automatically mean that all TR cpu's have 4 ryzen dies. Was it a retail stepping ES that was remarked? Are some shipping with blanks and some shipping with defective CCX's?
Click to expand...

He found four dies on the engineering sample. Then AMD forced him to remove the original video with the demonstration and said him that retail chips are different. He purchased a retail chip and found again four dies. Causality? I don't believe it.

Nathan_P said:
Since they can't be activated in any way the whole thread is pointless bollocks. AMD hater's will use this as evidence that AMD lied about the origins of the cpu, AMD fanboys are looking for a cheap pencil mod for extra cores.
Click to expand...

This thread is irrelevant for final users, as I stated a couple of times. It is relevant for certain other people that is neither a fanboy nor a hater, but is genuinely interested in the technology or in the economy behind it.
 
juanrga said:
I stated in more than one post that "this discussion about dies is irrelevant for final users," If you could read what I wrote...
Click to expand...

juanrga said:
This thread is irrelevant for final users, as I stated a couple of times, it is relevant for certain other people that is neither a fanboy nor a hater.
Click to expand...
then stop replying. there is nothing more to discuss anyways...
 
Meeho said:
Maybe not 4 GHz, but it should be easier for them to achieve higher clocks than Intel's big monolithic dies. And the cores are much more efficient in the lower 3.X range.


That doesn't seem likely if there are always the same two functional cores in TR.
Click to expand...

They are, but doubling up on them would most likely put them at a TDP they'd never be able to deal with until Zen+ or whatever it's going to be called comes out.
 
You must log in or register to reply here.
Back
Top