AMD '#Rekt son' reply to Intel's "four glued together" statement

[Not_Sean]

what I don't get is these are going to be used in large business, who have IT departments, I would love to meet the IT manager that would fall for this BS advertising, yeah maybe the other staff, but before any company outlays moneys like this thy'll do some research.

 

This go from Intel is something that my mom would say oh but I heard Intel is much better, but a It Professional I just can't see it working on. 

[Drak3]

Just now, Not_Sean said:

what I don't get is these are going to be used in large business, who have IT departments, I would love to meet the IT manager that would fall for this BS advertising, yeah maybe the other staff, but before any company outlays moneys like this thy'll do some research.

 

This go from Intel is something that my mom would say oh but I heard Intel is much better, but a It Professional I just can't see it working on. 

It doesn't have to work on the IT professional. It has to work on the guy that green lights tech purchases.

[XenosTech]

Just now, Not_Sean said:

what I don't get is these are going to be used in large business, who have IT departments, I would love to meet the IT manager that would fall for this BS advertising, yeah maybe the other staff, but before any company outlays moneys like this thy'll do some research.

 

This go from Intel is something that my mom would say oh but I heard Intel is much better, but a It Professional I just can't see it working on. 

Not all IT departments are going to be informed as others some business are like regular consumers and just buy what they're familiar with

[leadeater]

21 minutes ago, Hunter259 said:

It is straight up worse. You can't out do the laws of physics. Theres a reason 100ns is added to the base latency when talking outside even CCX's. You can make it as small as you can but there will always be a hit when you talk outside of the cores themselves.

That 100ns isn't related to distance between dies though. Intel QPI link between CPUs is only 40ns which is a greater difference also leaving the CPU package, what you are seeing is bad L3 latency not IF latency.

[SpaceGhostC2C]

2 hours ago, RadiatingLight said:

TBH that ecosystem slide looks very unimpressive.

That don't even have the Hitachi twins. Such a poor ecosystem. Sad. 

 

 

2 hours ago, RadiatingLight said:

I'm willing to believe that Hitachi was just an honest mistake, if it was really the only duplicated brand.

It wasn't. But the duplicates are a minor concern, they can be honest mistakes that you make while cramming in as many random logos as possible in a slide for no good reason. What we should ask ourselves is why is Hitachi (an many others) there even once. I'm which way is it part of any CPU manufacturer's  "ecosystem"? 

They may as well list the electrical companies for supplying 110/220V to Intel - based sereverd... 

 

(OP: I'm pretty sure "determinism" in the quote was just an autocorrect fail when transcribing the speech...) 

[XenosTech]

1 minute ago, leadeater said:

That 100ns isn't related to distance between dies though. Intel QPI link between CPUs is only 40ns which is a greater difference also leaving the CPU package, what you are seeing is bad L3 latency not IF latency.

Can't remember who the graph was from but it showed ryzen having lower latency than the x299 cpu's

[NumLock21]

2 hours ago, DeadEyePsycho said:

Hitachi

DDN

 

You actually went to look for it?!

 

Intel partners is seriously enormous compare to AMD's. Below is the actual link to those logos in their slide.

https://builders.intel.com/membership

[AnonymousGuy]

1 hour ago, Jito463 said:

Being a different method, does not automatically make it an inferior method.

It's a cheaper method and you get what you pay for.  No one is going to optimize their software for AMD so if your software and tools doesn't work right out of the gate with the assload of NUMA domains then you're going to buy Intel.

[AnonymousGuy]

3 minutes ago, XenosTech said:

Can't remember who the graph was from but it showed ryzen having lower latency than the x299 cpu's

Until you needed to hop outside the CCX and then it shit itself, and that wasn't even on TR or Epyc where it then has to hop across IF.

[XenosTech]

1 minute ago, AnonymousGuy said:

It's a cheaper method and you get what you pay for.  No one is going to optimize their software for AMD so if your software and tools doesn't work right out of the gate with the assload of NUMA domains then you're going to buy Intel.

That's a rather bold statement to make... People will end up optimising for AMD since software optimisation is a thing unless the company for teh software went out of business or something.

 

1 minute ago, AnonymousGuy said:

Until you needed to hop outside the CCX and then it shit itself, and that wasn't even on TR or Epyc where it then has to hop across IF.

And it still wasn't that much higher than x299  when it went across the IF

[2FA]

7 minutes ago, NumLock21 said:

You actually went to look for it?!

 

Intel partners is seriously enormous compare to AMD's. Below is the actual link to those logos in their slide.

https://builders.intel.com/membership

Mate, I knew about the Hitachi from the WAN show and I recognized the double DDN instantly when I looked at that slide. I didn't know the Costco one but I did know there was a third based off of the article on Tomshardware.

[AnonymousGuy]

1 minute ago, XenosTech said:

That's a rather bold statement to make... People will end up optimising for AMD since software optimisation is a thing unless the company for teh software went out of business or something.

 

And it still wasn't that much higher than x299  when it went across the IF

 

This is the one.  Doesn't include IF to another chip on the package.  Going up 3.5x in latency depending on your workload is pretty well qualified as "inconsistent performance" as mentioned on the Intel slides.

[RadiatingLight]

7 minutes ago, AnonymousGuy said:

 

This is the one.  Doesn't include IF to another chip on the package.  Going up 3.5x in latency depending on your workload is pretty well qualified as "inconsistent performance" as mentioned on the Intel slides.

True, but consistent bad < inconsistent bad/good

[Not_Sean]

17 minutes ago, Drak3 said:

It doesn't have to work on the IT professional. It has to work on the guy that green lights tech purchases.

 

17 minutes ago, XenosTech said:

Not all IT departments are going to be informed as others some business are like regular consumers and just buy what they're familiar with

While I agree with both of your points for normal IT, I've worked in corporate It (Worked for Audi & VW South Africa Head Office), this stuff is not cheap this isn't a little store going out and dropping this kind of money on a IBM / HP Solution for their server room, those guys are buying cheapest.

 

We did 19 new IBM servers and the only one who presented the quotes is the IT manager, the CFO and GM can comment all they want but business's understand that they hired a guy for the job, most they will say and we've had it is that option is to expensive come back with cheaper option, they do not care  what brand/ models are chosen. 

 

I don't dispute that this post with intel will cause big miss understanding with IT guys, but the people who are buying the big numbers and the dual socket Rack servers won't be fooled by this, and if they are than they shouldn't be in the position they are in. 

[Jito463]

44 minutes ago, Hunter259 said:

It takes engineering to get a solid manufacturing node that will work and to make a system inside the die for those cores to talk to each other. Good lord, AMD used to make fun of the Core 2 Quad for being MCM for the exact same reasons. 

No, not for the exact same reasons.  Hmm, I wish I had thought to mention what those reasons were.  If only I had posted that information somewhere.

 

Oh, wait.  I DID!

1 hour ago, Jito463 said:

True, but recall that the Intel method was flawed, due to their reliance on the NB bus for interconnects between the dies.

Once again, words and words without any comprehension of them.

31 minutes ago, Hunter259 said:

Oh it's a fantastic product that is much better for the price. My argument was never over if it was a good product or not. Just saying that calling Intel childish and wrong over this is silly.

It is a bit childish, but honestly I laugh at it and move on.  What I do consider childish, however, is your constant need to defend Intel's design choices by deriding AMD.  Yes, there is a latency hit for MCM (not like Intel's version, which I once again explained above), but saying it's the less elegant/cruder is a subjective opinion.  There is no objective basis for that statement.  And saying it's less technologically advanced - when it allows bypassing a current technological limitation while being cost effective - is simply being disingenuous.

 

I honestly don't know why I'm bothering to respond again, since you'll probably just ignore what I wrote and respond to your own preconceived notion of my post, but I figure it's worth one last attempt.

[SpaceGhostC2C]

 

2 hours ago, Hunter259 said:

. Multiple dies is a less elegant and technologically advanced,

That's a pretty empty claim to belong in engineering or computer science. I mean, "elegant"? leave that to Versace or whatever

2 hours ago, Hunter259 said:

I also never said it had bad performance just that it is inherently inferior

So it's inferior in a sense that doesn't involve worse performance? Meaning what then? Are we discussing technology or poetry? 

1 hour ago, Hunter259 said:

I already explained why it's worse and you don't deny that it is so I'm confused to why you can't understand that it is worse in a technical standpoint. 

Except you haven't. You see, you are focusing on a very narrow object, "cross-die latency", and basing your whole assessment of the approach on that. You forget that we are taking about server CPUs, and that many other factors are as important. Or more, like I/O. Have you seen how it works on monolithic high core count chips? How there is a hierarchy in terms of I/O access, with only some cores having direct access? Your latency claims get reveresed when we discuss access to I/O latency, a "slightly relevant" issue in the data center context... 

 

Si, no, it's not that easy to rank these designs. 

 

57 minutes ago, Drak3 said:

It doesn't have to work on the IT professional. It has to work on the guy that green lights tech purchases.

I don't know why people talk like it takes a Nobel prize to see through those slides... Frankly, they insult the intelligence of most any reader, no need for any specific training. 

 

(I also love this forum seeing how, when there was a discussion about the gender wage gap,  every form was super rational and the market would drive out anyone deviating from perfect profit maximization, but the moment we discuss corporate hardware, all firms are run by idiots and any attempt to fool them works ). 

[Drak3]

Just now, SpaceGhostC2C said:

I don't know why people talk like it takes a Nobel prize to see through those slides... Frankly, they insult the intelligence of most any reader, no need for any specific training. 

Seeing as Intel's slides were internal use, for Intel's salesmen, it doesn't matter what anyone outside would have thought about it.

 

1 minute ago, SpaceGhostC2C said:

I also love this forum seeing how, when there was a discussion about the gender wave gap,  every form was super rational and the market would drive out anyone deviating from perfect profit maximization, but the moment we discuss corporate hardware, all firms are run by idiots and any attempt to fool them works

There is a difference between knowing how to maximize profits by adjusting wages and most costs, and knowing how to balance IT Infrastructure to maximize profits.

[SpaceGhostC2C]

3 minutes ago, Drak3 said:

 

There is a difference between knowing how to maximize profits by adjusting wages and most costs, and knowing how to balance IT Infrastructure to maximize profits.

Oh, no, I'm not even going there again, but have fun! 

[XenosTech]

8 minutes ago, Drak3 said:

There is a difference between knowing how to maximize profits by adjusting wages and most costs, and knowing how to balance IT Infrastructure to maximize profits.

I'm confused... Is he assuming that everyone on this planet who's in charge of making purchases for IT departments are rational and sensible people 100% of the time ? If he so I'd like for him to meet the man who told me that an i5-650 and an i5-6500 were the same processor

[Drak3]

Just now, XenosTech said:

I'm confused... Is he assuming that everyone on this planet who's in charge of making purchases for IT departments are rational and sensible people 100% of the time ? If he is I'd like for him to meet the man talk told me that an i5-650 and an i5-6500 were the same processor

Apparently.

[sazrocks]

1 hour ago, bcguru9384 said:

curious? 

did you circle those using win 10?

1. Why do you care?

2. I didn’t make the picture, I just found it on r/amd.

[The Sloth]

4 minutes ago, XenosTech said:

i5-650 and an i5-6500 were the same processor

not that simple 

[XenosTech]

1 minute ago, Drak3 said:

Apparently.

Damn. That same man also told me that anything above 4gb of RAM for desktop use was waste. Shit that was the longest 3 weeks of my life trying to convince some one on how to properly upgrade their client's pcs in the future but at least I didn't have to tell him to put in a ssd

[leadeater]

@SpaceGhostC2C

Also have to wonder how many people commenting actually work in large data centers so have the working experience to back up what they are saying and not just regurgitating what is unbeknownst to them Intel's marketing and definition of how to do things .

 

Not to say that some of the stuff being said isn't true but if you don't know what the actual impact is and how it effects design and resourcing better phrasing to reflect that should really be used.

[mr moose]

1 hour ago, Trik'Stari said:

Not this time around lol

 

They might be eating bluegrass, but the green grass remains. (which is sad. I don't get how they hit such a home run here, but then decide to completely drop the GPU ball)

 

The cow is not AMD though.   The cow simply represents amusing event, it is how AMD responds to it that determines how much "green" gets eaten.