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

[Hunter259]

2 minutes ago, Jito463 said:

I can't say I agree with that.  As with anything in life, there are pros and cons to both methods.

 

True, but recall that the Intel method was flawed, due to their reliance on the NB bus for interconnects between the dies.  While IF does scale based on your RAM speed, the actual communication between CCX's still occurs on-die.  Yes, there will be a latency hit, but the extent of the latency determines whether it's a "worse way".  The primary thing holding back Ryzen isn't IF, it's the silicon process from GloFo.  If they could breach that 4GHz barrier, then Ryzen would be a knockout.

 

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

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. What it can do is make it easier to add a fuckton more cores in a single socket by upping the success rate of the manufacturing. It's extremely similar to QPI and UPI that Intel use for talking between CPU's and PCH which have similar bandwidth to IF but the same increased latency the moment you step outside the die. In a perfect work the entire computer would be inside the die talking directly between cores but that simply isn't possible.

[Marinatall_Ironside]

17 hours ago, Hunter259 said:

Uh. This is something that is not changing and is basic Computer Science/Engineering. Multiple dies is a less elegant and technologically advanced, although as we've pushed silicon this far it may be the only good way to get past 40 cores, and anytime you leave the die you will get a latency/bandwidth hit. It'll perform like Ryzen does just with more cores due to Ryzen already being a somewhat multi-die chip.

So does everything else in the Computer Science field, and literally anything else in life. Once it's outdated you have to update it so you stay relevant.

 

I still remember trying to talk a streamer with a Computer Science degree out of X299 with a set budget out of USD $600 (best you can get with USD $600 on X299 is an Intel Core i7-7740X + dirt-cheap X299 board, which is still not cheap - USD $219 is the lowest you can go right now, and the 7740X alone makes absolutely zero sense and should never, EVER be considered for anything on X299. Go 7700K and Z270 if you want Intel 4-core), and that got my ass kicked from her discord, banned from her Twitch chat, and blocked on Twitter (which is my public handle) after trying to flaunt her degree like a douchebag and asking whether I was going to college or not (which is very offending considering that I'm three years in, about to go back in the Fall, studying Multimedia. That's like telling someone "you're stupid/incompetent because you didn't go to college, hur, hur, hur"). That tells me that the streamer is really not competent in making not only sensible, but logical decisions in this field because she can (or could have if she already followed through with the X299 idea) get a whole hell of a lot more performance for what she does for a very similar price (hell, less, like USD $400-500 with Ryzen 7 1700 + pretty damn good X370 board like the Asus PRIME X370-PRO).

 

Ryzen being "somewhat" a multi-die chip doesn't mean it will have serious performance disparities compared to Intel. AMD's CPUs are already not very far away from Intel.

 

16 hours 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.  While IF does scale based on your RAM speed, the actual communication between CCX's still occurs on-die (and the communication between dies on TR/Epyc occurs on the silicon).  Yes, there will be a latency hit, but the extent of the latency determines whether it's a "worse way".  The primary thing holding back Ryzen isn't IF, it's the silicon process from GloFo.  If they could breach that 4GHz barrier, then Ryzen would be a knockout.

 

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

 

Edited July 19, 2017 by JurunceNK
Grammar corrections and added additional personal notes regarding X299 vs. AMD Ryzen

[Hunter259]

Just now, JurunceNK said:

So does everything else in the Computer Science field, and literally anything else in life. Once it's outdated you have to update it so you stay relevant.

 

I still remember trying to talk a streamer with a Computer Science degree out of X299 with a set budget out of USD $600 (best you can get with USD $600 on X299 is an Intel Core i7-7740X + dirt-cheap X299 board, which is still not cheap - USD $219 is the lowest you can go right now), and that got my ass kicked from their discord, banned from their Twitch chat, and blocked on Twitter, which is my public handle after trying to flaunt her degree like an douchebag and asking whether I was going to college or not (which is very offending considering that I'm three years in, about to go back in the Fall, studying Multimedia. That's like telling someone "you're stupid/incompetent because you didn't go to college, hur, hur, hur"). That tells me that the streamer is really not competent in making not only sensible, but logical decisions in this field because she can (or could have if she already followed through with the X299 idea) get a whole hell of a lot more performance for what she does for a very similar price (hell, less, like USD $400-500 with Ryzen 7 1700 + X370 board like the Asus PRIME X370-PRO).

 

Ryzen being "somewhat" a multi-die chip doesn't mean it will have serious performance disparities compared to Intel. AMD's CPUs are already not very far away from Intel.

 

 

Even people with degrees are idiots. I'm going into my second year for Computer Engineering and I can see similar people who can get grades but have no common sense. I also never said it had bad performance just that it is inherently inferior

[YedZed]

42 minutes ago, nerdslayer1 said:

they are good at eating things that are green( its a joke) 

I love the elegance of having every possible rate issued just once to a comment. Please, everyone, do not ruin this beautiful thing. 

[Kherm]

1 hour ago, nerdslayer1 said:

or having misleading once like Gigabyte and Microsoft. 

 

While I am certainly siding with AMD on this one -intel is digging a hole here- how is Microsoft misleading? Even AMD lists Ms as a partner three separate times with Windows Server, Visual Studio, and one again under "momentum across the industry"

[Jito463]

13 minutes ago, Hunter259 said:

What it can do is make it easier to add a f***ton more cores in a single socket by upping the success rate of the manufacturing

That doesn't sound less technologically advanced to me.  It sounds like a technological improvement that bypasses limitations of the current method.

7 minutes ago, Hunter259 said:

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

If you'll re-read my post, I never denied that latency was higher when using multi-die configurations.  What I was objecting to, was your claim that it was less elegant (elegance is a subjective term, in my opinion) and that it was less technologically advanced (which I addressed above).

[The Sloth]

3 minutes ago, Kherm said:

While I am certainly siding with AMD on this one -intel is digging a hole here- how is Microsoft misleading? Even AMD lists Ms as a partner three separate times with Windows Server, Visual Studio, and one again under "momentum across the industry"

it looked like Intel was trying to say that Microsoft had no AMD support, AMD just listed partners without attacking Intel. 

[Jito463]

3 minutes ago, Kherm said:

While I am certainly siding with AMD on this one -intel is digging a hole here- how is Microsoft misleading? Even AMD lists Ms as a partner three separate times with Windows Server, Visual Studio, and one again under "momentum across the industry"

In my personal opinion on the matter, I could see it being misleading because they list MS as part of their ecosystem, which would technically also apply to AMD as well.  Yet they do not list MS or any companies that partner with AMD on the AMD ecosystem side.  Thus making it appear that these companies are exclusively part of the Intel ecosystem.

[Hunter259]

3 minutes ago, Jito463 said:

That doesn't sound less technologically advanced to me.  It sounds like a technological improvement that bypasses limitations of the current method.

If you'll re-read my post, I never denied that latency was higher when using multi-die configurations.  What I was objecting to, was your claim that it was less elegant (elegance is a subjective term, in my opinion) and that it was less technologically advanced (which I addressed above).

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. The product performs excellently and I have never made shots on it's performance. I just said Intel's claim is correct in that it is a "worse" way of achieving it.

[Kherm]

1 minute ago, nerdslayer1 said:

it looked like Intel was trying to say that Microsoft had no AMD support, AMD just listed partners without attacking Intel. 

 

Just now, Jito463 said:

In my personal opinion on the matter, I could see it being misleading because they list MS as part of their ecosystem, which would technically also apply to AMD as well.  Yet they do not list MS or any companies that partner with AMD on the AMD ecosystem side.  Thus making it appear that these companies are exclusively part of the Intel ecosystem.

Oh, Of course. It didn't even cross my mind that Intel was trying to portray Microsoft as an exclusive partner. Which is, of course, completely absurd. 

[BuckGup]

Sick, screw intel. We need competition again

[DoctorZeus]

cmonBruh. intel. cause the 7740x isnt a glued together piece of trash

[Jito463]

8 minutes 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. The product performs excellently and I have never made shots on it's performance. I just said Intel's claim is correct in that it is a "worse" way of achieving it.

*sigh*

 

Insert words, more words, words that no one bothers to read or understand, words, words, yet more words.  Obligatory summary.

 

Hey, that was much easier than what I did previously.  I should just discuss like this from now on.  Saves me so much time, instead of writing up proper responses with an actual conclusion that no one bothers to read properly.

[bcguru9384]

1 hour ago, sazrocks said:

Here is a picture with the duplicates circled:

curious? 

did you circle those using win 10?

[Marinatall_Ironside]

34 minutes ago, Hunter259 said:

Even people with degrees are idiots. I'm going into my second year for Computer Engineering and I can see similar people who can get grades but have no common sense. I also never said it had bad performance just that it is inherently inferior

Like what @Jito463 said, the extent of the latency hit is dependent on not only RAM frequency (because Infinity Fabric runs at half the frequency of the RAM, so 3200 MHz RAM = 1600 MHz Infinity Fabric frequency), but also manufacturing from GloFo. Getting this gap closed is important to make multi-die CPUs with almost zero impact very possible. Just a side note AMD (or Intel if they would take a page from AMD and come up with a proper response) is that close to achieving that, because Moore's Law is limiting how many CPU cores you can squeeze into a single chunk of silicon. Fitting more CPU cores into a single package helps with server/node builders squeeze more CPU cores into the same amount of space (or less if wanting the same number of cores that you would get from Intel).

 

I've been also trying to figure out how Intel's Omni-Path 100G interconnect works on their Xeon Phi x200 Host CPUs, but they have Omni-Path switches that are used to hook together computer nodes together. I'll admit I am very likely wrong (because as cool as supercomputer tech is, it's not in my interests unlike consumer stuff) on how Omni-Path works outside of the Xeon Phi host CPUs (because the Omni-Path implementation in the video is connecting entire computer nodes in an "island" topology, not CPUs). LTT recently did a quick tour of the Cedar supercomputer will provide some insight into how it all works.

 

Already AMD Epyc is looking to be an impressive lineup due to the massive amounts of I/O with 128 PCIe lanes, and the 8-channel memory (which is absolutely absurd, I thought 6-channel memory from Intel with their Xeon Phi x200 Knights Landing was neck-twisting, but nuuu).

 

If you want any advice for your Computer Engineering pursuit, which I support wholeheartedly, don't be that idiot streamer I told you about, and I'm not naming anyone (don't ask me).

Edited July 19, 2017 by JurunceNK

[Drak3]

25 minutes ago, Jito463 said:

That doesn't sound less technologically advanced to me.  It sounds like a technological improvement that bypasses limitations of the current method.

MCMs have been around for quite some time, so EPYC and Thread Ripper being MCMs are less advanced in that regard. The underlying tech that the dies communicate through is significantly more advanced. At the end of the day, it's a solution to a problem, and new or old, basic or advanced, the only thing that matters is if it works. And we're in agreement that AMD's latest MCM implementation definitely works.

 

I just wonder how long it takes for Intel to do the same. What interconnect they will use, or if it'll even be half way decent compared to IF (according to MageTank, Intel's QPI, the fastest interconnect prior to Skylake SP, is slower than AMD's outdated and outmoded Hypertransport 3.1), and what size dies they'll use.

20 minutes 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. The product performs excellently and I have never made shots on it's performance. I just said Intel's claim is correct in that it is a "worse" way of achieving it.

You explained a con. Not why it is worse. MCM is a tradeoff, latency (and with Infinity Fabric, it's a minor tradeoff) for core count and price (and potentially clock speed). The EPYC 7601 comes in at ~$4000, the Xeon 8176 comes in at $8719, while packing 32 cores clocked at 2.2GHz vs 28 at 2.1GHz (theoretically making up for the ~5% IPC disadvantage for closely match single core perf).

4 minutes ago, JurunceNK said:

Already AMD Epyc is looking to be an impressive lineup due to the massive amounts of I/O with 128 PCIe lanes, and the 8-channel memory (which is absolutely absurd, I thought 6-channel memory from Intel with their Xeon Phi x200 Knights Landing was neck-twisting, but nuuu).

EPYC also might need those 8 lanes to match SLSP on a per core basis in AVX2, where memory has been a bottleneck on Ryzen. I just want to know if those 4 IMCs are able to somewhat overcome any additional latency incurred by adding channels.

[Hunter259]

3 minutes ago, JurunceNK said:

-edit-

Well of course the latency hit is lessened. You sped up the interface. Never have I said it was bad. Just that in terms of tech advancement it is a "worse" way.

 

2 minutes ago, Drak3 said:

-edit-

My argument has always the technically aspect of it. It is a trade off but it is a cruder design.

[Trik'Stari]

1 hour ago, nerdslayer1 said:

they are good at eating things that are green( its a joke) 

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)

[Drak3]

3 minutes ago, Hunter259 said:

My argument has always the technically aspect of it. It is a trade off but it is a cruder design.

The technological aspect is not of cruder design. If anything, throwing as many cores on a die and calling it good is the cruder solution. It takes engineering interconnects to make multiple dies work as a single processor, and to do so as effectively as Zen does, it's amazing that Ryzen is as inexpensive as it is.

[Hunter259]

Just now, Drak3 said:

The technological aspect is not of cruder design. If anything, throwing as many cores on a die and calling it good is the cruder solution. It takes engineering interconnects to make multiple dies work as a single processor, and to do so as effectively as Zen does, it's amazing that Ryzen is as inexpensive as it is.

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. 

[DoctorZeus]

9 minutes ago, Drak3 said:

The technological aspect is not of cruder design. If anything, throwing as many cores on a die and calling it good is the cruder solution. It takes engineering interconnects to make multiple dies work as a single processor, and to do so as effectively as Zen does, it's amazing that Ryzen is as inexpensive as it is.

ryzen is cheaper because it's easier to fab lowet core count chips and glue them together than make monolithic dies. This is partly why yields are so good

[The Sloth]

7 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. 

can we agree both are good? if you take cost into account Intel is losing. 

[Drak3]

6 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. 

Except that interconnects during the C2Q days were inferior to monolithic dies in every regard, when working on that small of a scale. Whereas QPI, Omnipath, and Infinity fabric make larger scale processors viable.

 

It's like saying RAM makes fun of the Geo Metro for having a 3 cylinder that requires anti-vibe mounts (due to it firing in a 1-2 pattern), then RAM announces a quad sequential turbo charged, balanced 3 cylinder that packs twice the power and maintains 90% of the fuel efficiency as being hypocritical.

[Hunter259]

Just now, Drak3 said:

Except that interconnects during the C2Q days were inferior to monolithic dies in every regard, when working on that small of a scale. Whereas QPI, Omnipath, and Infinity fabric make larger scale processors viable.

 

It's like saying RAM makes fun of the Geo Metro for having a 3 cylinder that requires anti-vibe mounts (due to it firing in a 1-2 pattern), then RAM announces a quad sequential turbo charged, balanced 3 cylinder that packs twice the power and maintains 90% of the fuel efficiency as being hypocritical.

Yes. Interconnects have evolved and become even better but they still have their issues that a single die does not have.

 

Well yeah. Everyone can be right at the very same time.

3 minutes ago, nerdslayer1 said:

can we agree both are good? if you take cost into account Intel is losing. 

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.

[The Sloth]

Just now, 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.

1

how? AMD didn't start attacking Intel first, also Ryzen is solid engineering, FYI best engineers are ones that make something that works for a cheaper price, long as you can keep it the same quality. 

 

"Intel childish and wrong over this is silly" because they are, they over simplified it, making it seem like XEON is worlds better. 

 

Competition is needed, arguing over how "primitive" ryzen is because of multiple CSM is dumb, end of the day companies will go for the "better" product for the price, something that can scale well.