AMD reports Threadripper 2990 is upto 50% faster than the Intel 7980

[TVwazhere]

[Taf the Ghost]

1 hour ago, Master Disaster said:

So does anyone have any idea when NDA lifts?

 

Would be nice to find out if AMD (or maybe even WCCF are lying) though for whatever reason I actually believe these numbers are not a million miles away.

August 13th, I believe, which is launch day.

[pas008]

3 hours ago, MMKing said:

You're not addressing the real point here. As i stated later on in the post you quoted earlier.

 

''Even if it does not retain the performance throne, the price is insane and i don't think AMD is selling it at low margins.''

 

The performance of the 2990WX is not that important, it doesn't have to beat Intel's flagship CPU. The point is that Intel is being forced to take their golden goose behind the barn and slaughter it. With a 32 core CPU at less than 2000 dollars, Intel can't keep selling 28 core CPUs at 10 000 dollars. Who knows what is next for AMDs Epyc lineup?

no my point was your comparison with server grade

but said nothing on price/etc I know tr32 is a steal on price area

and like leadeater posted below

2 hours ago, leadeater said:

Yep which is exactly why that 8180 OC is really good to look at, 33% OC is extremely good especially for BLK only OC. It's not like it's a AIO or anything either to get that.

 

 

Very sexy .

 

The only actual tweaks that matter is the clocks, disabling ECC does nothing heat/power/performance wise. Actual higher stock multiplier/clocks and unlock multiplier is obviously better than BLK OC but if you're hitting thermal or power limits it makes no difference between the two, it's just way harder to get stable clocks using BLK because it throws off ram and PCIe clocks as well.

 

Edit:

Plus I keep forgetting there is an LTT CES video covering one of these 28 core samples on custom loop water with the unlocked multiplier and a CB run shown.

6109 on water OC, so 5000 stock does seem reasonable to me.

exactly that was what I remembered and why i made the point of the comparison

[MMKing]

@pas008

16 phase

2 * 24 pin

4 * 8 pin

2 * 6 pin

4 fans cooling the power delivery

Custom water loop

 

Yes, the system you linked in the video seems very representative of what the final product will look like. Further more, this is not a store bought CPU. This is a CPU sent by the Intel marketing department to Asus for the explicit purpose of showing off at the Computex 2018 floor. And i'm not seeing any long term stress tests in the video either, which i doubt the system could handle. 

 

Meanwhile over at AMD. They are sending the press air cooler, closed loop water cooler, 3200MHZ ram and i grant you a very nice motherboard. AMD is comfortable with subjecting their product to media scrutiny.

[pas008]

11 minutes ago, MMKing said:

@pas008

16 phase

2 * 24 pin

4 * 8 pin

2 * 6 pin

4 fans cooling the power delivery

Custom water loop

 

Yes, the system you linked in the video seems very representative of what the final product will look like. Further more, this is not a store bought CPU. This is a CPU sent by the Intel marketing department to Asus for the explicit purpose of showing off at the Computex 2018 floor. And i'm not seeing any long term stress tests in the video either, which i doubt the system could handle. 

 

Meanwhile over at AMD. They are sending the press air cooler, closed loop water cooler, 3200MHZ ram and i grant you a very nice motherboard. AMD is comfortable with subjecting their product to media scrutiny.

lol my point was your comparison

and you continue on

 

and you think intels isnt going have tweaked that 28core to get better? but still isnt the point

point was your comparison plain and simple

 

should we start comparing mainstream to budget apus for single/multi core performance too?

[Citadelen]

10 hours ago, D13H4RD2L1V3 said:

Still pretty early so we'll have to wait.

 

If this pans out though, Intel really needs to respond. AMD is not playing around anymore. Maybe their SVP, which just happens to be Jim Keller might have a plan. 

He's just CVP of SoC related stuff, so he won't be developing any fancy new architectures anytime soon.

[Stefan Payne]

9 hours ago, Ryan_Vickers said:

Doesn't Intel have a 28 core "enthusiast" chip now (as in not server)?  The one from the chiller fiasco?  If so, AMD should really be comparing to that, otherwise we should be taking the 7980 vs 1950X.  If not, where is it and what was even the point of that demo?

Its not really known what Socket that one will use. The Rumormill is talking about the big 3k6 Pin Socket coming for that.

So that would mean that Intel has to deal with two HEDT Sockets, while AMD just has to deal with the one Socket...

 

 

9 hours ago, Ryan_Vickers said:

I think it's time for something new.  I don't know if sandy bridge was technically a big change vs what came before or not but everyone thinks of it that way and for good reason.  Just like Ryzen vs FX, etc.  I feel like it's just been small increments since Sandy on the Intel side and yeah it's time for another leap.

Sandy was just the usual rework of the original design. Kinda like K7 -> K8.

You can argue that both use the same cores while the K8 has a new Northbridge attached to it and some other additions. And that was one of the Problems at the time of AMD, that they didn't go for a really new architecture after K8 and instead a more or less improved architecture (with a shitty NB) and they also reused the garbage Northbridge design from the Phenom with the first bulldozers...

 

Well, at least now with Zen they have a good core Design and a workable/OKish Uncore/NB Design wich was AMD's biggest Problem.

 

PS: Intel had 2 clock Domains for Bloomfield IIRC and with the switch to Sandy Bridge they reworked it so that the L3 Cache is at the same clock as the Core.

 

8 hours ago, Ryan_Vickers said:

Ah, alright then.  I thought it was.  Well good job then AMD, they can quite comfortably claim the thrown on this one for now.

Yeah, and they will have <14nm before Intel ha

8 hours ago, Quadriplegic said:

Woah, chill out AMD.

Wouldn't want this to backfire at you, would you?

What are you talking about?!

Leaning back and doing nothing never works out.

They have to throw anything they have at the other guys right now to regain Market share.

 

And even if they have a far superior product, there are still people who won't buy the better part for less money...

8 hours ago, Humbug said:

It's not really surprising that Intel cannot respond fast.

Before Ryzen released a lot of people were viewing Intel as some kind of God company. People would post things on these forums like

Yeah, especially because people don't want to admit that there is no special sauce, that there can not be any special sauce as physics are the same for everyone and most of the other stuff is already known.

 

And the other side doesn't really know what the others are doing.

They can only react to predictions, if they are correct and not off by a grand canyon...

 

And often it takes years to react. Just look at Pentium 4 vs. Athlon 64!

 

8 hours ago, Humbug said:

Intel will come back with a bang obviously but not overnight. We have to give them time to sort out both their manufacturing and architectural issues. Meanwhile they easily have the majority market share in all segments of x86 computing and they have much stronger channel partnerships than AMD, they will rely on this to get them through the period where AMD leads in server performance.

We will have to wait and see how that will play out...

And how long they can survive with their old lineup...

And especially how AMD is able to keep up.

 

In the K8 area AMD was lazy and didn't improve their Architecture enough that was what allowed Intel to keep up.

 

Right now AMD is in overdrive and improves their Architecture with each iteration drastically, Zen isn't even a year old and already got some noticable improvements in clockrate and performance!

 

Back in 2004 that wasn't the case with K8. Although they released a Dual Channel Version/Socket, the Die was the same, there isn't a difference between a Socket 754 and 939 Chip, besides the Package.

Although the old socket was needed for the development of the 939 Boards because the MoBo Makers didn't have any experience designing Boards where the Memory Controller is in the CPU...

 

8 hours ago, GoldenLag said:

Would be interesting if intel did a BIGlittle core design. Big cores for singlethreaded and smaller ones to take on multithreaded.

That's not how it was done and what makes sense.

The BIGlittle concept is for saving energy, to increase the efficiency and the BIG Cores are totally clock gated -> switched off under lower load Conditions.

Its not something for inceasing Performance.

 

And  many little cores we already have, its called a GPU. Or XeonPHI...

5 hours ago, MMKing said:

Even if it does not retain the performance throne, the price is insane and i don't think AMD is selling it at low margins.

Exactly...

And it shows the world what quasi Monopolys do for a market:
You pay 10 fold for a chip...

And you don't have any choices...

4 hours ago, leadeater said:

That used to be quite true until Intel started lowering the base and boost clocks on the same model non K SKUs, jerks.

They have to keep the Chips inside their Definition of TDP, you know

And as Intel can violate their own TDP without any problems by 100% or more...

3 hours ago, mynameisjuan said:

Honestly what is the point of having a CPU unboxing video? 

Making money with easy to make content that is done in like 5minutes?!

 

What other reason do you need for that?

[spartaman64]

8 hours ago, GoldenLag said:

Thats the point of cost cuts. They have the money to win a monolithic war of attrition. Problem for intel is that the war isnt monolithic.

 

An 8 core 1 CCX CPU from AMD costs 300$ in the consumer market. Thats roughly 1250$ for an 32 core at consumer margins. They can cut it lower. 

 

8 hours ago, Humbug said:

 

Indeed it's not sustainable.

It's just an interim thing that they can do for the next two years until they have new architectural and process developments. They can take that hit financially for now (because they can afford it) if they want to completely block AMD from getting a foothold in the server market. I am not saying they will do this, I am just saying it's one option if they want to shut out AMD from the lucrative server market.

 

I agree the more likely scenario is that they will continue to sell their server CPUs at profit and maximize cash-flow, and resign themselves to the fact that AMD is going to get 5-10% CPU server market share.

 

The latter scenario is what AMD wants, because just these few percentage points of market share means loads of money for AMD. And it means they have got into the big data centers around the world and built the right relationships with server guys like Dell, HP etc. Breaking into the server market again is not easy, so this first hurdle is the biggest one for AMD.

another problem with this is that selling products at a loss to try to undercut a rival company might land you into antitrust territory

[Quadriplegic]

9 minutes ago, Stefan Payne said:

 

What are you talking about?!

Leaning back and doing nothing never works out.

They have to throw anything they have at the other guys right now to regain Market share.

 

 

I mean their quite aggressive marketing. 

I think they should be extremely careful with their choice of words, because they can't afford to screw up

[Stefan Payne]

6 minutes ago, Quadriplegic said:

I mean their quite aggressive marketing. 

I think they should be extremely careful with their choice of words, because they can't afford to screw up

No, they can't be careful, they have to break the mental bias of many people.

They have to be aggressive.

 

They don't have any other choice!

 

Just look at this Forum, how many People here want to build new things and only think about one manufacturer and don't even consider the other one because they don't know or they don't want them because someone somewhere told them that their hose will explode if they get it.

 

The only way to break this Mental Bias is very aggressive Marketing that really shows that they are the better choice.

[S w a t s o n]

Spoiler

 

[D13H4RD]

1 hour ago, Swatson said:

  Hide contents

 

Liquid nitrogen beats a fridge 

 

Kek 

[S w a t s o n]

1 minute ago, D13H4RD2L1V3 said:

Liquid nitrogen beats a fridge 

 

Kek 

*freezer

 

chiller was running at like -10c, closer to LN2 than water (not literally but in terms of getting clocks high)

[AnonymousGuy]

26 minutes ago, Swatson said:

*freezer

 

chiller was running at like -10c, closer to LN2 than water (not literally but in terms of getting clocks high)

The transistor performance scales linearly at "normal" temperatures like -10C where you're really only affecting the temperature and not the transistor's performance.  You need LN2 to affect transistor performance.  The same 28C Intel chip would probably go >6Ghz on LN2.

 

TLDR: LN2 is a huge advantage compared to chilled water.

[S w a t s o n]

4 minutes ago, AnonymousGuy said:

The transistor performance scales linearly at "normal" temperatures like -10C where you're really only affecting the temperature and not the transistor's performance.  You need LN2 to affect transistor performance.  The same 28C Intel chip would probably go >6Ghz on LN2.

 

TLDR: LN2 is a huge advantage compared to chilled water.

Tl;DR you provided no evidence, and I would LOVE to see that 28core hit 6ghz without blowing up that VRM, good luck

[S w a t s o n]

Not to mention having to find a platinum tier sample that can even do 6ghz on all 28 cores

[Master Disaster]

Didn't Linus do a phase change video once or am I imagining that?

[mynameisjuan]

Just now, Master Disaster said:

Didn't Linus do a phase change video once or am I imagining that?

He did on that sexy all white build

[S w a t s o n]

1 minute ago, Master Disaster said:

Didn't Linus do a phase change video once or am I imagining that?

He did

[leadeater]

2 hours ago, Swatson said:

  Reveal hidden contents

 

Is Tom's Hardware even allowed to post that ?!?!?

[pas008]

43 minutes ago, D13H4RD2L1V3 said:

Liquid nitrogen beats a fridge 

 

Kek 

didnt the 8180 get 6109 on non chilled water in a linus video

and get 5010 on bclk overclock on water

bet after some tweaks itll be like 5.5k stock score

 

[S w a t s o n]

4 minutes ago, leadeater said:

Is Tom's Hardware even allowed to post that ?!?!?

Hmm, depends on what the embargo covered. They have pictures from the event itself so maybe they got special rights

 

https://www.tomshardware.com/news/amd-threadripper_2-vs-intel-core_x,37550.html

[AnonymousGuy]

40 minutes ago, Swatson said:

Tl;DR you provided no evidence, and I would LOVE to see that 28core hit 6ghz without blowing up that VRM, good luck

On LN2 power consumption goes down.  Like I said, transistor properties change at the super low temperatures.

 

https://www.tomshardware.com/reviews/core-i9-7900x-overclock-ln2,5618-7.html

 

And that VRM had something like 32 phases on the Intel board...probably meant for 4 socket systems.   It's not a problem:

 

[S w a t s o n]

6 minutes ago, AnonymousGuy said:

On LN2 power consumption goes down.  Like I said, transistor properties change at the super low temperatures.

 

https://www.tomshardware.com/reviews/core-i9-7900x-overclock-ln2,5618-7.html

 

And that VRM had something like 32 phases on the Intel board...probably meant for 4 socket systems.   It's not a problem:

 

 

Uh just because it had 32 phases doesnt mean it's gonna be fine. Quad socket systems dont usually hit 1kw of power consumption (250w per chip), that single 28 core at 5ghz was pulling over 1000 watts

 

Edit: https://ark.intel.com/products/120496/Intel-Xeon-Platinum-8180-Processor-38_5M-Cache-2_50-GHz

That chip is rated at 205w, so at best ~800 watts in a quad socket lmfao

[WereCat]

5 hours ago, TVwazhere said:

The core feature of AMD CPUs is the amount of cores.