AMD Zen 8 core engineering sample runs at 3GHz

[shdowhunt60]

1 hour ago, LAwLz said:

That might be the biggest IF in the history of mankind, considering AMD's previous CPU track record.

I would not bet money on a 3GHz, 8 core Zen chip performing as well as a 5960X, and that chip will be more than 2 years old when Zen comes out.

Beyond Bulldozer, what exactly is the problem here? Pretty much everything from K5 to Thuban was good stuff, the shit show was only with Bulldozer. We don't dog Intel for incredibly bad the Pentium 4 was, do we? I'd argue that the Prescott was worse for the time than Bulldozer.

[ShiftHitTheFan]

Question, is the new Zen chip going to be a true 8 core? Or is it going to be like the fx chips where they are basically 4 cores split in two.

[Prysin]

1 hour ago, patrickjp93 said:

That's only when those few powerful players cooperate. Intel and Nvidia hate each other and would try their damn hardest to rip each other to pieces.

 

And it's the best thing for the industry. Intel releases tons of its patents every year. AMD and Nvidia don't. That's one huge reason the GPU market is so closed up. They're both incredible patent trolls. When was the last time you heard of Intel suing someone for the design of an instruction or circuit? It would be best for the market for AMD to step aside rather than just struggling along (which will continue in spite of any fan's delusions).

In the era where software isn't evolving and SISD instructions are already optimized near to their theoretical lowest possible latencies, rating by GHz is MORE accurate than it was in the old days. If software starts evolving again, then sure, it goes away.

the reason Intel isnt sueing left and right is because intel doesnt hold ALL the CPU patents. Some are held by AMD, some by IBM (a lot by IBM) and some by smaller companies that mostly make custom chips.

 

Aslong as an instruction is not a copy, it is hard for intel to sue for those too.

 

But most importantly, and most notably, is the fact that nobody is able to COMPETE with intel. They can make all the circuits and instructions they want, but i promise you that even if the circuit and instruction was not of intels own design, if someone managed to get a large enough market share to threaten intel, or enough traction to actually GET to market. Intel would bury them head first before they could even chip a dime off intels bottom line.

 

Just look at AMD, every time AMD gets close to 25% market share, Intel swings their hammer and crushes them down, one way or the other. Be it by cheating benchmarks, subsidicing OEMs, raw marketing power or whatever.

So if a Fortune 500 company like AMD (if they're even in F500 anymore) cannot threaten intel, then why would intel bother spending money on lawyers against petty students or small companies that doesnt present any harm?

 

[Castdeath97]

4 hours ago, patrickjp93 said:

Those were I3s mind you.

The i3 now runs at 3.7 GHz compared to the testing sample 2.6 GHz, still a big jump.

[Prysin]

9 minutes ago, Castdeath97 said:

The i3 now runs at 3.7 GHz compared to the testing sample 2.6 GHz, still a big jump.

i wouldnt expect AMD to manage the same jump, not for a 8-core chip.

 

Going from 3.0 to 3.5 or 3.6, maybe, with a 3.7-3.9 Turbo speed at best. But nothing more.

[Sakkura]

32 minutes ago, ShiftHitTheFan said:

Question, is the new Zen chip going to be a true 8 core? Or is it going to be like the fx chips where they are basically 4 cores split in two.

True 8 core, with an AMD version of hyperthreading on top (for 16 threads).

[patrickjp93]

19 minutes ago, Prysin said:

the reason Intel isnt sueing left and right is because intel doesnt hold ALL the CPU patents. Some are held by AMD, some by IBM (a lot by IBM) and some by smaller companies that mostly make custom chips.

 

Aslong as an instruction is not a copy, it is hard for intel to sue for those too.

 

But most importantly, and most notably, is the fact that nobody is able to COMPETE with intel. They can make all the circuits and instructions they want, but i promise you that even if the circuit and instruction was not of intels own design, if someone managed to get a large enough market share to threaten intel, or enough traction to actually GET to market. Intel would bury them head first before they could even chip a dime off intels bottom line.

 

Just look at AMD, every time AMD gets close to 25% market share, Intel swings their hammer and crushes them down, one way or the other. Be it by cheating benchmarks, subsidicing OEMs, raw marketing power or whatever.

So if a Fortune 500 company like AMD (if they're even in F500 anymore) cannot threaten intel, then why would intel bother spending money on lawyers against petty students or small companies that doesnt present any harm?

 

Oh please, Intel and IBM could both sue ARM out of existence for their vector instructions. Intel and IBM don't because they understand having the CREDIT is prestige, but it's pointless to chase down all these companies for coming to the same obvious conclusion.

 

Eh, more like AMD stumbles and faceplants again. Intel is constantly fighting IBM and Oracle in HPC, so let's not kid ourselves. Intel is never swinging a big hammer at AMD. The last time that happened was more than a decade ago, and the people responsible for it have all left Intel (Otellini and all executive officers from 2005-2007 are gone).

 

Because lawsuit victories = royalties, but they also invite Intel being labeled a blight on the industry, something far more damaging. I'm sorry but ARM blatantly violates more than 30 IBM and Intel patents on branch prediction, superscalar engines, and vector instructions alone. However, they are obvious approaches to making hardware go faster. Intel invented and patented the small loop detector with Nehalem. IBM copied it for Power 8. Oracle copied it for Sparc M7. Neither of them pays Intel any licensing fees to use the technique.

[byalexandr]

5 hours ago, Humbug said:

AMD said the flagship highest end Zen parts will have 95 watt TDP.

 

we have no idea about the TDP of this engineering sample part or where it fits in the product line up.

 

even with the massive ipc gains AMD will need to clock above 3.5Ghz on the top parts in order to compete with Intel so let's hope Zen clocks well.

Haswell like IPC and eight cores is better than a 5820K or Xeon equivalent I'd imagine.

[Prysin]

43 minutes ago, patrickjp93 said:

Oh please, Intel and IBM could both sue ARM out of existence for their vector instructions. Intel and IBM don't because they understand having the CREDIT is prestige, but it's pointless to chase down all these companies for coming to the same obvious conclusion.

 

Eh, more like AMD stumbles and faceplants again. Intel is constantly fighting IBM and Oracle in HPC, so let's not kid ourselves. Intel is never swinging a big hammer at AMD. The last time that happened was more than a decade ago, and the people responsible for it have all left Intel (Otellini and all executive officers from 2005-2007 are gone).

 

Because lawsuit victories = royalties, but they also invite Intel being labeled a blight on the industry, something far more damaging. I'm sorry but ARM blatantly violates more than 30 IBM and Intel patents on branch prediction, superscalar engines, and vector instructions alone. However, they are obvious approaches to making hardware go faster. Intel invented and patented the small loop detector with Nehalem. IBM copied it for Power 8. Oracle copied it for Sparc M7. Neither of them pays Intel any licensing fees to use the technique.

copied or derived?

That is the question.

 

And yes, AMD aren't without fault when it comes to their misery. Hopefully, their current leadership has a good enough long term memory to know what NOT to do (before Q1 2019).

[patrickjp93]

1 hour ago, Prysin said:

copied or derived?

That is the question.

 

And yes, AMD aren't without fault when it comes to their misery. Hopefully, their current leadership has a good enough long term memory to know what NOT to do (before Q1 2019).

Copied. The paradigm of vector addition, multiplication, and primitive ops is all Intel IP. Scatter/Shuffle, reduce, and more advanced ones such as in SSE and AVX are IBM and Intel. ARM has blatantly copied these exact instructions under a different name. MIPS 64 is doing the same thing. If Intel really wanted, no one in the world other than AMD could do any primitive vector instruction.

[patrickjp93]

1 hour ago, byalexandr said:

Haswell like IPC and eight cores is better than a 5820K or Xeon equivalent I'd imagine.

Only if for the subset of instructions you'll be using. If it's only an overall average, you could very well be screwed.

[deviant88]

7 hours ago, LAwLz said:

Zen can't come soon enough.

 

I am trying not to get my hopes but, but I really want them to succeed. Hopefully the release chips will be clocked higher than 3GHz. It would suck if AMD managed to boost their IPC by 40% just to have it ruined by low clocks.

You know bulldozer chips didnt suck because of the clocks, its not a gpu raw clocks dont mean that much if the ahitecture is shit. Last time i checked a fx 8000 was bottlenecking most of games and applications that didnt need raw power. Same as my old phenom II 3.8ghz even tough usage was not even close to 100% my gtx 670 was being held back in games, when i switched to an intel 4570 at 3.2ghz/3.4 turbo  the bottleneck became the gpu in all games/benchmarks, IPC > frequency.

Nothing new really, but i do expect overclocking models to go up to 4.0 on air and 4.5ghz on water, silicon lottery luckers only ofcourse.

[patrickjp93]

20 minutes ago, deviant88 said:

You know bulldozer chips didnt suck because of the clocks, its not a gpu raw clocks dont mean that much if the ahitecture is shit. Last time i checked a fx 8000 was bottlenecking most of games and applications that didnt need raw power. Same as my old phenom II 3.8ghz even tough usage was not even close to 100% my gtx 670 was being held back in games, when i switched to an intel 4570 at 3.2ghz/3.4 turbo  the bottleneck became the gpu in all games/benchmarks, IPC > frequency.

Nothing new really, but i do expect overclocking models to go up to 4.0 on air and 4.5ghz on water, silicon lottery luckers only ofcourse.

Hehe, not quite. And even the following is crude . Performance ~ IPC * Clockspeed.

[EChondo]

I'm really hoping Zen is at least comparable to Intel's next release. If we can get some good competition going again then it'll help everyone out.

 

I'm on a Z68 chipset and have been itching for an upgrade even though I don't really need it. Even if Zen isn't amazing, if we can get a price war going then I'm definitely upgrading sooner than previously anticipated.

[Ren]

7 hours ago, valdyrgramr said:

Then came the 9xxx series, for overclockers, which was kinda a mistake.  The 8320=9370 & 8350=9590 just factory OCed.  On one hand you're getting more speed, but on the other hand they did kinda wasted money on that.  The 9xxx chips were nice for their target, and ya I did buy one, but at the same time they lost more money on that than they made for "higher clock speeds".

Pretty sure the 9370 and 9590 were sort of equivalent to an Intel i7-5930k in terms of their place on the board.

 

That and the 9370/9590 are binned chips just like the 860k.

[Prysin]

4 minutes ago, Ren said:

Pretty sure the 9370 and 9590 were sort of equivalent to an Intel i7-5930k in terms of their place on the board.

 

That and the 9370/9590 are binned chips just like the 860k.

Fx9 isnt even remotely close to a 5930k even in productivity software that can use all the cores/threads 

[Ren]

Forgot to post that I expect:

 

 

Low voltage and standard parts for desktop.

 

8C/T, 95W, 3.7-4.2ghz turbo. OCing to 4.6+. Pricing at $339.99

8C/T, 65W, 2.9-3.3ghz turbo. Why would you OC? Pricing at $269.99

4-core chips that run at same wattage but higher default clocks, lower turbo. OC probably far more mild. $129.99-$179.99

 

No idea what to expect below $120-$150.

 

I am almost 100% sure AMD plans to try and sell as many low-end and top-end + low voltage parts as possible. The middle-of-the-road shit, i.e. ~i5 territory, they'd probably leave alone. I also expect them to go deep on OEM sales in return for what they will fail to sell to consumers initially. I hope they plan on buying some advertisement on television and such for this event as well..

[Ren]

13 minutes ago, Prysin said:

Fx9 isnt even remotely close to a 5930k even in productivity software that can use all the cores/threads 

"Then came the 9xxx series, for overclockers, which was kinda a mistake.  The 8320=9370 & 8350=9590 just factory OCed."

 

'Pretty sure the 9370 and 9590 were sort of equivalent to an Intel i7-5930k in terms of their place on the board', as in their SKU. They are not simply overclocked Bulldozer+ FX-8000 processors. They were sold as prosumer+ chips, like the 5930k and such. And more than just "on average" they have higher IPC than FX-8000/6000/4000. The only chip I know of that can beat them in IPC from AMD are the 860k/7860k+ (most recent 4-core FM2+ Excavators).

[shdowhunt60]

17 minutes ago, Ren said:

Forgot to post that I expect:

 

 

Low voltage and standard parts for desktop.

 

8C/T, 95W, 3.7-4.2ghz turbo. OCing to 4.6+. Pricing at $339.99

8C/T, 65W, 2.9-3.3ghz turbo. Why would you OC? Pricing at $269.99

4-core chips that run at same wattage but higher default clocks, lower turbo. OC probably far more mild. $129.99-$179.99

 

No idea what to expect below $120-$150.

 

I am almost 100% sure AMD plans to try and sell as many low-end and top-end + low voltage parts as possible. The middle-of-the-road shit, i.e. ~i5 territory, they'd probably leave alone. I also expect them to go deep on OEM sales in return for what they will fail to sell to consumers initially. I hope they plan on buying some advertisement on television and such for this event as well..

Those frequencies are far too optimistic for 8C/16T, especially on Samsung's/GloFo's 14nm LPP node. At 95 watts, I'd consider 2.9-3.3ghz a more realistic number. Maybe 3.2-3.6ghz at the most. Everything I've heard from Zen sounds like it's basically going to be Haswell-like in terms of efficiency.

[SparkyRain]

16 minutes ago, shdowhunt60 said:

Everything I've heard from Zen sounds like it's basically going to be Haswell-like in terms of efficiency.

Haswell-like? So already outdated when it's released. It's better be cheap or *poof* amd.

[shdowhunt60]

Just now, SparkyRain said:

Haswell-like? So already outdated when it's released. It's better be cheap or *poof* amd.

Since when was Haswell outdated? I mean, shit, still plenty of people out there doing just fine on Phenom II's and Bulldozer, so I definitely fail to see how Haswell is outdated.

[SparkyRain]

1 minute ago, shdowhunt60 said:

Since when was Haswell outdated? I mean, shit, still plenty of people out there doing just fine on Phenom II's and Bulldozer, so I definitely fail to see how Haswell is outdated.

Zen is gonna be outdated at release not haswell. You have trouble with reading? Benchmarks will be held against skylake, we can can already predict amd will be behind - as always. Not excited for this line up. They better focus on gpu's instead crippling behind intel for taking too long with their releases. 

[shdowhunt60]

Just now, SparkyRain said:

Zen is gonna be outdated at release not haswell. You have trouble with reading? Benchmarks will be held against skylake, we can can already predict amd will be behind - as always. Not excited for this line up. They better focus on gpu's instead crippling behind intel for taking too long with their releases. 

Okay? And? Skylake might be %5-10 faster. I think Zen will do just fine.

 

And we're seriously going to go there with the whole "AMD is taking too long for Zen" again? They put just about every other major CPU project on the backburner to get Zen out for the commercial market. The process required to even make the damned chips in the first place JUST started ramping up. They're doing everything they can to get Zen out as early as possible.

[CyberMist]

On 4/3/2016 at 5:56 PM, SamStrecker said:

This title is like AMD selling their chips. GHZ IS NOT EVERYTHING! There are a lot of other factors

True. Ghz, cache, core and threads count is not final factor.. 

If u don't have  software developed to utilize fully of hardware capability u will suck. 

Even u have fastest car in world. without proper road u can't use full its capability and your speed is limited.. 

 Same goes with pc hardware..  

And this rule goes for pc overall not only for amd cpu. 

[BuckGup]

10 minutes ago, CyberMist said:

True. Ghz, cache, core and threads count is not final factor.. 

If u don't have  software developed to utilize fully of hardware capability u will suck. 

Even u have fastest card in world. without proper road u can't use full its capability and your speed is limited.. 

 Same goes with pc hardware..  

And this rule goes for pc overall not only for amd cpu. 

AMD just uses this to sell outdated non efficient high 1.5 core count at high clock speeds