AMD Zen processors to have "double the performance of the FX 8350", performance pictures posted

[Tomsen]

38 minutes ago, TheDefinitionOfInsanity said:

Where does it say 2x? The bar being 2 times higher doesn't mean shit if there's no scale. Orochi can be 1x, which makes Summit 1.5x just like on the left chart.

It is all the way to the left, at the kaveri - carrizo - bristol ridge comparison, as it is one slide.

I think they could end in legal trouble if they tried to deceive partners, shareholders and the common people with such marketing slide.

[LAwLz]

7 hours ago, patrickjp93 said:

No, Cinebench is all integer and int-vector.

 

FX only had 1 vector unit per module as well.

Got a source on that? I can't find any definitive answers from Googleing around. If it is very float heavy then that could explain the huge performance leap these slides are talking about.

 

 

2x 8350 sounds too good to be true, and I am pretty set on getting the i7-6800K when it is released next month (hopefully).

[porina]

29 minutes ago, Tomsen said:

It is all the way to the left, at the kaveri - carrizo - bristol ridge comparison, as it is one slide.

I think they could end in legal trouble if they tried to deceive partners, shareholders and the common people with such marketing slide.

Same slide, yes. Same scale, we can't say that. Note the horizontal grid lines don't line up. We see Orochi at 2 units height, and Summit at 4 units height, but there is no explicit zero stated, or any other value. If it was double, why not put 2x over Summit in a similar style to the left? So, they're not strictly lying, because they're not giving you the information.

9 minutes ago, LAwLz said:

Got a source on that? I can't find any definitive answers from Googleing around. If it is very float heavy then that could explain the huge performance leap these slides are talking about.

I'm also curious about the instruction mix of Cinebench. In recent looking at R15, I didn't see any major steps between the last few Intel architectures and Sandy Bridge. I would have hoped for more scaling if it was more float heavy, unless it is simply not optimised for newer instructions. How does existing generation AMD compare against Intel in Cinebench? If AMD are at all competitive, then Cinebench probably isn't doing that much float!

[LAwLz]

4 minutes ago, porina said:

I'm also curious about the instruction mix of Cinebench. In recent looking at R15, I didn't see any major steps between the last few Intel architectures and Sandy Bridge. I would have hoped for more scaling if it was more float heavy, unless it is simply not optimised for newer instructions. How does existing generation AMD compare against Intel in Cinebench? If AMD are at all competitive, then Cinebench probably isn't doing that much float!

AMD gets destroyed in Cinebench.

8350 - 640

i7-6700K - 919

 

The i7 got half the "cores" and still ends up being ~44% faster.

[RagnarokDel]

9 hours ago, Z3DT said:

[...]

 

AMD seems to be making quite the comeback. I'm interested to see whether the Zen chips will make buying an AMD chip for a higher-end machine will be worth it again.

 

Source

And Intel wants to slow cpus in favor of efficiency... Just try Intel, I dare you.

[porina]

11 minutes ago, LAwLz said:

AMD gets destroyed in Cinebench.

8350 - 640

i7-6700K - 919

 

The i7 got half the "cores" and still ends up being ~44% faster.

Actually, the scores at that link does suggest Cinebench doesn't use FPU. The multi-thread score for the 8350 is about 6.7x that of single-thread, assuming clocks don't change turbo style. If it were FPU intensive, you'd expect it to be capped at 4x or less. Actually, the Intel results are also suggestive that Cinebench doesn't use FPU, since in my experience with other software, HT is unable to extract more performance in intensive FPU applications whereas Cinebench does gain a good 30% or so from HT.

 

There is another possibility that could fit the observations, that Cinebench does use the FPU but in a highly inefficient way. I would hope this isn't the case.

[Colonel_Gerdauf]

20 minutes ago, RagnarokDel said:

And Intel wants to slow cpus in favor of efficiency... Just try Intel, I dare you.

I really have to wonder how this misinterpretation stuck around for so many people. Intel does NOT want to slow CPUs for efficiency; rather, the rate of increase (delta) of IPC improvements will decrease.

 

Graphically speaking, there is a difference between a slow increase and a decrease at any rate.

[Energycore]

This is no surprise. The i7 5960X is on Haswell architecture and it's 8/16 cores.

 

So a 8/16 core Zen chip, if we're to believe the 40% IPC improvement, will rival the 5960X.

 

Also, the 2x performance number is probably taken from both the IPC gains and the performance increase from Simultaneous multithreading (not to mention that it makes sense for an 8-core Zen chip to have 8 FPUs instead of the 4 of 8350).

[Godlygamer23]

9 hours ago, GtaSeriesFan said:

"big performance leap expected" so not yet confirmed right?

High performance was expected with Bulldozer. That didn't turn out too well, did it?

[Kamina]

Going to keep my expectations low.

[Tomsen]

11 hours ago, porina said:

Same slide, yes. Same scale, we can't say that. Note the horizontal grid lines don't line up. We see Orochi at 2 units height, and Summit at 4 units height, but there is no explicit zero stated, or any other value. If it was double, why not put 2x over Summit in a similar style to the left? So, they're not strictly lying, because they're not giving you the information.

We can't, but they would most likely be in legal troubles if not, as that is deceptive otherwise. 

Because this is a marketing slide, marketing people want clean and sexy slides. 

You are right they wouldn't strictly be lying, but they would strictly be deceiving.

[Citadelen]

12 hours ago, patrickjp93 said:

Nope, because it's very stable and going nowhere for now. I can repurchase closer to the conference if needed, or start shorting.

It recently closed at $4.20...

Going no where. Look at that flat line.

[porina]

1 hour ago, Tomsen said:

We can't, but they would most likely be in legal troubles if not, as that is deceptive otherwise. 

Because this is a marketing slide, marketing people want clean and sexy slides. 

You are right they wouldn't strictly be lying, but they would strictly be deceiving.

Marketing includes the dark art of presenting info in the most positive was possible while not technically lying. That does include what a reasonable person would consider intentionally deceptive, but with enough vagueness to not be outright wrong. Like adverts for battery life, up to 8 hours. Well, zero is up to 8 hours, so could still be right. Or a big "Free Something", then in tiny small print "only if you buy this other expensive rubbish". 

[Tomsen]

44 minutes ago, porina said:

Marketing includes the dark art of presenting info in the most positive was possible while not technically lying. That does include what a reasonable person would consider intentionally deceptive, but with enough vagueness to not be outright wrong. Like adverts for battery life, up to 8 hours. Well, zero is up to 8 hours, so could still be right. Or a big "Free Something", then in tiny small print "only if you buy this other expensive rubbish". 

You can say many things without technical lying which will still hold you in legal troubles. Holding in information and presenting information in a very deceptive manner (as this slide would be) can bring legal consequences.

The big difference with this and the examples you provide, is in the examples they both clarify themselves with "up to" and the tiny print. And I don't see any tiny print on AMDs slide.

[Thony]

The day AMD could finally compete/beat Intel's i5 might be around the corner, fingers crossed !

[Thony]

14 hours ago, LAwLz said:

Got a source on that? I can't find any definitive answers from Googleing around. If it is very float heavy then that could explain the huge performance leap these slides are talking about.

 

 

2x 8350 sounds too good to be true, and I am pretty set on getting the i7-6800K when it is released next month (hopefully).

It's super higly unlikely for AMD to even get close to X99 unfortunately.

 

 

[Tomsen]

14 hours ago, LAwLz said:

2x 8350 sounds too good to be true, and I am pretty set on getting the i7-6800K when it is released next month (hopefully).

First of: It is a bulldozer die, so it would be the FX 8150 not the FX 8350.

Second: 2x 8150 would that really be unbelievable? Dedicated cores (not suffering from MT scaling as CMT-designs did with a performance penalty on ~20%), however the clockspeed isn't specified afaik (so it could very well run at the same).

[Sauron]

Good, hopefully the price is much lower than the 5960x

[LAwLz]

7 hours ago, Tomsen said:

First of: It is a bulldozer die, so it would be the FX 8150 not the FX 8350.

Second: 2x 8150 would that really be unbelievable? Dedicated cores (not suffering from MT scaling as CMT-designs did with a performance penalty on ~20%), however the clockspeed isn't specified afaik (so it could very well run at the same).

Double the performance of an 8150 would make it more than 20% faster than the 6700K. I do find that hard to believe considering AMD's track record. 

 

But I did look up the Broadwell-e benchmarks again and the six core version is getting a Cinebench score of 1311. I guess it isn't unreasonable that AMD could make an 8 core that has 80% of the performance of Intel's 6 cores.

 

Even though I am set on getting the 6800K I am looking forward to Zen. Would be great if AMD could compete with Intel again. It sucks having to go LGA 2011-3 to get a decent upgrade these days. I hope Intel get forced to give us 6 cores on their mainstream platform. 

[Castdeath97]

7 minutes ago, LAwLz said:

Double the performance of an 8150 would make it more than 20% faster than the 6700K. I do find that hard to believe considering AMD's track record. 

 

But I did look up the Broadwell-e benchmarks again and the six core version is getting a Cinebench score of 1311. I guess it isn't unreasonable that AMD could make an 8 core that has 80% of the performance of Intel's 6 cores.

 

Even though I am set on getting the 6800K I am looking forward to Zen. Would be great if AMD could compete with Intel again. It sucks having to go LG 2011-3 to get a decent upgrade these days. I hope Intel get forced to give us 6 cores on their mainstream platform. 

If not having a 6 cores sucks you have got to think about how bad it's to not have any unlocked CPU below 220 dollars from intel (with the exception of the G3258 which still requires Z97 after the microcode update if I recall).

[RagnarokDel]

20 hours ago, Colonel_Gerdauf said:

I really have to wonder how this misinterpretation stuck around for so many people. Intel does NOT want to slow CPUs for efficiency; rather, the rate of increase (delta) of IPC improvements will decrease.

 

Graphically speaking, there is a difference between a slow increase and a decrease at any rate.

10% is already slow.

[Prysin]

6 hours ago, LAwLz said:

Double the performance of an 8150 would make it more than 20% faster than the 6700K. I do find that hard to believe considering AMD's track record. 

 

But I did look up the Broadwell-e benchmarks again and the six core version is getting a Cinebench score of 1311. I guess it isn't unreasonable that AMD could make an 8 core that has 80% of the performance of Intel's 6 cores.

 

Even though I am set on getting the 6800K I am looking forward to Zen. Would be great if AMD could compete with Intel again. It sucks having to go LG 2011-3 to get a decent upgrade these days. I hope Intel get forced to give us 6 cores on their mainstream platform. 

like they sort of did with Nehalem/Westmere?

[LAwLz]

14 minutes ago, Prysin said:

like they sort of did with Nehalem/Westmere?

What did they do with Nehalem/Westmere?

The cheapest 6 core from that generation was the i7-970, and that was like 600 dollars. Not what I'd call "mainstream".

[Prysin]

17 minutes ago, LAwLz said:

What did they do with Nehalem/Westmere?

The cheapest 6 core from that generation was the i7-970, and that was like 600 dollars. Not what I'd call "mainstream".

meh, 600 dollas are cheap.

[LAwLz]

1 minute ago, Prysin said:

meh, 600 dollas are cheap.

If you say so, Mr Millionaire.