AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

[OrangeKhrush]

@inf64

I see Juan has been bending the results a little.

I will wait for reviews, but if that score is legit then my ancient

Zen IPC ~ Sandy Bridge IPC

has to be finally written like

RyZen IPC = 1.09 Sandy Bridge IPC

For the sake of comparison

Skylake IPC = 1.22 Sandy Bridge IPC

I don't agree with this graph at all, all those chips got a free point or two which doesn't agree with Cinebench at all.

2600K (3.4/3.8) 135 * 3 /3.8 = 106
3770K (3.5/3.9) 143 * 3 /3.9 = 110
4770K (3.5/3.9) 156 * 3/3.9 = 120
5775C (3.3/3.7) 157 * 3/3.7 = 127
6700K (4/4.2) 181 * 3/4.2 = 129
7700K (4/4.5) 193 * 3/4.5 = 128.66 (round off 129)
Zen (3.4) 134 * 3/3.4 = 118

So in reality

Zen vs Sandybridge = 1.12

12% higher is a huge miss on a prediction, that is well over margin of error. No Juan 12% is not Sandy = Zen bro, try another one.

 

[Flash831]

My expectations of Intel and the market:

As Zen, Zeppelin and Raven Ridge gets released, Intel will see their revenue drop significantly. Even if they release some new chips just to fight Zen. The last couple of years has been some golden years for Intel, and now they are over. Time to fight for your survival!

As the revenue falls, Intel will have a problem filling their fabs and supporting their R&D of new fabs.

As a result, dividend and share buybacks will be stoped or at least significantly reduced. Brian Krzanich gets replaced as CEO.

A new CEO takes place (someone like Rory Read). A couple years later Intel releases a new arcitecture and claims back lost ground.

 

[Crumpet]

Unfortunately with the army of people purchasing Intel 7700k Spaceheater™ their bottom line won't be effected that much. The average person is still going to buy Intel, even if AMD offers the superior product, which is why it's up to us to explain to people that they might be able to get better bang for their buck with Team Red, assuming that from everything we've seen they are releasing a quality and relatively competitive product.

(seriously, who buys a brand new high end processor less than 2 weeks away from a competitor launch date?)

 

[OrangeKhrush]

Would they sell Intel CPU's with free deliding coupons?

 

[IntelUser2000]

Long term its its a bit difficult. Intel imo needs to differentiate their cpu core in different arch for consumer and heavy computing prof.
Atom is nice on serverside for some things but vs a low cost 7nm small variant zen its hopeless and will have to fight that and samsung/qcom trying to enter that market.

Atom has issues but this I would say this isn't one of them. While the performance is lower, so is the price.
Die-size wise, the 14nm Airmont core is less than quarter of the size of either Zen or Skylake.

 

[IntelUser2000]

Unfortunately with the army of people purchasing Intel 7700k Spaceheater™ their bottom line won't be effected that much. The average person is still going to buy Intel, even if AMD offers the superior product,

Average folks aren't that ignorant. Maybe a bit "slow". What I mean is the market eventually gets it. It'll take a few years of building trust. Most folks are being practical by not focusing on details of the CPU they buy in their computers, but on their lives instead. Sorry to be cynical considering the interests of most of us but it's true. There are more important things in life than buying a CPU because you weren't as knowledgeable and it isn't a big deal.

The original Athlon launched in 1999. Which I believe is a parallel to Ryzen. They won't take the absolute dominance. But what they'll do is prove most people wrong. It's the belief that AMD can't make a competitive core again. 4 years later in 2003, the Athlon 64 launched. Now that chip became the dominant one. 2 more years in 2005, AMD took over 50% retail marketshare. Think a competitive CPU from a small brand can't take majority marketshare? Think again. If Intel did not have Core uarch as a backup and they need to make a new one, Intel might have been the one in financial trouble now.

 

[GroundZero7]

As Zen, Zeppelin and Raven Ridge gets released, Intel will see their revenue drop significantly. Even if they release some new chips just to fight Zen. The last couple of years has been some golden years for Intel, and now they are over. Time to fight for your survival!

Especially since AMD, ARM, Imagination, Samsung, and IBM, maybe others have all put their heads together to take out Intel. That's a lot of patent power to throw around

 

[Atari2600]

I don't agree with this graph at all, all those chips got a free point or two which doesn't agree with Cinebench at all.

2600K (3.4/3.8) 135 * 3 /3.8 = 106
3770K (3.5/3.9) 143 * 3 /3.9 = 110
4770K (3.5/3.9) 156 * 3/3.9 = 120
5775C (3.3/3.7) 157 * 3/3.7 = 127
6700K (4/4.2) 181 * 3/4.2 = 129
7700K (4/4.5) 193 * 3/4.5 = 128.66 (round off 129)
Zen (3.4) 134 * 3/3.4 = 118

Where are you getting the numbers above from?

 

[OrangeKhrush]

Where are you getting the numbers above from?

I took the database scores off Anandtech then simple math:

2600K = 135 single thread max turbo. 135 time by 3 divided by max turbo 3.8 = 106

The numbers are correct, did them with my 5960X from 2.8 all the way to 4.5 and I will never go through that excercise again.

 

[sirmo]

What a ride this has been. Zen+ is going to be fun too. Having CPU competition back is awesome.

 

[OrangeKhrush]

Zen+ will be less dramatic though

 

[Atari2600]

What a ride this has been. Zen+ is going to be fun too. Having CPU competition back is awesome.

I'm actually quite interested to see if AMD can make any headway now with APUs in HPC as the fusion vision from circa 2007 is really starting to come together.

Now with:
(1) Competitive x86 core
(2) Shared memory controller and address space between x86 and GPU
(3) Highly programmable GPU "hardware"
(4) Expansion of very-parallel programming languages (i.e. OpenCL / ROCm)
(5) Infinity Fabric large memory architecture

They have most/all of the basic toolset required for people to put together some really powerful HPC software.

Then when you think of things like using augmented reality headsets as surgical tools, with the computing power to perform real-time interpolation of scans and present them on a headset for the surgeon to get a much better understand of what they are doing inside a patient, then accelerated compute has potential far beyond the "big clusters" and pure research you'd typically consider.

 

[Crumpet]

Zen+ will be less dramatic though

I'll still be here.. i'll still be excited.

I just like technology.

 

[Atari2600]

Zen+ will be less dramatic though

True, but due to the nature of the Zen gestation*, its bound to have rough edges where things didn't quite balance out as the designers hoped. There should be some relatively low hanging fruit lying around that the design crew can grab in their next iteration.


*basically coming from nowhere competitive, changing the fundamental direction re. BD, implementing SMT, doing so without any useful internal reference frames and doing it with a small team and relatively scant resources...

 

[majord]

Mike Clark himself mentioned several areas where comprimise had been made due to time /complexity. He was also quite specific about Zen being 'just the beginning' . I think it would be unwise to think just because we've only seen sub 10% IPC uplift for recent Intel 'tock's, that Zen will automatically be the same.

PS, again, they have had some experience in SMT vi

True, but due to the nature of the Zen gestation*, its bound to have rough edges where things didn't quite balance out as the designers hoped. There should be some relatively low hanging fruit lying around that the design crew can grab in their next iteration.


*basically coming from nowhere competitive, changing the fundamental direction re. BD, implementing SMT, doing so without any useful internal reference frames and doing it with a small team and relatively scant resources...

Mike Clark confirmed as much during his presentation.. (perhaps a little too eagerly.. he's clearly not a marketing guy!)

 

[OrangeKhrush]

True, but due to the nature of the Zen gestation*, its bound to have rough edges where things didn't quite balance out as the designers hoped. There should be some relatively low hanging fruit lying around that the design crew can grab in their next iteration.


*basically coming from nowhere competitive, changing the fundamental direction re. BD, implementing SMT, doing so without any useful internal reference frames and doing it with a small team and relatively scant resources...

There is some definite low laying tweaks on the horizon that could see maybe another 10% odd gain by update which is about on what David Kanter had said.

 

[Agent-47]

Zen+ will be less dramatic though

WTF is that display picture about?

Everyone is entitled to their own opinion be it alternate facts or actual facts.

 

[lobz]

Unlikely, unless of course all the leaks were a fluke and it is much worse.

I mean, that sounds like the most logical thing ever. FXs generally choke on a single thread in games , of course from putting load on the rest of CPU they stay flat.

however backwards you'd like to interpret what I've said, it's still fine by me, it won't change any facts

 

[CentroX]

Ryzen has better IPC then kaby lake

https://youtu.be/f-Kw4XFXB30

 

[lobz]

6900K is a terrible value and it doesn't have the "halo" effect of being the top SKU/having 10 cores.

Just a bad, overpriced SKU from Intel.

doesn't matter too much anymore, the entire Intel HEDT lineup will be rendered totally obsolete at their current prices. -> spare some very rare cases of course, when you want a 3-4 way SLI/crossfire setup, or you - for some strange reason - really need more than 2 channels of already high data throughput memory. But I don't think it will be very-very long before you can purchase a Naples based CPU for that, with tons and tons of connectivity.

 

[OrangeKhrush]

Ryzen has better IPC then kaby lake

https://youtu.be/f-Kw4XFXB30

No it doesn't, I would not lead yourself into being that ambitious, Haswell+ is about the best case realistic expectation that seems to have worked out.

 

[CatMerc]

I can believe higher IPC than Kaby in specific workloads, especially integer heavy ones, but on average it should be Broadwell-ish.

 

[JDG1980]

6950x has the most atrocious perf/$ ratio of all intel CPUs. Heck it makes 6900K look good actually. 57% higher price for 25% more cores and roughly the same OC with a much higher power draw.

Yet even so, people bought it - some, presumably, just so they could say they had the absolute best (consumer) CPU out there.
Assuming the Ryzen launch is a success, I wonder if AMD might release a consumerized version of Naples somewhere down the road? Sell the higher-leakage, higher-clocking 16C/32T chips into the mainstream market, presumably as "R9" series (since 3, 5, and 7 are already taken). If Intel doesn't drop their existing prices on 6950X or its successor, then AMD might even be able to charge $999 for this halo part - more cores than anyone else (without going up to non-OC workstation parts) and clock speeds decent enough not to cripple single-threaded performance. Even if the Skylake-X successor to 6950X is a 12-core CPU, Zen at 16C/32T would probably overcome Intel's now-slight archtectural edge to give AMD the advantage in most benchmarks.

 

[inf64]

Ryzen has better IPC then kaby lake

https://youtu.be/f-Kw4XFXB30

That guy is discussing CPUz benchmark and assuming this will reflect all workloads? That is not smart at all. We have been all over the CPUz numbers and although they are impressive(IPC and total throughput) they cannot be used as a stick to measure general IPC of a core.

Zen will not have higher IPC than SKL or BDW or maybe even HSW. I think it will be a smidge slower than HSW on average but this is a HUGE result since IPC improvements were minuscule since then. Zen will also perform slower in AVX(2)/FMA heavy workloads due to how AMD built its FP co-processor but this will affect only a certain number of workloads.

 

[sm625]

As a result, dividend and share buybacks will be stoped or at least significantly reduced. Brian Krzanich gets replaced as CEO..

That would be the rational response. But what's more likely is they accelerate the share buybacks to record highs and beyond. The buybacks wont stop until about 18 months after the yield curve inverts, which is likely still 3 years out. I wouldn't be surprised to see Intel's debt reach $40 billion by 2020.