[Deustche Bank Conference] AMD's New x86 Core is Zen, WIll Launch WIth K12

[mrmt]

The plan is servers, targeting data centers.
[/url]

And that's low volume.

 

[Cerb]

SteamRoller Modules can easily compete in MT loads even on Clock to Clock against Haswell 2C 4T CPUs.

Ray tracing, and highly multithreaded 7-zip compression are tasks I do all day every day.

http://www.hardcoreware.net/intel-core-i3-4340-review/5/

There's a 4M8T showing less smooth gameplay, across multiple games, than a 2C4T. Meanwhile, the 2C4T uses much less power to do it, with but a single exception. The sad part of the exception: none of the CPUs spent too much time under 60FPS, so while one feather in AMD's cap, not compelling.

http://www.xbitlabs.com/articles/cpu/display/core-i3-4340-4330-4130_5.html#sect0
When competitive in terms of raw performance (save that the i3 had its games, there, too), power use is not just a wee little bit higher, but far higher, much like back when the A64 was taking the P4 to task:
http://www.xbitlabs.com/articles/cpu/display/core-i3-4340-4330-4130_7.html#sect0

And still there are workloads that FX is still faster than any i5 today.

I never stated otherwise. Where they are sufficiently fast, they use too much power. Where they aren't sufficiently fast, Core i3s typically meet or best them, plus use less power. In the few cases they are nominally similar, there aren't enough buyers (lacking IGP does not help, either), so they have to price them low.

 

[Abwx]

I never stated otherwise. Where they are sufficiently fast, they use too much power. Where they aren't sufficiently fast, Core i3s typically meet or best them, plus use less power. In the few cases they are nominally similar, there aren't enough buyers (lacking IGP does not help, either), so they have to price them low.

Before making bold statements about the i3 please read this review, where would you put the i3 on the following perfs graphs.?.

http://www.hardware.fr/focus/99/amd-fx-8370e-fx-8-coeurs-95-watts-test.html

And remember, your i3 should be better than a 4670K if you want your claim to have the slightest meaning.

 

[Cerb]

Before making bold statements about the i3 please read this review, where would you put the i3 on the following perfs graphs.?.

http://www.hardware.fr/focus/99/amd-fx-8370e-fx-8-coeurs-95-watts-test.html

http://www.hardware.fr/articles/901-4/performances-jeux-3d.html
There are a few missing between them, but there you go. No need to extrapolate. I hunted down the review I did for frame time comparisons, to show just how much the difference means (I had trouble finding such reviews of low-end CPUs at other sites that give such results).

My statements are not at all bold. They are simply some of the reasons most of the AMD lineup is cheaper than Intel's, that has nothing to do with conspiracy theories, abusive monopolistic behavior (not that Intel isn't guilty of such all the time), untapped potential, etc.. Both of the above reviews show very similar results, including in overall efficiency, to what I linked to prior.

And remember, your i3 should be better than a 4670K if you want your claim to have the slightest meaning.

Or...not. That's a ridiculous notion.

 

[MiddleOfTheRoad]

How so? It would lose just as bad to 22nm Xeons as it does to 32nm Xeons. CPU wise, 28nm SR cant even remotely compete with 32nm SB either.

Remember this?

You're definitely distorting the facts -- AMD decided to allocate a lot more silicon to the GPU design at 28nm than Intel. If the CPU's were of equal size (AMD could really use some L3 cache)..... They'd be pretty damn close to each other.

The 22nm i7 3770k is slower than a 32nm FX-8350 in 7 benchmarks on Phoronix. And am I the only one that noticed that you love to compare AMD chips with Intel chips that cost twice as much as the AMD?

 

[mrmt]

You're definitely distorting the facts -- AMD decided to allocate a lot more silicon to the GPU design at 28nm than Intel. If the CPU's were of equal size (AMD could really use some L3 cache)..... They'd be pretty damn close to each other.

AMD allocates more silicon for the cpu and still gets inferior performance

 

[MiddleOfTheRoad]

Last, but not least, AMD always had a professional business, they just couldn't get it right because they can't fix their software support and devrel.

I think you bashed AMD a little too soon -- the latest numbers indicate that AMD picked up sizeable marketshare in Professional Graphics.... AMD just jumped from 15% to 25% in Q2 2014.... AMD got a nice pop from the design win in the new Mac Pro.

http://seekingalpha.com/article/2491735-amds-high-margin-business-grows-market-share

Enthusiasts may not be thrilled with many of the decisions coming from Rory Read -- but he is definitely remaking AMD into a more stable / healthy company. Under the previous management, they only made a profit 2 out of the previous 8 years.

 

[Abwx]

http://www.hardware.fr/articles/901-4/performances-jeux-3d.html
There are a few missing between them, but there you go. No need to extrapolate. I hunted down the review I did for frame time comparisons, to show just how much the difference means (I had trouble finding such reviews of low-end CPUs at other sites that give such results).

The i3 perfs on games wont last long given that we re going to have more MT, besides the page you linked is already obsolete as Hardware.fr updated their games suite wich was still stuck with BF3 and Crysys 2 for exemple, the result is that the low core count CPUs got a severe hit, the 4670K advantage over the FX8350 was halved on the process :

http://www.hardware.fr/articles/924-19/indices-performance.html

The link i posted show a quite different behaviour for the i7 and the FX, it is clear that it s not single thread perf that is causing the FX to perform not as good as the intel CPUs, looking at the average of the games we can see that the 4770K scale almost linearly with frequency when comparing to the 4790K while the FX get only 18% improvement when frequency is increased by 42%, so there s definitly something at work that is not ST perf, it could be caches latencies and bandwith or whatever else but not ST perfs.

Old graph :

New graph with updated game suite :

 

[MiddleOfTheRoad]

AMD allocates more silicon for the cpu and still gets inferior performance

That is not true -- 47% of the silicon in AMD's latest chips is dedicated to the GPU.
Only about 32% of the real estate in Haswell is dedicated to the GPU.... 68% of
Intel's die is dedicated to the CPU, while AMD only allocates 53%..... They clearly
have different design goals.

 

[Cerb]

The i3 perfs on games wont last long given that we re going to have more MT, besides the page you linked is already obsolete as Hardware.fr updated their games suite wich was still stuck with BF3 and Crysys 2 for exemple, the result is that the low core count CPUs got a severe hit, the 4670K advantage over the FX8350 was halved on the process

The low core count CPUs took a severe hit, yet they aren't even listed? The FX is a bit faster in games that can use many cores now, and over time. Those that can't are not dead, however, nor will they be any time soon. MMOs have a harder time of it, and strategy and sim type games are just plain hard to scale out too much, so will consistently lag behind in that regard. Fast paced action games will benefit from more cores. But, as long as Intel's i5 series is, on average, faster and cooler, they are better buys, and thus justifiably cost more. Other reviews show similar results, over and over again. Likewise, the i3s being able to beat AMD's 6 and 8 thread CPUs provides justification for that pricing. Given that they are huge and run hot, and are not popular among OEMs, it also justifies calling them failures, from the company that has before bested Intel with a fraction of its budget.

The link i posted show a quite different behaviour for the i7 and the FX, it is clear that it s not single thread perf that is causing the FX to perform not as good as the intel CPUs, looking at the average of the games we can see that the 4770K scale almost linearly with frequency when comparing to the 4790K while the FX get only 18% improvement when frequency is increased by 42%, so there s definitly something at work that is not ST perf, it could be caches latencies and bandwith or whatever else but not ST perfs.

42% frequency increase? 8350 to 9590 is about 18%. I don't see any 42% anywhere.

 

[Enigmoid]

That is not true -- 47% of the silicon in AMD's latest chips is dedicated to the GPU.
Only about 32% of the real estate in Haswell is dedicated to the GPU.... 68% of
Intel's die is dedicated to the CPU, while AMD only allocates 53%..... They clearly
have different design goals.

Kaveri = 245 mm^2 * 53% = 130 mm^2 non GPU
Sandy Bridge = 216 mm^2 * 0.83 = 179 mm^2 non GPU

Kaveri is 28 nm (density optimized) vs 32 nm so both have roughly have the same amount of normalized die area for the CPU.

4C8T SB destroys kaveri 2M4T in single and multithread.

The i3 perfs on games wont last long given that we re going to have more MT, besides the page you linked is already obsolete as Hardware.fr updated their games suite wich was still stuck with BF3 and Crysys 2 for exemple, the result is that the low core count CPUs got a severe hit, the 4670K advantage over the FX8350 was halved on the process :

http://www.hardware.fr/articles/924-19/indices-performance.html

The link i posted show a quite different behaviour for the i7 and the FX, it is clear that it s not single thread perf that is causing the FX to perform not as good as the intel CPUs, looking at the average of the games we can see that the 4770K scale almost linearly with frequency when comparing to the 4790K while the FX get only 18% improvement when frequency is increased by 42%, so there s definitly something at work that is not ST perf, it could be caches latencies and bandwith or whatever else but not ST perfs.

Old graph :

New graph with updated game suite :

This is true. The 8350 should age fairly well. i3s will likely age poorly.

Note, old games, 4670k is 42% faster than the 8350, newer gaming index it is only 26% faster.

 

[Abwx]

4C8T SB destroys kaveri 2M4T in single and multithread.

Just that SB 1C/2T is as big as a whole module, the comparison is 1C/2T with 1M/2T, in ST SB has the advantage but in MT that s not the case, the 2600K as well as the 2500K used to perform very well against a 8350 but that was almost two years ago, current benches show the 2500K almost oudated in MT and the 2600K globaly outmatched by the 8350 in MT, Kaveri can only be compared to i3s and in his register it does quite well overall.

 

[Cerb]

That is not true -- 47% of the silicon in AMD's latest chips is dedicated to the GPU.
Only about 32% of the real estate in Haswell is dedicated to the GPU.... 68% of
Intel's die is dedicated to the CPU, while AMD only allocates 53%..... They clearly
have different design goals.

And, it shows:
http://www.tomshardware.com/reviews/amd-a10-7800-kaveri-apu-efficiency,3899-6.html
http://www.tomshardware.com/reviews/amd-a10-7800-kaveri-apu-efficiency,3899-8.html
http://shopping1.hp.com/is-bin/INTE...se?CatalogCategoryID=EOcQ7EN6eEUAAAFFh1kRvdw1
Just that they really needed that, especially power-wise, to be the BD respin in 2012, to avoid the situation they are in now.

AMD has to accept that Intel's combination of talent, money, and process lead will leave them without a processor that can be considered an overall superior one to whatever Intel has, much less what they will have soon (not decent looking in a few benchmarks, but in aggregate, including power/heat). But, they can, should, and in the future, need to be, excellent for the OEMs making those products for the end users. They don't need to be great to you and me, but to HP, Acer, Lenovo, et al. They need to be compelling for HP to put in Proliants, for Supermicro to do something quirky with, for Synology to start using, and so on and so forth.

 

[Abwx]

The low core count CPUs took a severe hit, yet they aren't even listed?

They havent got the time to do the measurement using their new suite, it will be updated in the next weeks/months but you have the 4670K as reference, if the 8350-4670K difference was drasticaly reduced i think that you are not expecting the 2C/4T to resist better than a real 4C.

But, as long as Intel's i5 series is, on average, faster and cooler, they are better buys, and thus justifiably cost more.

Remember the 2500K two years ago when 8350 was launched and that softs werent as MThreaded as actualy...

42% frequency increase? 8350 to 9590 is about 18%. I don't see any 42% anywhere.

I picked the number on the 8370E review, base frequency increase from 3.3 to 4.7 from the 8370E to the 9590.

http://www.hardware.fr/focus/99/amd-fx-8370e-fx-8-coeurs-95-watts-test.html

 

[Cerb]

4C8T SB destroys kaveri 2M4T in single and multithread.

I would hope it would. But, AMD can't compete in die size with Intel, period. Even when their processes have been similar, Intel can put more man hours into optimizing size than AMD can. While power use has hampered them greatly, 4M8T was always intended to compete against 4C8T, regardless of die size.

 

[Cerb]

I picked the number on the 8370E review, base frequency increase from 3.3 to 4.7 from the 8370E to the 9590.

OK. Missed that.

Let's say they stay at base frequencies. Anno 2070 is, and well should be, among the worst as far as single-threaded performance goes (and a type of game that will generally have a single thread the rest are waiting on, no matter how many total it scales out to). In base clock speed:
9590/8350: 17%
8350/8370E: 21%
In Turbo clock speed:
9590/8350: 19%
8350/8370E: -2.4%
Average if running at 50% between base and Turbo, on average:
9590/8350: 18%
8350/8370E: 7.9%

In Anno:
9590/8350: 14%
8350/8370E: 15%

If the 8370E can spend a good bit of time in faster states, that could very well explain it, and it would make sense for it to do so, since samples are likely good bins. The 9590 v. 8350, and most performance results of other PD speeds at different speeds, work pretty linearly, but they all have much tighter ranges between base and Turbo.

http://www.anandtech.com/show/8427/amd-fx-8370e-cpu-review-vishera-95w/3
That very well may be it. Looking at more gaming results, it's all over the map. But its base v. Turbo is +30%, plus they're likely good bins, and it matches well to one end of its clock speed range often. Only the 8370E is out of place, generally, as well.

Remember the 2500K two years ago when 8350 was launched and that softs werent as MThreaded as actualy...

As today, you mean? Not quite, no. But, the faster i5 CPUs still have an edge absolutely, a larger edge with fewer usable threads, and requires ignoring heat entirely to not look even better than the longer raw performance graphs indicate. At the same price as an i5-4460 (though often at $20-40 less), the FX-8350 is certainly good enough if you're going to make use of it where it can be better, but most people will be better served with the i5, all things considered (like power, noise, chipset, and programs that don't scale out really well, if at all).

 

[ShintaiDK]

You're definitely distorting the facts -- AMD decided to allocate a lot more silicon to the GPU design at 28nm than Intel. If the CPU's were of equal size (AMD could really use some L3 cache)..... They'd be pretty damn close to each other.

The 22nm i7 3770k is slower than a 32nm FX-8350 in 7 benchmarks on Phoronix. And am I the only one that noticed that you love to compare AMD chips with Intel chips that cost twice as much as the AMD?

Again, compare SB Xeons with Pilediver Opterons. And you can easily see how bad AMDs uarch is.

What is the work again on your 7 selected benchmarks in Linux? How big of a crowd does it apply to? See the issue?

How is Kaveri holding up with the SR uarch? Pretty terrible, thats how. So unless you use 300mm2 on multiple modules to run some very scaling INT MT load. Its just terrible slow and inefficient compared to the competition.

And again, even AMD stated the uarch is a total failure. It cant be much more clear than that.

 

[mrmt]

That is not true -- 47% of the silicon in AMD's latest chips is dedicated to the GPU.
Only about 32% of the real estate in Haswell is dedicated to the GPU.... 68% of
Intel's die is dedicated to the CPU, while AMD only allocates 53%..... They clearly
have different design goals.

You can't compare SNB 4C with AMD APU, do it with the 2C.

 

[AtenRa]

Ray tracing, and highly multithreaded 7-zip compression are tasks I do all day every day.

http://www.hardcoreware.net/intel-core-i3-4340-review/5/

There's a 4M8T showing less smooth gameplay, across multiple games, than a 2C4T. Meanwhile, the 2C4T uses much less power to do it, with but a single exception. The sad part of the exception: none of the CPUs spent too much time under 60FPS, so while one feather in AMD's cap, not compelling.

http://www.xbitlabs.com/articles/cpu/display/core-i3-4340-4330-4130_5.html#sect0
When competitive in terms of raw performance (save that the i3 had its games, there, too), power use is not just a wee little bit higher, but far higher, much like back when the A64 was taking the P4 to task:
http://www.xbitlabs.com/articles/cpu/display/core-i3-4340-4330-4130_7.html#sect0

First of all we are talking about applications, not gaming. Games behave differently than all other applications. Haswell core i3 is only faster than Kaveri APUs in Cinebench, they loose or are equal in the vast majority of MT workloads out there.
Secondly there aren't any SteamRoller products like Kaveri APUs in those links you provided above.

 

[AtenRa]

You can't compare SNB 4C with AMD APU, do it with the 2C.

Yes please,

Compare 2M 4T Kaveri to 2C 4T GT3 Haswell.

 

[frozentundra123456]

Why can't you compare 7850k to an i5? Just because it loses badly in everything but igp performance? Although some of the lower end kaveri models are reasonably priced, top end kaveri is very close to a locked i5.

 

[mrmt]

Why can't you compare 7850k to an i5? Just because it loses badly in everything but igp performance? Although some of the lower end kaveri models are reasonably priced, top end kaveri is very close to a locked i5.

Actually you can compare against Xeon if you want, but it wouldn't make sense. This is what happened:

- Middleoftheroad said that Intel devotes more silicon for the CPU than Intel does.

- I said that it doesn't. In fact, the entire SNB-2C die is roughly the size of Richland/Trinity CPU part only, therefore AMD indeed devotes more area for the CPU and still gets inferior performance. You can do the same with i3 Haswell and Kaveri and reach the same conclusion.

- i5 indeed has more CPU area than AMD CPU, but it's a de-featured part. If you want to see the full potential of the chip, you have to take the more complete part of both products, and i7 gives performance levels far beyond you can get with AMD APUs in the CPU department, hence the comparison would be flawed. i3 is much more comparable to Trinity/Kaveri than i5 or i7.

 

[frozentundra123456]

I was not referring to your post specifically. TBH, I don't really think about the die area. It just seems like a parameter more relevant to the manufacturer cost than me, the end user. All I was trying to point out is that the top end kaveri is really in low end i5 territory, and gets destroyed in pretty much anything but igpu performance.

 

[Idontcare]

It's a great slide, really, but unfortunately it is one to which AMD's shareholders respond "Awesome! So when will you actually make some money on this whole 'the future is fusion' phenomenon? Cause all we see is that Intel is making all the profits along the way..."

 

[AtenRa]

Why can't you compare 7850k to an i5? Just because it loses badly in everything but igp performance?

Then it doesnt loses badly in everything because i5 loses badly in iGPU performance.

You can compare the two through price if you want, one is better in CPU the other is better in iGPU. Depending on what you are after, both will give you best performance for what you need.

But technically Kaveri 2M 4T with 512Radeon Cores can be compared directly to Haswell 2C 4T with GT3 graphics. And although Intel Core i3 + GT3 is using 22nm FF, the Kaveri can actually compete in CPU performance, iGPU performance and die sizes are not that far apart between the two.
Core i3 + GT3 has better power consumption only because of the node and FF advantage.