bitsandchipsNo Dual and Four Cores Zen CPUs, at least initially

[majord]

What is known is the following Gate Pitch x Metal Pitch

Intel 14nm = 70nm x 52nm

Samsung 14nm LPE = 78nm x 64nm



But, I also believe that 8-Core ZEN die should be smaller than 200mm2. It could be close to 160mm2 as Silverforce11 estimates.

Believe That's 70x 60 for Intel's "hp" process.

Also, SRAM is about 10% larger on Samsung ff.

I think the variables are too great to have this debate quite yet though. Highly Likely the core would end up smaller than Skylake even with a process density handicap, how much smaller is the unknown. E.g. Shrinking excavator module (with its HDL layout and GP stack) to 14nm would leave it roughly 30% smaller than a Skylake core.

Then you have twice as much L2 per core to consider, and an unknown amount of L3?

Intel have always tended to sacrifice area for larger speed optimised designs, using process node to their advantage (have their cake and eat it too). Now that density advantage is somewhat eroded, so it will be interesting to see there approach moving forward

 

[Phynaz]

No, not that close, but it doesn't need to be because the difference, ultimately, is not going to result in a much bigger chip.

Think about this and let it sink in. 4C/8T SMT of Skylake quality is ~60mm2.

Read that again, maybe you understand, these chips are small.

The cost is mostly R&D at this point, because manufacturing these small chips are cheap!

Manufacturing semiconductors at today's leading edge nodes is far from cheap. Go to ASML and get a quote on the equipment to create a line that can start a very modest 10,000 wafers per month. That billion dollars combined with crappy yields makes this generation of manufacturing anything but cheap.

If manufacturing was cheap AMD wouldn't have sold off their factories.

 

[Erenhardt]

Very interesting. They will cover whole market with 2 products it seems. 6-8 cores (6? to 16 threads)for power users and servers, and 2 to 4 cores (2? to 8 threads) for mainstream coupled with some GCN and HBM for an APU that could cover everything up to mainstream GPUs.

No cut down FX part from 8 to 4 cores could indicate they don't have good power gating on Zen.

 

[thilanliyan]

If manufacturing was cheap AMD wouldn't have sold off their factories.

I thought they sold to make some quick cash?

On topic, as long as Zen is competitive, I'd definitely buy an 8-core. Not interested in sticking with 4-core.

 

[seitur]

Not surprising and good decision by AMD.

Who needs 2 core processors without iGPU these days? Considering Zen cores will be weaker than Intel Core ones similar should apply to 4 core situation too.

You will get 2 and 4 Zen CPU too. Just with iGPU attached at a later date.

 

[stockwiz]

dual core should be dead anyways. I made sure my mother got a quad core laptops as you can often find both dual and quad core at similar price ranges.. it flies compared to the dual core she bought from best buy and returned.

It should be 4-8 cores by now.. it only isn't because of lack of competition.

 

[Dresdenboy]

I thought they sold to make some quick cash?

That was a strategic decision knowing, that they won't fab for others anytime soon.

With your own fabs, you need to ensure having products, which can be sold at good margins and volumes most of the time to offset process development costs. As the latter are constantly growing, you'd need to get volumes and margins growing at roughly similar rates. And the risk of fluctuating market performance is completely on your side (both directions - low sales and capacity constrained sales). If you're standing at a financial cliff, that's not a good position to work from in any direction.

 

[hojnikb]

dual core should be dead anyways. I made sure my mother got a quad core laptops as you can often find both dual and quad core at similar price ranges.. it flies compared to the dual core she bought from best buy and returned.

It should be 4-8 cores by now.. it only isn't because of lack of competition.

I'll gladly take faster dual core than slower quad core any day of the week.

Good case in point is pentium N series and i3. You'd think i3s sux because they are dual cores, but they simply destroy pentiums (as they should).

 

[Dresdenboy]

I'll gladly take faster dual core than slower quad core any day of the week.

Good case in point is pentium N series and i3. You'd think i3s sux because they are dual cores, but they simply destroy pentiums (as they should).

You're actually not talking about quad core vs. dual core, but about Atom vs. bigger cores.

 

[poohbear]

dual core should be dead anyways. I made sure my mother got a quad core laptops as you can often find both dual and quad core at similar price ranges.. it flies compared to the dual core she bought from best buy and returned.

It should be 4-8 cores by now.. it only isn't because of lack of competition.

So according to this logic, the AMD FX CPUS should kickass. But they don't. Could it be....There's more to a CPU'S performance than just the core count? Quad cores gor mass market have been around since 2006. Doesn't mean u just pop in a quad core from 2006 & voila! It trumps a modern dual core.

 

[The Stilt]

EDIT: Wrong thread.

 

[MiddleOfTheRoad]

So according to this logic, the AMD FX CPUS should kickass. But they don't.

They actually do, but their main strength was never gaming.

FX chips can still render video and crunch integers with amazing speed considering their age.

 

[jhu]

They actually do, but their main strength was never gaming.

FX chips can still render video and crunch integers with amazing speed considering their age.

FX is decent, but not amazing, which is why I replaced my FX8350 with dual Xeon E5 2670s (at least 2.5-3x rendering performance).

 

[SAAA]

FX is decent, but not amazing, which is why I replaced my FX8350 with dual Xeon E5 2670s (at least 2.5-3x rendering performance).

These old server chip really make for a steal deal... What about 16 cores, 8 modules opterons? CPU makers really compete with themselves these days rather than each other...

 

[Dresdenboy]

These old server chip really make for a steal deal... What about 16 cores, 8 modules opterons? CPU makers really compete with themselves these days rather than each other...

That Hackintosh GB score was of such a chip@4+GHz IIRC.

 

[jhu]

These old server chip really make for a steal deal... What about 16 cores, 8 modules opterons? CPU makers really compete with themselves these days rather than each other...

Those chips are really cheap too, but the motherboards are at least twice the price of the Xeon boards. Unfortunately, their performance isn't that great either: takes 32 modules to beat my 16 core machine. That would mean getting a 4 CPU board, and those are even more expensive than dual boards.

 

[HiroThreading]

They actually do, but their main strength was never gaming.

FX chips can still render video and crunch integers with amazing speed considering their age.

That's exactly what the Intel fanboys said about the Pentium 4 and Pentium D...

 

[Abwx]

Those chips are really cheap too, but the motherboards are at least twice the price of the Xeon boards. Unfortunately, their performance isn't that great either: takes 32 modules to beat my 16 core machine. That would mean getting a 4 CPU board, and those are even more expensive than dual boards.

The 32M opterons should score about 3000 in this test, for instance taking account the CMT penalty a FX8350 has a MT score of 640, that is 80/core at 4GHz and hence 48 at 2.4GHz...

Actualy you could do a lo of oher things while rendering with this software that scale at only 80%, in this case that s about 12.8 cores left unused..

 

[jhu]

The 32M opterons should score about 3000 in this test, for instance taking account the CMT penalty a FX8350 has a MT score of 640, that is 80/core at 4GHz and hence 48 at 2.4GHz...

Actualy you could do a lo of oher things while rendering with this software that scale at only 80%, in this case that s about 12.8 cores left unused..

You're forgetting memory access penalties on multi-processor systems. So the scores are correct.

 

[MiddleOfTheRoad]

FX is decent, but not amazing, which is why I replaced my FX8350 with dual Xeon E5 2670s (at least 2.5-3x rendering performance).

LOL, talk about the understatement of a century.

One would hope that dual 16 thread CPU's would be faster than a single 8 thread chip.

Did you just compare 32 threads of CPU throughput to 8? Seriously? Talk about Apples To Oranges.
I'd bet the dual E5 owns on the World Community Grid as well..... But a consumer FX was never
designed to compete with server chips.

 

[MiddleOfTheRoad]

That's exactly what the Intel fanboys said about the Pentium 4 and Pentium D...

Doesn't make it any less true.

 

[Abwx]

You're forgetting memory access penalties on multi-processor systems. So the scores are correct.

If it s correct then it wasnt a relevant comparison, i can as well find an app where the opteron is maxed and where your Xeon is used at 80% and then point how great the Opteron is "scoring"...

 

[jhu]

If it s correct then it wasnt a relevant comparison, i can as well find an app where the opteron is maxed and where your Xeon is used at 80% and then point how great the Opteron is "scoring"...

Let's see it then. Because Povray and Blender are going to show similar scaling. Can't get around Amdahl's law.

 

[jhu]

LOL, talk about the understatement of a century.

One would hope that dual 16 thread CPU's would be faster than a single 8 thread chip.

It's about total performance, not just about cores or threads. Otherwise a Mediatek Helio X10 would be a great rendering CPU too (which it's not).

Did you just compare 32 threads of CPU throughput to 8? Seriously? Talk about Apples To Oranges.
I'd bet the dual E5 owns on the World Community Grid as well..... But a consumer FX was never
designed to compete with server chips.

Go to eBay now. How much is an FX8350? Now how much are the dual Xeons? Yeah, currently the pair of Xeons are less.

 

[Abwx]

Let's see it then. Because Povray and Blender are going to show similar scaling. Can't get around Amdahl's law.

Amdhal law is not at play in CB R15, the test is supposed to scale at almost 100% even with 64 cores.

Other than that this list is somewhat not accurate since the 2500K is "scoring" 646 at "stock" frequency (it s actualy a 4.8GHz...), that is 3 pts better than a FX8350 at stock, and also better than a stock 2600K...