*** AMD "Zen" thread (inc AM4/APU discussion) ***

[Boomstick777]

The price is representative of actual performance.
They're not pricing low because they care, they price it depending on what the market will pay.

Martini1991 already said this, if they have a product that is worthy and command a decent price then good on em. Speaking for myself I would be happy to pay for a decent AMD product just like I do with Intel / Nvidia.

Just in the past few years AMD haven't had the products, hopefully this will change next year... So far away though..

 

[humbug]

AMD can charge more than they currently do if the performance is this high, 130 4 core, 200 6 core and perhaps even £300+ for the 8 core, thats way less than Intel whose prices are going up not down.

 

[humbug]

Question, AM3+ Excavator this year?

 

[ALXAndy]

Question, AM3+ Excavator this year?

From what I recall excavator is not going to be released on AM3+. I'm sure I remember it being an APU for their other socket.

 

[humbug]

From what I recall excavator is not going to be released on AM3+. I'm sure I remember it being an APU for their other socket.

Bah

 

[ALXAndy]

Bah

Humbug

 

[Orangey]

The 40% will be absolute best case, which for Haswell is 70% vs Piledriver.

The 20nm-class intra-chip wiring on the GF process will be a problem as well vs Intel's.

 

[joeyjojo]

your having a laugh, Haswell is about 30 to 40% faster than piledriver.

The 40% will be absolute best case, which for Haswell is 70% vs Piledriver.

IPC? It's in excess of 50%. Or at least can be.

With single-threaded Cinebench

8370E: 1.1 CBMarks @ 3.3 GHz = 0.333 CBMarks/GHz/thread
4790K: 1.7 CBMarks @ 3.5 GHz = 0.486 CBMarks/GHz/thread

In this workload, efficiency is about 46% higher in the 4790K than the 8370E. This is probably the largest difference you'll find.

It should also be noted that the whole idea of running a single threaded benchmark favours the Intel architecture (penalises the modular AMD architecture). In threaded apps comparing the 4790K to the 8370E (8 threads) the difference in "efficiency" shrinks to 27%.

 

[Martini1991]

With single-threaded Cinebench

8370E: 1.1 CBMarks @ 3.3 GHz = 0.333 CBMarks/GHz/thread
4790K: 1.7 CBMarks @ 3.5 GHz = 0.486 CBMarks/GHz/thread

In this workload, efficiency is about 46% higher in the 4790K than the 8370E. This is probably the largest difference you'll find.

It should also be noted that the whole idea of running a single threaded benchmark favours the Intel architecture (penalises the modular AMD architecture). In threaded apps comparing the 4790K to the 8370E (8 threads) the difference in "efficiency" shrinks to 27%.

You're misusing the term "threaded" and running a single thread doesn't penalise the AMD modular architecture, running 2 threads on a module would result in a performance (Which translates to IPC) penalty.
Look at ; http://forums.overclockers.co.uk/showpost.php?p=27688774&postcount=573

Look at Nkata's 5 GHZ and Tonesters 5GHZ, Tonesters score is 67% higher. Hell, if you take a look at my 4.8GHZ score versus Damo's, it's again about 64%. Take the highest FX83 at 5.3GHZ to my 4.8GHZ result, and it's 48% (And that's not directly comparing IPC). How can you tell me that the IPC difference (At least in this circumstance) can ever be 30-40%? It's just Humbug living in his own world again. How much difference do you think would be if it was X fully loaded PD cores versus the same amount of fully loaded Haswell Cores at the same clock? It's not going to be pretty.

Either way, people can throw about "An Intel core is 40-50% whatever percent faster than an AMD core" all they want. A possible 40% over Excavator in an 8 core package *should* be very good indeed.

And you're making up your own definition of efficiency by the sounds of it too.

But like I say, it's a meaningless argument.

 

[cheesyboy]

You're misusing the term "threaded" and running a single thread doesn't penalise the AMD modular architecture, running 2 threads on a module would result in a performance (Which translates to IPC) penalty.

Bingo.

I compared my Athlon x2 270 to a similarly clocked one of the dual core (single module) APUs in Cinebench - the APU won hands down in single core bench, but they were almost identical in multi core bench, where the shared floating point held the APU back.

If the +40% IPC quoted by AMD is for single core performance, rather than against fully saturated modules, then they're potentially on to a winner.

 

[Martini1991]

All I'm seeing is people arguing their flawed view points all over again, and doing it without the knowledge to do so.

 

[cheesyboy]

Anyway, back to yesterday's info...

Anyone notice anything interesting about this chart;


Looks to me like no Zen for the 2016 consumer APUs - presumably that's where Excavator is going to show up, with DDR4 support though.

 

[CAT-THE-FIFTH]

A 40% IPC increase over Excavator would mean at least a 50% IPC increase over Piledriver. Looking at the Cinebench R15 thread for the single thread scores for Piledriver and IB and Haswell,it would put Zen as having nearly the same IPC as IB and Haswell in that software.

However,clockspeeds are going to be another concern though and whether the IPC increase is peaky or not,ie,massive for some software and much less for others.

 

[Davedree]

With single-threaded Cinebench

8370E: 1.1 CBMarks @ 3.3 GHz = 0.333 CBMarks/GHz/thread
4790K: 1.7 CBMarks @ 3.5 GHz = 0.486 CBMarks/GHz/thread

In this workload, efficiency is about 46% higher in the 4790K than the 8370E. This is probably the largest difference you'll find.

It should also be noted that the whole idea of running a single threaded benchmark favours the Intel architecture (penalises the modular AMD architecture). In threaded apps comparing the 4790K to the 8370E (8 threads) the difference in "efficiency" shrinks to 27%.

You can't compare intel vs amd on a synthetic workload,
Use handbrake or something real.

 

[Boomstick777]

Either way, people can throw about "An Intel core is 40-50% whatever percent faster than an AMD core" all they want. A possible 40% over Excavator in an 8 core package *should* be very good indeed.

^^ This, Martini got it spot on.

I think we can take from this that at least AMD are focusing on traditional CPU performance space over other things, i.e prioritised over Zen APU and K12 ARM which are coming later in 2017.

This renewed focus along with a potential 40% uplift are promising.

Whether AMD can deliver it we'll see at launch but at least they are focusing here again. 2016 is a long ways off so until then it's Intel all the way, but good to know what AMD are trying to achieve.

 

[Orangey]

http://www.kitguru.net/components/c...r-direct-nvme-storage-promontory-pch-usb-3-1/

No PCIe4 and only 16 lanes?

WHY.

 

[joeyjojo]

You're misusing the term "threaded" and running a single thread doesn't penalise the AMD modular architecture, running 2 threads on a module would result in a performance (Which translates to IPC) penalty.
Look at ; http://forums.overclockers.co.uk/showpost.php?p=27688774&postcount=573

Look at Nkata's 5 GHZ and Tonesters 5GHZ, Tonesters score is 67% higher.

Thanks for pointing that out. I was taking the numbers from Anandtech's Cinebench 11.5 charts (which they filed under "CPU 2015 Professional Performance"), but the numbers you're looking at are Cinebench R15 (which Anandtech have under "2015 CPU Legacy" for some strange reason since it seems to be newer ).

The difference between Intel and AMD in R15 is much larger as you say, I make it 83% between the 4690K and the 8350.

(As you know) whether running two threads or one per module will favour one architecture or the other depending on the workload. I'm just trying to point out that there's no one value of IPC, so these promises of x% increases in IPC are open to being spun any which way.

 

[drunkenmaster]

AMD can charge more than they currently do if the performance is this high, 130 4 core, 200 6 core and perhaps even £300+ for the 8 core, thats way less than Intel whose prices are going up not down.

Yes and no, it's a catch 22, if they price too high then for some people they will feel no reason to switch from Intel to AMD, particularly those who have an upgrade option on their current mobo(though thankfully Intel doesn't do that too often ).

AMD could do with clawing back marketshare and giving the likes of Dell no reason to not push AMD chips. So offering a 8 core at 4970k type pricing and a quad core around the £100 mark would help push sales significantly. The problem being if/when sales increases, if you just whack up the price of chips £50-100 per price bracket you look like a bit of a ****.

I think AMD have to be price competitive, extremely so, till they have a year or two being deemed competitive again. Volume and higher margins are essential for AMD in a push to pay off debt and become safely profitable again.


Currently the biggest implication it has for enthusiasts/gamers is, potentially FX series chips for mainstream. Instead of paying £250 for a 4 core + ht + entirely unused iGPU as we do with Intel currently. We have the option of likely getting 4 cores at £100-120 with no igpu, and 8 cores at £200 again with no iGPU. then have a range of chips with iGPU for the places that makes the most sense, lower end, mobile, etc.

I've had a 2500k for an age now, £250+ to go from a quad core to quad core with HT + a hugely improved iGPU and what 15-20% ipc improvement. It's not an attractive upgrade. I, like lets be honest most of this forum, don't use the iGPU, it's an entire waste to have it on silicon and pay for a larger chip that has it.

FX series with 4/8 cores at way cheaper price brackets could clean up in enthusiasts/gamers... might even force Intel to release their own mainstream 4/8 cores without a damn iGPU.

The x99 platform isn't terrible but expensive mobos, expensive chips and insane priced if you want more than 16 pci-e lanes makes it a very niche market. bringing only CPU's back to mainstream pricing is what enthusiasts/performance/gamers have needed for the past 5 years.

 

[Orangey]

AMD need an option for mGPU people, there must be a bigdaddy waiting in the wings.

 

[d_brennen]

I've had a 2500k for an age now, £250+ to go from a quad core to quad core with HT + a hugely improved iGPU and what 15-20% ipc improvement. It's not an attractive upgrade. I, like lets be honest most of this forum, don't use the iGPU, it's an entire waste to have it on silicon and pay for a larger chip that has it.

You, like the rest of this forum, do not represent the entire market. Having an iGPU capable of high resolution video, PowerPoint and nice infographics is hugely important to the markets that actually matter, the box shifters. It reduces the component count and therefore BoM, no descrete GPU needed, no onboard GPU needed, cheaper motherboards, less assembly, smaller chassis, lower weight... The list goes on.

While the 2500K is one of Intel's concessions to the DIY crowed, it's cut from the same cloth as the CPUs destined for Dell, HP et al. It makes little business sense to have several dies in design, validation and production, when you can have one and fuse off bits that don't work and segment your line that way.

I find it a curious move that Zen will be available iGPU-less - AMD has made a huge amount of noise about heterogeneous compute, and the potential for is is immense. But I guess this could be down to the server market, where pure x86 performance is in demand, and once again we get the scraps that don't make Opteron grade.

The more things change, the more they stay the same