[Xbitlabs]AMD Excavator Core May Dramatic Performance Increases.

[mikeymikec]

Excavator should have 50% jump in IPC over Bulldozer/Piledriver

Has a company ever produced an x86 processor that has a 50% jump in IPC over its predecessor?

Sandy Bridge's IPC improvement over Conroe was about 30% IIRC, which is the most dramatic improvement I can think of. The second most dramatic improvement was probably the Athlon 64 X2 over the A64, but I can't find a decent set of benchmarks to compare that easily.

Since the A64 X2 I think AMD have been managing a 10% IPC improvement, ignoring the Ph1 and BD. If those two 'dodgy' generations are taken into account, the % IPC improvement looks pretty shaky.

 

[inf64]

Has a company ever produced an x86 processor that has a 50% jump in IPC over its predecessor?

One problem with that question is that Excavator's predecessor is not PD/BD like you think but SR which hasn't even launched yet. If SR is 25% over PD then Excavator needs ~20% over SR to make it to ~50% over PD (1.25x1.2).

 

[AtenRa]

BD
PD + 15
SR + 15
XC + 15

BD to XC ~50%

 

[frozentundra123456]

BD
PD + 15
SR + 15
XC + 15

BD to XC ~50%

BD to PD was a nice performance increase, but I am not sure it was 15% clock for clock. More like 10% overall I think, with some higher of course.

SR and XC are both still conjecture, and dependent on maintaining the same clocks as well for overall performance.

 

[Soulkeeper]

I don't think performance is nearly as important as power usage.
Or perf/watt specifically.

In fact, I think most people would be content with the FX 8350's performance if it used ivy bridge's power ... Things really wouldn't look too bad then

 

[Insert_Nickname]

So you're suggesting we distort reality because the performance of their 8 cores isn't what an 8 core should be? Performance has NOTHING to do with core count. Similarly, modules are NOTHING like hyperthreading. They don't even perform anything like it.

No, I'm simply suggesting that AMD instead of calling a 2M/4T a quadcore call it a dualcore-with-extra-execution-resources. Because that is what a module is, a single core with some duplicated hardware, so it can handle two threads instead of one. Its not really a quadcore, so why should AMD market it as one...?

There is no shame in Intel having a 2C/4T i7, for example.

 

[TerryMathews]

Has a company ever produced an x86 processor that has a 50% jump in IPC over its predecessor?

I can't seem to find IPC data that far back, but K6-> K7 sure felt like at least 50%.

 

[Abwx]

I can't seem to find IPC data that far back, but K6-> K7 sure felt like at least 50%.

Or pentium to pentium pro...

 

[PCunicorn]

I really hope AMD releases this on AM3+. But I guess it's more likely they will release on a new FMx socket

 

[Fjodor2001]

How long can AMD deliver 15% CPU performance increase per generation when Intel is delivering 8%?

Is AMD picking low hanging fruit that Intel picked some generations ago, so Intel now has to grab the fruit higher up which doesn't bring as much CPU performance increase? I.e. the CPU performance gap between AMD and Intel will continue to shrink going forward?

 

[mikeymikec]

Or pentium to pentium pro...

The Pentium Pro wasn't the successor to the Pentium. The P2 was. The Pentium Pro was kind of like the Athlon FX compared to the Athlon 64.

The Athlon 600 versus the K6:
http://www.xbitlabs.com/articles/cpu/display/amd-athlon.html

Doesn't look like 50% IPC improvement to me.

How long can AMD deliver 15% CPU performance increase per generation when Intel is delivering 8%?

Prescott/Cedar Mill to Conroe was an 8% improvement, really? How about Conroe to SB? I'm pretty sure that the reason why IVB and Haswell weren't revolutionary in performance terms was because they didn't have to be, and Intel knows that desktop CPU development isn't where their R&D should be poured into.

 

[Enigmoid]

BD to PD was a nice performance increase, but I am not sure it was 15% clock for clock. More like 10% overall I think, with some higher of course.

SR and XC are both still conjecture, and dependent on maintaining the same clocks as well for overall performance.

It was around 10% because there was a jump in clockspeed from the 8150 to the 8350.

Piledriver is still around phenom II or slightly below it levels.

Phenom II at 4.0 Ghz above FX 4350 at 4.2 ghz.

Also notice the the Athlon x4 750K at 4.3 ghz is basically identical to the FX 4350 at 4.2 ghz. L3 does basically nothing (difference is smaller than Athlon 640 and phenon II).

Of course this does not take into account the scaling issues present in Piledriver due to the modular design.

But singlethread performance is still slightly behind phenom 2 in terms of IPC.

 

[erunion]

How long can AMD deliver 15% CPU performance increase per generation when Intel is delivering 8%?

?

AMD isn't delivering 15% IPC improvements.

Bulldozer was a reduction in IPC.
Piledriver delivered up to 7% IPC, plus 7% higher clocks for 15% total( The official number from AMD)
SR is still speculation.

 

[Sweepr]

It was around 10% because there was a jump in clockspeed from the 8150 to the 8350.

Piledriver is still around phenom II or slightly below it levels.

Precisely.

7.7% faster at equal clocks in applications.
3.8GHz-4GHz FX4300 is a bit slower than a 4.2-4.3GHz FX4170.
3.2GHz quad-core Phenom II matches a 3.8-4.0GHz 2M/4C Piledriver. Time to catch up to Phenom II level IPC Steamroller.

 

[Fjodor2001]

AMD isn't delivering 15% IPC improvements.

Bulldozer was a reduction in IPC.
Piledriver delivered up to 7% IPC, plus 7% higher clocks for 15% total( The official number from AMD)
SR is still speculation.

Note that I wasn't talking about IPC iprovements, but CPU improvements, i.e. regardless if that comes from IPC or frequency increase.

But do you mean this is incorrect?:

BD
PD + 15
SR + 15
XC + 15

BD to XC ~50%

 

[Abwx]

Precisely.
7.7% faster at equal clocks in applications.
3.8GHz-4GHz FX4300 is a bit slower than a 4.2-4.3GHz FX4170.
3.2GHz quad-core Phenom II matches a 3.8-4.0GHz 2M/4C Piledriver. Time to catch up to Phenom II level IPC Steamroller.

And 13.5% clock for clock in games according
to the very same link...

hint : "moyenne jeux" means "average in games"

 

[mikeymikec]

Has a company ever produced an x86 processor that has a 50% jump in IPC over its predecessor?

Admittedly when I made this comment I was thinking about PD's successor, not the successor to that, which makes me wonder: Why are we even discussing two (essentially) imaginary generations into the future?

IMO, it's imaginary until it's been released. Anyone not remembering the PR farce leading up to Bulldozer should bear that in mind. AMD probably has a vague idea what they want to accomplish with Excavator but if they knew how to design it, they wouldn't wait another generation before doing it.

http://en.wikipedia.org/wiki/Steamroller_(microarchitecture)

AMD estimates that these improvements will increase instructions per cycle up to 30% while maintaining Piledriver high clock rates with decreased power consumption

That's some pretty big talk considering the non-event that was BD and the modest improvement that is PD. Though I just noticed the all-important "up to".

 

[BallaTheFeared]

As happy as this would make me, AMD needs to increase IPC by about 100% on their modular design for them to go toe to toe with Intel per core, per clock.

Really start to wonder exactly what AMD could do to increase their per core/clock performance to those levels, and at the same time dramatically decrease their power consumption.

I love fairy tales though.

 

[Elixer]

I'm pretty sick and tired of talk about how AMD is going to eventually bend Intel over in x years time when blahblah is released. After nearly 8 years of playing second fiddle I think its about time people gave it a rest. Not to mention predicting Excavator performance at this stage is sort of ridiculous.

People sure do have short memories...
AMD did it once, the question is, can they do it again ?

Competition is a good thing.

 

[mrmt]

IIRC Steamroller is to be launched in Q114, and we are yet to hear anything of that processor. And instead of bringing attention to the soon-to-be launched product, AMD is trying to bring attention to... Excavator, a product which is at least 6 quarters away.

Gentlemen, get ready for another deception once Steamroller hit the review sites.

 

[inf64]

Precisely.



7.7% faster at equal clocks in applications.
3.8GHz-4GHz FX4300 is a bit slower than a 4.2-4.3GHz FX4170.
3.2GHz quad-core Phenom II matches a 3.8-4.0GHz 2M/4C Piledriver. Time to catch up to Phenom II level IPC Steamroller.

Or you can look at the 6350 and 4350 review here so you can straight compare 4350 and 4170 as they are basically identical clock wise(4.2Ghz base and 4.3Ghz Turbo) .

Apps:
4350- 109
4170- 100.6
Average IPC difference 8.3% in apps.

Games:
4350-121.1
4170- 106.2
Average IPC difference 14% in apps.

A can be seen on average PD is around 11% (rounded) faster than BD. That is basically the same core with minor tweaks. Now imagine what will SR do with changes like this:

• Store to load forwarding optimization
• Dispatch and retire up to 2 stores per cycle
• Improved memfile, from last 3 stores to last 8 stores, and allow tracking of dependent stack operations.
• Load queue (LDQ) size increased to 48, from 44.
• Store queue (STQ) size increased to 32, from 24.
• Increase dispatch bandwidth to 8 INT ops per cycle (4 to each core), from 4 INT ops per cycle (4 to just 1 core). 4 ops per cycle per core remains unchanged.
• Accelerate SYSCALL/SYSRET.
• Increased L2 BTB size from 5K to 10K and from 8 to 16 banks.
• Improved loop prediction.
• Increase PFB from 8 to 16 entries; the 8 additional entries can be used either for prefetch or as a loop buffer.
• Increase snoop tag throughput.
• Change from 4 to 3 FP pipe stages.

PD had none of above and was a pretty hefty upgrade of BD (~14% in games in terms of IPC for a minor core tweak is massive). I think SR will give us ~15% in terms of average IPC( somewhere more and somewhere less of course) and will eliminate the module scaling penalty which is around ~10-15% depending on the workload. So ST performance ~15% and MT 1.15x1.15~=1.32x or around 30% more IPC , both at the same clock. If you now compare SR to bulldozer, MT performance difference can be a significant 1.32x1.11=1.46x or 46%, just like VR-zone's source claimed a while back . Now if SR regresses in terms of clock then above % will be meaningless of course.

 

[TerryMathews]

The Athlon 600 versus the K6:
http://www.xbitlabs.com/articles/cpu/display/amd-athlon.html

Doesn't look like 50% IPC improvement to me.

Where are you drawing that conclusion from? None of those graphs have data for any of the K6 series.

 

[escrow4]

Will it even match the i7 4770 I currently have? I seriously doubt it. AMD is far far behind Intel. Doubt this will be the revolution they need.

 

[Erenhardt]

Source

AMD to Tape Out First 20nm, 14nm FinFET Chips Within Next Two Quarters.




Source


My thoughts/opinion;
I know that AMD is behind Intel and so are TSMC and GloFO though if AMD and their partners manage to get viable 20nm and even 14nm FinFET yields and production capacity would not that be dangerously close to Intel also we know that Intel has some problems with 14nm because of nature of physics and those chips will be red hot because of density and increased requirements of proper cooling.


I am sure that 14nm FinFET's will be for ARM and maybe MIPS chips from AMD while 20nm will be used for Excavator, so... What kind of performance can we expect from Excavator? On par Haswell or exceed those kind of performances?

Don't know if I should comment on that or just report this and move along

 

[CHADBOGA]

Has a company ever produced an x86 processor that has a 50% jump in IPC over its predecessor?

Sandy Bridge's IPC improvement over Conroe was about 30% IIRC, which is the most dramatic improvement I can think of. The second most dramatic improvement was probably the Athlon 64 X2 over the A64, but I can't find a decent set of benchmarks to compare that easily.

I'm surprised you didn't think of the IPC jump Conroe had over Prescott.