"Steamroller" could crush "Haswell"?!

[chernobog]

Here are links containing various data/information's:

http://www.xbitlabs.com/news/cpu/di...h_Steamroller_Micro_Architecture_in_2013.html

http://www.tomshardware.com/news/AMD-Kaveri-APU-Gaming,22947.html

http://forums.anandtech.com/showthread.php?t=2322010

http://www.brightsideofnews.com/new...core-architectural-enhancements-unveiled.aspx

It seems "Steamroller" will be a real threat to Intel's Haswell and Intel coud make a run for their money, if that's so then I can't wait for an outcome.

So if this all applies then we are looking at these results:

FX 8350 "Piledriver" cinebench 11.5 multi-threaded: 6.89pts (result from anandtech)

i7 4770k "Haswell" cinebench 11.5 multi-threaded 8.07pts (result from anandtech)

FX 8550 "Steamroller" cinebench 11.5 multi-threaded: 9.92pts(estimation) - 4Ghz

FX 8550 "Steamroller" cinebench 11.5 multi-threaded: 11.90pts(estimation) - 5Ghz

Please open the links and read the whole articles, thank you.

Alright, the point has been made and then driven into the ground. This has gone on long enough.
-ViRGE

 

[ViRGE]

Just curious, where are the Cinebench estimates coming from? I'm not seeing anything regarding Cinebench in any of the linked articles.

 

[chernobog]

Just curious, where are the Cinebench estimates coming from? I'm not seeing anything regarding Cinebench in any of the linked articles.

Well Virge, I looked at cinebench results from various sites and then I look at Piledriver FX 8350 results and with all improvements that Steamroller introduces I calculated.

Steamroller will introduce a fix to multi-threaded performance thus removing penalty of 20% and then I added IPC/IPO improvements of 15 to 20% from Kaveri and thus I got these results.

 

[SithSolo1]

Well Virge, I looked at cinebench results from various sites and then I look at Piledriver FX 8350 results and with all improvements that Steamroller introduces I calculated.

Steamroller will introduce a fix to multi-threaded performance thus removing penalty of 20% and then I added IPC/IPO improvements of 15 to 20% from Kaveri and thus I got these results.

So basically: Pixie dust

 

[chernobog]

So basically: Pixie dust

Its a "pixie dust" ultil Steamroller FX and benchmarks of it arrives/show up.

FX 8350 cinebench 11.5 multi-threaded is 6.89 points

Steamroller fixing multi-thread penalty = 20%+ multi-threaded

6.89+1.378 = 8.268

Kaveri(Steamroller cores) 15-20%+ IPC/IPO single/multi-threaded

8.268+1.6536(20%)/1.2402(15%)

Result:
9.9216 or 9.5082 pts - Cinebench 11.5 Multi-Threaded FX 8550 aprox.

Simple math...

 

[3DVagabond]

Its a "pixie dust" ultil Steamroller FX and benchmarks of it arrives/show up.

FX 8350 cinebench 11.5 multi-threaded is 6.89 points

Steamroller fixing multi-thread penalty = 20%+ multi-threaded

6.89+1.378 = 8.268

Kaveri(Steamroller cores) 15-20%+ IPC/IPO single/multi-threaded

8.268+1.6536(20%)/1.2402(15%)

Result:
9.9216 or 9.5082 pts - Cinebench 11.5 Multi-Threaded FX 8550 aprox.

Simple math...

You are basing all of this on marketing claims though. Not a typically very reliable source.

 

[Sable]

Its a "pixie dust" ultil Steamroller FX and benchmarks of it arrives/show up.

FX 8350 cinebench 11.5 multi-threaded is 6.89 points

Steamroller fixing multi-thread penalty = 20%+ multi-threaded

6.89+1.378 = 8.268

Kaveri(Steamroller cores) 15-20%+ IPC/IPO single/multi-threaded

8.268+1.6536(20%)/1.2402(15%)

Result:
9.9216 or 9.5082 pts - Cinebench 11.5 Multi-Threaded FX 8550 aprox.

Simple math...

Well that's all the proof I need.

 

[ShintaiDK]

I think someone forgot to check AMDs roadmaps besides making up pixie dust. 2M/4T for desktop only with APUs. And servers will keep using Pilediver.

 

[chernobog]

You are basing all of this on marketing claims though. Not a typically very reliable source.

Even if all improvements are just put FX 8550 "just" 20% above FX 8350, it will be still slightly faster than i7 4770k, this is just all estimation in case if all those promises or estimations from AMD become true.

FX 8350 "Piledriver" 4Ghz 6.89pts (anandtech bench)

FX 8550 "Steamroller" 4Ghz 9.9216/9.5082 pts

FX 8550 "Steamroller" 4.8Ghz 11.90592/11.41434 pts
(20%+ in clocks/20% gains)

i7 4770k "Haswell" 3.5Ghz 8.07pts (anandtech bench)

i7 4770k "Haswell" 4Ghz 9.2805pts (estimation)

@ShintaiDK

"I think someone forgot to check AMDs roadmaps besides making up pixie dust. 2M/4T for desktop only with APUs. And servers will keep using Pilediver."

I am sorry to spoil your day, but I did not forgot to check AMD's latest server roadmap and server roadmap does not indicate desktop roadmap my pal. I know that AMD's FX series are coming from Opteron chips but it does not indicate that Steamroller is delayed or canceled at all.

Its called estimation, it would be "pixie dust" if someone claimed that an Pentium 3 is faster than Core 2 Duo.

There is a 4 core Kaveri(Steamroller) APU and 12-16 core Piledriver+ server chips but there is a gap, 6-8 core server chips, so do you see an 6-8 core server chip on AMD's latest server roadmap? No, ok.

Only 8 core server chip from AMD is ARM64 "Seattle", we are talking about x86-64 chips not ARM based.

 

[3DVagabond]

Even if all improvements are just put FX 8550 "just" 20% above FX 8350, it will be still slightly faster than i7 4770k, this is just all estimation in case if all those promises or estimations from AMD become true.

FX 8350 "Piledriver" 4Ghz 6.89pts (anandtech bench)

FX 8550 "Steamroller" 4Ghz 9.9216/9.5082 pts

FX 8550 "Steamroller" 4.8Ghz 11.90592/11.41434 pts
(20%+ in clocks/20% gains)

i7 4770k "Haswell" 3.5Ghz 8.07pts (anandtech bench)

i7 4770k "Haswell" 4Ghz 9.2805pts (estimation)

I saw where you got your numbers. I highly doubt the 20% IPC improvement. I also wouldn't bank on a 20% increase in clocks. 20% overall, including clock increase would be a more likely scenario, IMO.

BTW, a 20% improvement over 6.89 is 8.27. Cinebench does scale really well with cores and clocks, way better than typical applications, but even Cinebench isn't 100%.

 

[chernobog]

I know that, this is just rough estimation even though theoretically it should apply while reality it is 25% higher clocks equals 20% gains. Right?

I am looking at my response to ShintaiDK and it seems I nailed it in the middle like a Quake 3 guy with a rail gun.

 

[guskline]

chernobog, do you know galego?

Callouts are not acceptable
-ViRGE

 

[chernobog]

chernobog, do you know galego?

I do not know him, I wonder why you ask but its off topic.

Lets not continue this so we don't derail the threads topic, you should have PM me.

I wonder what idontcare would say... hmmm..

 

[Sheep221]

Even if all improvements are just put FX 8550 "just" 20% above FX 8350, it will be still slightly faster than i7 4770k, this is just all estimation in case if all those promises or estimations from AMD become true.

FX 8350 "Piledriver" 4Ghz 6.89pts (anandtech bench)

FX 8550 "Steamroller" 4Ghz 9.9216/9.5082 pts

FX 8550 "Steamroller" 4.8Ghz 11.90592/11.41434 pts
(20%+ in clocks/20% gains)

i7 4770k "Haswell" 3.5Ghz 8.07pts (anandtech bench)

i7 4770k "Haswell" 4Ghz 9.2805pts (estimation)

@ShintaiDK

"I think someone forgot to check AMDs roadmaps besides making up pixie dust. 2M/4T for desktop only with APUs. And servers will keep using Pilediver."

I am sorry to spoil your day, but I did not forgot to check AMD's latest server roadmap and server roadmap does not indicate desktop roadmap my pal. I know that AMD's FX series are coming from Opteron chips but it does not indicate that Steamroller is delayed or canceled at all.

Its called estimation, it would be "pixie dust" if someone claimed that an Pentium 3 is faster than Core 2 Duo.

There is a 4 core Kaveri(Steamroller) APU and 12-16 core Piledriver+ server chips but there is a gap, 6-8 core server chips, so do you see an 6-8 core server chip on AMD's latest server roadmap? No, ok.

Only 8 core server chip from AMD is ARM64 "Seattle", we are talking about x86-64 chips.

I seriously doubt your estimations are going to hit the reality. And if even AMD will take the performance crown for a while, it won't still endanger Intel technically.
AMD was leading in enthusiast CPUs since cca 2000 till late 2006(estimate), but Intel still had 10 times more revenue and sales anyway, and once the Intel took over with Core2, the AMD began to fall very fast, compared to Intel, which held same stable position on the market for many years when the AMD sold better CPUs.
AMD made good stuff before, they probably will make good CPUs sometime in the future again, but they never overtook Intel and technically won't be able to do so in many years to come.

 

[chernobog]

I seriously doubt your estimations are going to hit the reality. And if even AMD will take the performance crown for a while, it won't still endanger Intel technically.
AMD was leading in enthusiast CPUs since cca 2000 till late 2006(estimate), but Intel still had 10 times more revenue and sales anyway, and once the Intel took over with Core2, the AMD began to fall very fast, compared to Intel, which held same stable position on the market for many years when the AMD sold better CPUs.
AMD made good stuff before, they probably will make good CPUs sometime in the future again, but they never overtook Intel and technically won't be able to do so in many years to come.

AMD was previously lead by two engineers that did not have any business sense, most executions were bad and partially thankful to its fabrications and later to Global Foundries and TSMC.

AMD was tehnically superior to Intel circa 1999 to 2006, first with Athlon then Athlon 64(AMD64 that Intel later reverse engineered) and then a dual core processor. Ultil Core 2 Duo arrived that was based/derived from ancient Pentium 3 and thus all current Intel processors are practically heavily modified Pentium 3's

AMD will overtook performance crown sooner or later.

AMD was always a smaller company to Intel and it will still be also do not forget that Intel played dirty everytime it could not won/beat the competition. It intentionally sabotaged AMD. Now Intel is doing a similar thing in india as of recent.

People, please do not derail the thread if its not directly involved with the theme of my thread.

 

[Durp]

chernobog, do you know galego?

I was going to say the same thing.

It's best you don't
-ViRGE

 

[frozentundra123456]

I do not know him, I wonder why you ask but its off topic.

Lets not continue this so we don't derail the threads topic, you should have PM me.

I wonder what idontcare would say... hmmm..

Don't really understand sarcasm do you?

 

[chernobog]

Moderators, can you delete response 16 and 17, why? Offtopic.

Thank you, and delete this response also. I want a clean thread.

I understand sarcasm but I don't know if Galego is an member or its a thing, something other.

 

[Erenhardt]

Don't really understand sarcasm do you?

As someone who lost both of his legs i feel offended!

 

[Idontcare]

FX 8550 "Steamroller" cinebench 11.5 multi-threaded: 11.79pts(estimation) - 5Ghz

Please open the links and read the whole articles, thank you.

What do you estimate the single-thread pts will be?

 

[guskline]

Look, you started a thread captioned Steamroller could " crush" Haswell putting the "crush" in quotations, not me. Moteover you make this claim based upon AMD marketing claims without much substantiation. My question is a valid one, NOT a personal attack.

 

[postmortemIA]

I *could* crush Chuck Norris too, just could...

 

[chernobog]

@guskline

This is all just a theory, estimation in case all of these is true and we are talking about "Steamroller" and not a "Piledriver" refesh. Steamroller is a 3rd Gen Bulldozer architecture and as some people even here on anandtech have discussed about Steamroller and alleged leak of the shot depicting Steamroller core and agreed that it could give a substantially better performance over Piledriver.

@idontcare

My estimation is fairly sane, Steamroller single threaded at 4Ghz in cinebench 11.5 should score around 1.20 to 1.25 points and at 5Ghz it should score 1.40 to 1.50.

Its all just rough estimation, some people are seriously taking my estimations as a claim that it will happen rather than just a theory and even derailing the thread!

@postmortemIA

Bruce Lee crushed/killed him or at least defeated him. Please, everyone. Don't derail the thread.

 

[SammichPG]

I was going to say the same thing.

Too much stuff for his sign, he doesn't have enough space.

 

[jfelano]

LOL

So it 'may' beat Haswell, in one benchmark, based on your calculations, wow.