Overclock.net › Forums › Industry News › Hardware News › [xbitlabs] AMD: We Are On Track With Steamroller Micro-Architecture in 2013.
New Posts  All Forums:Forum Nav:

[xbitlabs] AMD: We Are On Track With Steamroller Micro-Architecture in 2013. - Page 39  

post #381 of 718
Quote:
Originally Posted by S.M. View Post

Your overclock comparisons are unfair. You're comparing the highest clocked chip from one with a downclocked chip from the other.

3.0Ghz to 4.0Ghz is 33% (i7-870~)
4.0Ghz to 4.8Ghz is 20% (FX-8350)
3.5Ghz to 4.8Ghz is 37% (i5-3770k)
3.5Ghz to 4.8Ghz is 37% (FX-8320)
3.4Ghz to 4.8Ghz is 41% (i5-3570k)
2.66Ghz to 4.0Ghz is 53% (i7-750)

I think I'm being pretty fair saying Ivy's average overclock on air is 4.8Ghz, grabbed from here.

I think we can agree Bulldozer has the pedigree to clock high.

AMD's approach makes sense to me. Create a smaller, scale-able and modular architecture that matches the old one, get 8 of them on silicon, then from there, any improvements they make to the new architecture is multiplied heavily. I'm excited to see how dedicated decode effects multi-core scaling.

I used that comparison, because that's what started this whole discussion:
http://www.overclock.net/t/1377248/xbitlabs-amd-we-are-on-track-with-steamroller-micro-architecture-in-2013/250_50#post_19670699

I can use my chip as an example too. Its stock clocks are 2.8, and I easily hit 4.2. Again, that's 50%.

Agree with you on the Ivy Bridge percentages, Intel dropped the ball on that one, but due to their slightly higher IPC, they still put out good performance. We'll see come Haswell if they clock to 5 GHZ. Remember Sandy Bridge? Tons of chips hit 5.0.

I wish you didn't mention the WR run, you were on a roll, how many users can realistically hit that? That's just pure hype.

See Alatar's results, he hit 5.6 with Phase. We should stick to discussing realistic attainable results for Air & Water because that's the type of cooling 99.4% of users will be using.

For the record, I also agree Steamroller is the right direction for AMD. wink.gif
Quote:
Originally Posted by malmental View Post

AMD should have never of abandoned Deneb in the first place, it still had some tweaks left in it.
yes new tech is the move but Bulldozer was a downgrade.
Quote:
Originally Posted by 2010rig View Post

Steamroller is the right direction for them IMO. Remember, AMD's goals are HSA oriented, and Deneb wasn't going to get them there.

Quote:
Originally Posted by Brutuz View Post

Because people still think IPC is the one and only measurement of performance that matters, for some reason.
No, it doesn't because it comes at higher clocks stock..You and I both know Intel could easily have their chips run at 4Ghz stock, but you're missing my point...Yes, AMD has significantly lower IPC..But IPC means nothing on its own, to use the logic that it does matter could mean that an ancient chip like those old Cyrix's could compete, etc.

Clock for clock doesn't really have its place...Measure an average and max OC (eg. Say you have a 5Ghz capable i5, measure it at stock, 4.5Ghz and 5Ghz, do the same for AMD even if its max OC is only 4.7Ghz, it's much closer to the real world unlike this which puts a massive focus on IPC, which by itself is as useless as clock speeds.)

And yes, that's why I think SR will be good to compete with Ivy but not Haswell, even if SR beats all expectations and matches Ivy in single-threaded IPC it'll still clock lower and be slower in IPC than Haswell unless you're talking 8 threaded tasks.
This is another reason I say only ever compare synthetic benchmarks if there's utterly no other choice, they don't generally line up with the real world that well in my experience.
As someone who has a lot of experience with Phenom II and FX, I can definitely say that FX was a bit faster in games...Did it make any noticeable difference? Hell no. The last time I saw a big difference in an upgrade was when I upgraded from a Core 2 Duo E8300 to a Phenom II x4 B50 around the same clock speed, before that was from an Athlon XP to a Pentium Dual Core E2180.

Yes, in multi-threaded, best case scenario it may match a 3770K in some cases. Most of our software and games are NOT multi-threaded today. One day, they will be, by that time, Excavator and beyond will be more viable options.

I'm optimistic for Steamroller because it will be a more well rounded chip, not a chip you have to find specific scenarios for where it shines, while making excuses for its weaknesses where the majority of software is utilized today.

Worst case scenario it's a bottleneck even for a single card.



A 3770K is getting 50% more performance out of a single Titan, and it almost always beats the 8350 in multi-threaded. No excuses need to be made for it.

With the 8350 you HAVE to find specific scenarios where it shines, while downplaying its weaknesses, this is where I don't agree with your posts in general.
Quote:
Originally Posted by Brutuz View Post

Nah, there's a reason it's priced at 3570k levels. It sometimes creeps up on an i7 but mainly competes with the i5.

It's what I've been saying for a long time on these forums. AMD isn't pricing an 8350 at $200 because it wants to, but rather because it has to due to its performance. I'm glad we agree on something.
Edited by 2010rig - 4/4/13 at 10:21am
2010rig
(14 items)
 
Galaxy S3
(8 items)
 
 
CPUMotherboardGraphicsRAM
X5660 @ 4.5  ASUS P6X58D-E 980TI? 12GB OCZ Platinum - 7-7-7-21 
Hard DriveCoolingOSMonitor
1 80GB SSD x25m - 3TB F3 + F4 NH-D14 Windows 7 Ultimate LG 47LH55 
KeyboardPowerCaseMouse
Natural Wireless Keyboard Corsair 750HX CM 690 II Advanced MX 518 
CPUGraphicsRAMHard Drive
Snapdragon S4 Dual core 1500mhz Adreno 225 Samsung 2GB 16GB Onboard Flash 
OSMonitorPowerCase
Android 4.4.2 - CM11 4.8" AMOLED 1280x720 2100 mAh battery Otterbox Defender 
  hide details  
2010rig
(14 items)
 
Galaxy S3
(8 items)
 
 
CPUMotherboardGraphicsRAM
X5660 @ 4.5  ASUS P6X58D-E 980TI? 12GB OCZ Platinum - 7-7-7-21 
Hard DriveCoolingOSMonitor
1 80GB SSD x25m - 3TB F3 + F4 NH-D14 Windows 7 Ultimate LG 47LH55 
KeyboardPowerCaseMouse
Natural Wireless Keyboard Corsair 750HX CM 690 II Advanced MX 518 
CPUGraphicsRAMHard Drive
Snapdragon S4 Dual core 1500mhz Adreno 225 Samsung 2GB 16GB Onboard Flash 
OSMonitorPowerCase
Android 4.4.2 - CM11 4.8" AMOLED 1280x720 2100 mAh battery Otterbox Defender 
  hide details  
post #382 of 718
Quote:
Originally Posted by computerparts View Post

Quote:
Originally Posted by Biorganic View Post

Holy crap, everyone knows BD/PD need better IPC, lets not make this about something it's not, ie core addition scaling... rolleyes.gif

Your reply demonstrates the lack of knowledge you have. Maybe you could go onto elaborate exactly which instruction set it is lacking per cycle? And don't go about posting Cinemark or any other Intel biased benches as that's not even relevant. Or you could just admit you have no clue what you're talking about and go and actually do some research before posting.

I am not an electrical engineer but here is my partial understanding of the situation. Bulldozer/PD has a bottleneck in the front end decoder units. The front end decode unit feeds 2 integer cores, but cannot feed both simultaneously, it must alternate on a per clock basis. This is one thing reported to be fixed in SR. IIRC there were memory latency or cache latency issues as well.

The decoder issues should affect highly threaded workloads more so than lightly threaded. Whereas a Cache latency issue would affect all operations. However PD does not have as much issue keeping up with intel solutions in heavily threaded apps, so I don't know where the primary problem is.

It is pretty obvious from various benchmarks that BD/PD underperform relative to the competition, they need to increase IPC to compete with Ivy or Haswell (not likely). PD and Ivy will both clock to around the same frequency, so the chip with better per core IPC in many cases will perform better. On certain highly threaded tasks, the BD arch shines, but not everything is so designed.

Also next time could you be less of an A$$ in your tone. And if you understand it all so well, why don't you explain how to fix BD/PD and whats wrong with them. Thanks
Quote:
Originally Posted by SCollins View Post


I await your in depth analisys of which particular instructions, dispatch, prefetch, cache, pipeline etc areas of the chip, that you feel under perform.

Do you even know how a CPU actually works ?
 
Fat HTPC
(9 items)
 
 
CPUMotherboardGraphicsRAM
2700k Asrock Z77 extreme 4 Sapphire HD 7950 Boost 8 GB gskill ripX 
Hard DriveHard DriveOptical DriveCooling
Samsung F3 WD 500 Blue ASUS 212+ 
OSMonitorMonitorKeyboard
Win7 x64 Prof ASUS AS248H-P LG 21.5 1080p Keys of Boardness 
PowerCaseMouseAudio
XFX 750W XXX Antec 300, fanned out! Death Adder Klipsch Promedia 2.1 
AudioAudio
Xonar Dg Superlux HD 668B 
CPUMotherboardGraphicsRAM
Phenom II 955 ASUS M4A785T-M XFX HD 5670 1 GB Crucial Ballistix 2x2 GB 
Hard DriveCoolingOSPower
WD 500 Green Hyper 212+  Win 7 x64 Prof antec earthwatts 650 
Case
NZXT source 210 
  hide details  
 
Fat HTPC
(9 items)
 
 
CPUMotherboardGraphicsRAM
2700k Asrock Z77 extreme 4 Sapphire HD 7950 Boost 8 GB gskill ripX 
Hard DriveHard DriveOptical DriveCooling
Samsung F3 WD 500 Blue ASUS 212+ 
OSMonitorMonitorKeyboard
Win7 x64 Prof ASUS AS248H-P LG 21.5 1080p Keys of Boardness 
PowerCaseMouseAudio
XFX 750W XXX Antec 300, fanned out! Death Adder Klipsch Promedia 2.1 
AudioAudio
Xonar Dg Superlux HD 668B 
CPUMotherboardGraphicsRAM
Phenom II 955 ASUS M4A785T-M XFX HD 5670 1 GB Crucial Ballistix 2x2 GB 
Hard DriveCoolingOSPower
WD 500 Green Hyper 212+  Win 7 x64 Prof antec earthwatts 650 
Case
NZXT source 210 
  hide details  
post #383 of 718
Still don't believe AMD's claim of 30% ipc improvement with ONE release... that has never been done before? History being made when they are in such crushing downfall? Meh.. I see 5-10% increase. Business as usual.
post #384 of 718
Quote:
Originally Posted by j3st3r View Post

Still don't believe AMD's claim of 30% ipc improvement with ONE release... that has never been done before? History being made when they are in such crushing downfall? Meh.. I see 5-10% increase. Business as usual.

AMD did not said they will have 30% of improvements with Steamroller, Xbit labs did, they speculated... Please read.
post #385 of 718
Quote:
Originally Posted by vampirr View Post

AMD did not said they will have 30% of improvements with Steamroller, Xbit labs did, they speculated... Please read.

"AMD believes that Steamroller will provide 30% improvement in ops per cycle."

Sorry who needs to read? Oh yeah. You.
post #386 of 718
Quote:
Originally Posted by vampirr View Post

Quote:
Originally Posted by j3st3r View Post

Still don't believe AMD's claim of 30% ipc improvement with ONE release... that has never been done before? History being made when they are in such crushing downfall? Meh.. I see 5-10% increase. Business as usual.

AMD did not said they will have 30% of improvements with Steamroller, Xbit labs did, they speculated... Please read.
doh.gif
post #387 of 718
Realistic IB clock is more like 4.5ghz. 4.8ghz and 5ghz is the extreme side of the spectrum, and if you're doing something besides taking the top OC list of max overclock of a giant community, you'll see that it's much lower.

What AMD needs is to close the gap on desktop/server, not to set the world on fire. As long as Steamroller is a bigger improvement over PD than Haswell to IB, then AMD has a win in my book and they're on the right path.

Also, you are forgetting about turbo when comparing overclock %. 3570k does not run single thread benchmarks at 3.5ghz, more like 3.8ghz, which makes a big difference. To go for 3.8ghz turbo single core to 4.5ghz without turbo, it's only about an 18% increase in clock.

So to say that 3570k gets about 37% overclock and that that applies to all workloads is misleading at best. Even if you clock it to 4.8ghz, that's still only a 26% increase in single thread clock speed. 37% is 42% larger than 26%, so that's a pretty big margin to overly-inflate 3570k's single thread when overclocked performance extrapolation.
Tyrant Kuma
(13 items)
 
Starscythe
(13 items)
 
Mobility
(6 items)
 
CPUCPUCPUCPU
Opteron 8431 Opteron 8431 Opteron 8431 Opteron 8431 
MotherboardGraphicsRAMHard Drive
Super Micro h8qme-2+ Sapphire 4870 Samsung DDR2 ECC 667mhz western digital caviar blue 
CoolingCoolingOSPower
FX 8350 stock cooler FX 6300 stock cooler Gentoo Linux 2x Silverstone PSU 
Case
Custom Fabbed Steel Case 
CPUMotherboardGraphicsRAM
AMD FX 8350 gigabyte 990FXA-UD5 7970 Mushkin Enhanced Blackline 16GB 
OSOSMonitorKeyboard
Gentoo Linux Windows 7 x64 Yamakasi Catleap Q270 Corsair K90 
PowerCaseMouseAudio
Silverstone Strider Gold Evolution 850W 80 PLUS... Antec 1200 Logitech G9x Asus Xonar D2X 
Other
XSPC RS360 Raystorm with custom radiator mounts 
CPUGraphicsRAMOS
a4 5000 Radeon HD 8330 8GB G. Skill DDR3 1600 9-9-9 Gentoo Linux 
OSOther
Windows 7 Lenovo Thinkpad x140e 
  hide details  
Tyrant Kuma
(13 items)
 
Starscythe
(13 items)
 
Mobility
(6 items)
 
CPUCPUCPUCPU
Opteron 8431 Opteron 8431 Opteron 8431 Opteron 8431 
MotherboardGraphicsRAMHard Drive
Super Micro h8qme-2+ Sapphire 4870 Samsung DDR2 ECC 667mhz western digital caviar blue 
CoolingCoolingOSPower
FX 8350 stock cooler FX 6300 stock cooler Gentoo Linux 2x Silverstone PSU 
Case
Custom Fabbed Steel Case 
CPUMotherboardGraphicsRAM
AMD FX 8350 gigabyte 990FXA-UD5 7970 Mushkin Enhanced Blackline 16GB 
OSOSMonitorKeyboard
Gentoo Linux Windows 7 x64 Yamakasi Catleap Q270 Corsair K90 
PowerCaseMouseAudio
Silverstone Strider Gold Evolution 850W 80 PLUS... Antec 1200 Logitech G9x Asus Xonar D2X 
Other
XSPC RS360 Raystorm with custom radiator mounts 
CPUGraphicsRAMOS
a4 5000 Radeon HD 8330 8GB G. Skill DDR3 1600 9-9-9 Gentoo Linux 
OSOther
Windows 7 Lenovo Thinkpad x140e 
  hide details  
post #388 of 718
I forget, was there any word on Steamroller having a retooled IMC? Ordinarily those are considered Keller's specialty if I remember correctly, and considering how bandwidth-hungry the APUs are, either Kaveri or the XV APU with a Keller-designed DDR4 IMC would be drool.gif

Not quite as much of an issue on the desktop systems, but it'd still be nice if one of the weaker points of the FX design ATM could get revised.
Claire
(8 items)
 
Asus N56JR
(5 items)
 
Die "Neue" FXe
(11 items)
 
CPUMotherboardGraphicsRAM
AMD Ryzen 5 1600 ASRock X370 Taichi MSI Radeon R9 390 GAMING 8G G.Skill Flare X 3200 MHz CL14 
Hard DriveOSPowerCase
Samsung 960 EVO 500 GB Solus Seasonic SSR-550RM NZXT S340 Elite Matte White 
CPUGraphicsHard DriveOS
Intel Core i7-4700HQ GeForce GTX 760M Samsung 840 EVO 250 GB Microsoft Windows 10 Education 
Other
Intel Wireless-AC 7260 
CPUMotherboardGraphicsRAM
AMD FX-8120 Asus SABERTOOTH 990FX MSI Radeon R9 390 GAMING 8G G. Skill Ripjaws (DDR3-1600) 
Hard DriveOptical DriveCoolingOS
Samsung 840 EVO 250 Asus DRW-21B1ST Cooler Master Hyper 212 Evo Microsoft Windows 7 Home Premium 
PowerCaseAudio
Cooler Master RS-600-AMBA-D3 Cooler Master CM690 II Advanced Asus Xonar DG 
  hide details  
Claire
(8 items)
 
Asus N56JR
(5 items)
 
Die "Neue" FXe
(11 items)
 
CPUMotherboardGraphicsRAM
AMD Ryzen 5 1600 ASRock X370 Taichi MSI Radeon R9 390 GAMING 8G G.Skill Flare X 3200 MHz CL14 
Hard DriveOSPowerCase
Samsung 960 EVO 500 GB Solus Seasonic SSR-550RM NZXT S340 Elite Matte White 
CPUGraphicsHard DriveOS
Intel Core i7-4700HQ GeForce GTX 760M Samsung 840 EVO 250 GB Microsoft Windows 10 Education 
Other
Intel Wireless-AC 7260 
CPUMotherboardGraphicsRAM
AMD FX-8120 Asus SABERTOOTH 990FX MSI Radeon R9 390 GAMING 8G G. Skill Ripjaws (DDR3-1600) 
Hard DriveOptical DriveCoolingOS
Samsung 840 EVO 250 Asus DRW-21B1ST Cooler Master Hyper 212 Evo Microsoft Windows 7 Home Premium 
PowerCaseAudio
Cooler Master RS-600-AMBA-D3 Cooler Master CM690 II Advanced Asus Xonar DG 
  hide details  
post #389 of 718
Quote:
Originally Posted by j3st3r View Post

"AMD believes that Steamroller will provide 30% improvement in ops per cycle."

Sorry who needs to read? Oh yeah. You.

It's worth pointing out that the 30% improvement claim was with regards to what the instruction decoders are capable of decoding. That isn't the same thing as saying each core is going to be able to complete 30% more instructions every second. Not every issue is a matter of the cores being idle when they could be chewing on more instructions.
Edited by Particle - 4/4/13 at 11:56am
post #390 of 718
can't amd come up with a name that doesn't somehow seem to reference pro-wrestling or construction? granted intels seem to be typically pretty abstract and/or meaningless, but they're still preferable
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
This thread is locked  
Overclock.net › Forums › Industry News › Hardware News › [xbitlabs] AMD: We Are On Track With Steamroller Micro-Architecture in 2013.