AMD "Steamroller/Excavator" -info, speculations and experience

[Evantaur]

dropping the clock from 4 to 3 GHz is a steep drop. Seems there is a lot changing under the hood this time around....

it's actually haswell with radeon graphics on it

sent from future: April 1st 2014

[tbone8ty]

"The power savings comes from not having to route clocks and signals as far, while the area savings are a result of the computer automated transistor placement/routing and higher density gate/logic libraries. The tradeoff is peak frequency. These heavily automated designs won?t be able to clock as high as the older hand drawn designs. AMD believes the sacrifice is worth it however because in power constrained environments (e.g. a notebook) you won?t hit max frequency regardless, and you?ll instead see a 15 - 30% energy reduction per operation. AMD equates this with the power savings you?d get from a full process node improvement."

http://www.anandtech.com/show/6201/a...r-architecture

[Fottemberg]

Hi guys!

About the frequency of the GPU, we have to evaluate that a Radeon HD7750 (512SP, 8CU, GCN) generates 920 GFlops @900 MHz (SP). Also, I know that actually Kaveri ESamples generate a similiar amount of GFlops at the same frequency. So Kaveri will have the GPU at 1GHz (or more) to reach the goal of 1050 GFlops (as AMD slides reported one year ago).

[xVeinx]

"The power savings comes from not having to route clocks and signals as far, while the area savings are a result of the computer automated transistor placement/routing and higher density gate/logic libraries. The tradeoff is peak frequency. These heavily automated designs won?t be able to clock as high as the older hand drawn designs. AMD believes the sacrifice is worth it however because in power constrained environments (e.g. a notebook) you won?t hit max frequency regardless, and you?ll instead see a 15 - 30% energy reduction per operation. AMD equates this with the power savings you?d get from a full process node improvement."

http://www.anandtech.com/show/6201/a...r-architecture

Except those are in use for excavator, not steamroller .

[tbone8ty]

seems like it would be worth it for Amd to release a steamroller cpu for all the am3+ users, even with out an updated chipset. potential for the whole user base to upgrade.

but then again i don't seem them doing it without having an updated chipset? hmm i dunno.

i for one would upgrade my aging am3+ mobo with a steamroller FX chip.

[xVeinx]

Not sure. They have a lot of older processors/stock floating about and I'm not sure how much income it would produce relative to the costs of production/chipset/etc.. For servers, they'll use the C32/G34 socket, so it isn't simply a matter of repackaging an Opteron as an FX. If they are looking to consolidate lines, now may be the time to do it. I understand the sentiment completely; my only question is whether or not it really benefits AMD in the long run.

[Johan45]

seems like it would be worth it for Amd to release a steamroller cpu for all the am3+ users, even with out an updated chipset. potential for the whole user base to upgrade.

but then again i don't seem them doing it without having an updated chipset? hmm i dunno.

i for one would upgrade my aging am3+ mobo with a steamroller FX chip.

Me too, would absolutely love another CPU line to drop in my system but as xVeinx said it'll all come down to the money and AMD doesn't have a lot of that kicking around at the moment.

[tbone8ty]

I just wanna see how a 28nm FX chip performs! Gotta be some power savings!

[vario]

Doing a 4 to 3ghz .Hum.
Is AMD doing a "conroe" ? Jeez it would be nice.

Stilt , i know you cant say anything of importance.But is your comment "hammer time" in expectation of kaveris performance ? Wil it Athlonize(K7) the market ? :P

[Astennu]

Hi guys!

About the frequency of the GPU, we have to evaluate that a Radeon HD7750 (512SP, 8CU, GCN) generates 920 GFlops @900 MHz (SP). Also, I know that actually Kaveri ESamples generate a similiar amount of GFlops at the same frequency. So Kaveri will have the GPU at 1GHz (or more) to reach the goal of 1050 GFlops (as AMD slides reported one year ago).

Could be but 1GHz compared to the 800/850 we have in current APU's is quite high.
Maybe a part of the GFlops are comming from the newer GCN architecture?

Doing a 4 to 3ghz .Hum.
Is AMD doing a "conroe" ? Jeez it would be nice.

Stilt , i know you cant say anything of importance.But is your comment "hammer time" in expectation of kaveris performance ? Wil it Athlonize(K7) the market ? :P

Could be that they shortened the Pipleline again.
Bulldozer's pipline was quite a bit longer compared to the K10,5 Architecture (Deneb and Thuban cores).

Lower clocks could also help to get more performance per watt. And CPU's for tablets and Laptops are more important then every. The Desktop market is getting smaller and smaller.

Top performance might go down a bit. But if the performance per watt is similar to Intel they can still be winners in the mobile space.

[EniGmA1987]

Doing a 4 to 3ghz .Hum.
Is AMD doing a "conroe" ? Jeez it would be nice.

There is already a Kaveri ES sample we have seen with a 3.5GHz base speed, so the rumor that these will top out around 3GHz is quite obviously wrong.

Here are the pics:

[FlanK3r]

seems, you have right. IM looking forward for some benchmarks.

[The Stilt]

The model code on ES parts doesn't always specify the clock frequency (usually it does thou).
For example ES Trinity parts with 4500 model number have base frequency of 1900MHz.

[SUPERKAMES]

There is already a Kaveri ES sample we have seen with a 3.5GHz base speed, so the rumor that these will top out around 3GHz is quite obviously wrong.

So A10-7xx0K may have max boost frequency around 4ghz.

[tbone8ty]

Should probably find out by Wednesday

[tbone8ty]

[SUPERKAMES]

New slide.

http://www.pcworld.com/article/20624...n-january.html
http://www.anandtech.com/show/7507/a...p-january-14th

Second of all, we have confirmation on what the highest shipping APU configuration will be. Kaveri will have up to 4 CPU core (2 modules), which will be based on AMD’s latest revision of their desktop CPU architecture, Steamroller. Meanwhile the GPU will be composed of 8 GCN 1.1 CUs, which would put the SP count at 512 SPs (this would be equivalent to today's desktop Radeon HD 7750). Furthermore AMD is throwing around a floating point performance number – 856 GFLOPS – which thanks to some details found in AMD's footnotes by PCWorld gives us specific clockspeeds and even a product name. A10-7850K CPU clockspeed 3.7GHz, GPU clockspeed 720MHz.

[G.Foyle]

FYI: 3.7 GHz is base clock, not turbo.

[BeepBeep2]

Will be interesting how the HUMA Memory Controller affects performance in different situations...

[tbone8ty]

I thought it was suppose to hit 1 Tflops?

Still cool to see kaveri finally emerging!

Huma will be interesting

[informal]

They didn't hit 1TF as it seems they have missed the target GPU clock range. If it was ~900Mhz the 1TF would have been achieved. That's a good 25% lower clock speed, I hope the GPU has some Turbo that can clock up at least 10% more.

[FlanK3r]

my tip is 7850K will be 3.7GHz/4.1 GHz (turbo) and OC around 4500-4600 MHz stable max. Next tips: Around 10% better than A10-6800K (no clocks to clocks), single tread around 5%+, multi 10-17%+

[informal]

my tip is 7850K will be 3.7GHz/4.1 GHz (turbo) and OC around 4500-4600 MHz stable max. Next tips: Around 10% better than A10-6800K (no clocks to clocks), single tread around 5%+, multi 10-17%+

My prediction.
Stock vs stock, 7850K (3.7/4.1Ghz) vs 6800K (4.1/4.4Ghz)
In ST: ~10-12% better performance on average. IPC ~20% higher.
In MT: ~22-25% better performance on average. "IPC" ~35% higher. I put it in quotation marks since IPC is really "only" 20% higher like in ST case but this time we don't have the ~10-15% penalty from shared decoder. I took mean number of ~13% and got to: 1.2x1.13=1.35 or 35% higher throughput.

[tbone8ty]

Power efficiency will be better though

[justapost]

Seen this Kaveria 3.5GHz ES results over at Overclock.net http://browser.primatelabs.com/geekbench3/223722. As far as i remember geekbench tends to vary alot between run's so it's not that usefull. Memory speed has an huge impact on the overall score aswell. Anyway found this A8-6500 (3.5GHz) results with similar memory results but on an different os. http://browser.primatelabs.com/geekb...aseline=223722