AMD CPU speculation... and expert conjecture

[juanrga]

Wasn't Basilisk an early code name for Kaveri?

No.

iirc basilisk was supposed to come out after kaveri. but the exc. based apus were supposed to be called carrizo, which was "leaked" earler than basilisk's codename. i don't know about basilisk's specs.

Kaveri, its sucessor, and Basilisk are different projects.

Worked in the APU Emulation group on the Single Emulation/Simulation Work Area and Build Flow project for the Kaveri, Carrizo and Basilisk APUs.

 

[juanrga]

Especially that 8350rocks said Carrizo is mobile only.

We will see Carrizo beyond mobile. In fact, AMD has already announced officially that Carrizo will be used in some "All-in-One systems".

http://www.youtube.com/watch?v=I6zX2IqBI7A

What the hell? Carrizo and Carrizo-L in one package? Is it some kind of big.LITTLE?

It couldn't be more far from big.LITTLE!!!!!

AMD is sharing the same platform (FP4) and same packaging (BGA) for reducing costs, this is a kind of pre-Skybridge approach. Nowhere AMD said that both small cores and big cores will be on the same CPU.

 

[logainofhades]

More price cuts to current APUs.. this is really indicative of a nearing launch http://www.kitguru.net/components/cpu/anton-shilov/amd-to-drop-prices-of-a-series-kaveri-apus-to-discontinue-a10-7700k/

The 7700k seemed kind of pointless, and does that man excavator is on its way then?

 

[szatkus]

Especially that 8350rocks said Carrizo is mobile only.

We will see Carrizo beyond mobile. In fact AMD has already announced officially that Carrizo will be used in some "All-in-One systems".

Guess which one of you I belive. AIOs with mobile parts aren't uncommon.

http://www.youtube.com/watch?v=I6zX2IqBI7A

What the hell? Carrizo and Carrizo-L in one package? Is it some kind of big.LITTLE?

No. AMD is sharing the same platform and same packaging for reducing costs, this is a kind of pre-Slybridge approach. Nowhere AMD said that both small cores and big cores will be on the same CPU.[/quotemsg]

You're right. I understood "same package" wrong

http://www.hardware.fr/medias/photos_news/00/45/IMG0045881_1.jpg
http://www.hardware.fr/medias/photos_news/00/45/IMG0045882_1.jpg.

 

[logainofhades]

The hospital I used to work at, had AIO's with Atoms, so Carrizo in an all in one wouldn't surprise me in the least.

 

[Reepca]

FX 8350 is doing horrible in far cry 4 competing with an I3 and Amd is performing better on a Nvidia game?
http://www.techspot.com/review/917-far-cry-4-benchmarks/page6.html

What I find most interesting is in page 5: Both the i7-4770k and FX-8350 gain performance almost linearly with clockspeed. This indicates, even OCd to 4.5GHz, the game is still being limited to some extent by the CPU. The Intel chips are clustered together less because of a GPU bottleneck, and more because all their chips offer comparable performance.

What's also interesting is the rate of increase. The i7 is getting about 3 FPS per 500MHz. The FX is getting about 2 FPS per 500MHz. That's the IPC difference between the two showing up again, and accounts for all of Intels performance lead.

In the end, it looks like two strong cores simply outperform 8 weak ones. I'd be interested to see latency graphs, as the i3 is likely a lot less stable performance wise, but it's quite clear where Cry Engine 3's performance characteristics lie at this point.

It's bizarre - from what I've read, in most multi-threaded benchmarks the a8-7600 and i3-4330 (I think I got the number right) share around the same compute capability (usually within 10%), so would it be reasonable to infer that an FX-8350 being beaten by it probably means the game doesn't scale well off multiple cores? But then again in the consoles aren't they programming for 8 (light) cores?

I'm sure there's an ubisoft-blaming answer that satisfies the demands of reason somewhere here...

 

[szatkus]

FX 8350 is doing horrible in far cry 4 competing with an I3 and Amd is performing better on a Nvidia game?
http://www.techspot.com/review/917-far-cry-4-benchmarks/page6.html

What I find most interesting is in page 5: Both the i7-4770k and FX-8350 gain performance almost linearly with clockspeed. This indicates, even OCd to 4.5GHz, the game is still being limited to some extent by the CPU. The Intel chips are clustered together less because of a GPU bottleneck, and more because all their chips offer comparable performance.

What's also interesting is the rate of increase. The i7 is getting about 3 FPS per 500MHz. The FX is getting about 2 FPS per 500MHz. That's the IPC difference between the two showing up again, and accounts for all of Intels performance lead.

In the end, it looks like two strong cores simply outperform 8 weak ones. I'd be interested to see latency graphs, as the i3 is likely a lot less stable performance wise, but it's quite clear where Cry Engine 3's performance characteristics lie at this point.

It's bizarre - from what I've read, in most multi-threaded benchmarks the a8-7600 and i3-4330 (I think I got the number right) share around the same compute capability (usually within 10%), so would it be reasonable to infer that an FX-8350 being beaten by it probably means the game doesn't scale well off multiple cores? But then again in the consoles aren't they programming for 8 (light) cores?

I'm sure there's an ubisoft-blaming answer that satisfies the demands of reason somewhere here...

Games on consoles can use 6 cores. There's no evidence that FC4 uses all of them on consoles. Actually threads structure are usually the same on consoles and PC, that's why most games from past-gens use scale to only 3-4 cores (with drivers and system overhead).

 

[blackkstar]

FX 8350 is doing horrible in far cry 4 competing with an I3 and Amd is performing better on a Nvidia game?
http://www.techspot.com/review/917-far-cry-4-benchmarks/page6.html

What I find most interesting is in page 5: Both the i7-4770k and FX-8350 gain performance almost linearly with clockspeed. This indicates, even OCd to 4.5GHz, the game is still being limited to some extent by the CPU. The Intel chips are clustered together less because of a GPU bottleneck, and more because all their chips offer comparable performance.

What's also interesting is the rate of increase. The i7 is getting about 3 FPS per 500MHz. The FX is getting about 2 FPS per 500MHz. That's the IPC difference between the two showing up again, and accounts for all of Intels performance lead.

In the end, it looks like two strong cores simply outperform 8 weak ones. I'd be interested to see latency graphs, as the i3 is likely a lot less stable performance wise, but it's quite clear where Cry Engine 3's performance characteristics lie at this point.

It's bizarre - from what I've read, in most multi-threaded benchmarks the a8-7600 and i3-4330 (I think I got the number right) share around the same compute capability (usually within 10%), so would it be reasonable to infer that an FX-8350 being beaten by it probably means the game doesn't scale well off multiple cores? But then again in the consoles aren't they programming for 8 (light) cores?

I'm sure there's an ubisoft-blaming answer that satisfies the demands of reason somewhere here...

Games on consoles can use 6 cores. There's no evidence that FC4 uses all of them on consoles. Actually threads structure are usually the same on consoles and PC, that's why most games from past-gens use scale to only 3-4 cores (with drivers and system overhead).

Look at the Intel results too. There's 5 fps difference between i3 3220 and 3930k, yet 13fps difference between 2.5ghz 4770k and 4.5ghz 4770k. 3770k and 4690x is 2fps, within margin of error and 4690x has 100 extra mhz.

I would have liked to see Pentium G results just to see core scaling on a dual core without HT, and the unlocked one would be nice to see frequency scaling.

This looks like the only CPU benchmark so far that I can find. Reading reviews it looks like Ubisoft didn't do much to favor AMD hardware at all.

 

[juanrga]

The hospital I used to work at, had AIO's with Atoms, so Carrizo in an all in one wouldn't surprise me in the least.

The announced AIO will use the 35W (FP4/BGA) Carrizo variants.

 

[juanrga]

After announcing that will use HSA-enabled APUs for future supercomputers, AMD announces now that will use HSA-enabled APUs for future workstations as well

 

[gamerk316]

FX 8350 is doing horrible in far cry 4 competing with an I3 and Amd is performing better on a Nvidia game?
http://www.techspot.com/review/917-far-cry-4-benchmarks/page6.html

What I find most interesting is in page 5: Both the i7-4770k and FX-8350 gain performance almost linearly with clockspeed. This indicates, even OCd to 4.5GHz, the game is still being limited to some extent by the CPU. The Intel chips are clustered together less because of a GPU bottleneck, and more because all their chips offer comparable performance.

What's also interesting is the rate of increase. The i7 is getting about 3 FPS per 500MHz. The FX is getting about 2 FPS per 500MHz. That's the IPC difference between the two showing up again, and accounts for all of Intels performance lead.

In the end, it looks like two strong cores simply outperform 8 weak ones. I'd be interested to see latency graphs, as the i3 is likely a lot less stable performance wise, but it's quite clear where Cry Engine 3's performance characteristics lie at this point.

It's bizarre - from what I've read, in most multi-threaded benchmarks the a8-7600 and i3-4330 (I think I got the number right) share around the same compute capability (usually within 10%), so would it be reasonable to infer that an FX-8350 being beaten by it probably means the game doesn't scale well off multiple cores? But then again in the consoles aren't they programming for 8 (light) cores?

I'm sure there's an ubisoft-blaming answer that satisfies the demands of reason somewhere here...

Games on consoles can use 6 cores. There's no evidence that FC4 uses all of them on consoles. Actually threads structure are usually the same on consoles and PC, that's why most games from past-gens use scale to only 3-4 cores (with drivers and system overhead).

Look at the Intel results too. There's 5 fps difference between i3 3220 and 3930k, yet 13fps difference between 2.5ghz 4770k and 4.5ghz 4770k. 3770k and 4690x is 2fps, within margin of error and 4690x has 100 extra mhz.

I would have liked to see Pentium G results just to see core scaling on a dual core without HT, and the unlocked one would be nice to see frequency scaling.

This looks like the only CPU benchmark so far that I can find. Reading reviews it looks like Ubisoft didn't do much to favor AMD hardware at all.

I'd assume scaling is similar to Crysis 3, which had two heavy threads and about 12 smaller rendering threads account for most of the work, I'd need to see someone look at a GPUView analysis or something though, since I have no plans to personally buy the game. It is quite possible though Ubi just did a really shoddy coding job, which given AC:U, may be the simpler, but correct, explanation.

To me, these results simply are the result of the Intel/AMD IPC difference.

 

[8350rocks]

This very much mirrors the performance envelope MWO is getting from Cry3. Their engine runs on, at most, 4 cores. I think they have done what PGI did by making an attempt to make cry3 better on more entry systems. Of course, ubi is horrible at optimizing, so take that as you will. However, in cry3 the game scaled well to 8 cores, beyond that, loads were not equal enough and you had 2 cores bottlenecking while the rest were barely loaded. Where as on 6-8 core setups most cores loaded at least 60-70%. This explains why overclocking yields so well. The 2 main threads, even on cryengine3 are still very heavy. The engine threads well enough, but if you cut back on the threading it will generate stronger single thread tendencies than the engine originally had to begin with. Interesting, nonetheless.

 

[cdrkf]

This very much mirrors the performance envelope MWO is getting from Cry3. Their engine runs on, at most, 4 cores. I think they have done what PGI did by making an attempt to make cry3 better on more entry systems. Of course, ubi is horrible at optimizing, so take that as you will. However, in cry3 the game scaled well to 8 cores, beyond that, loads were not equal enough and you had 2 cores bottlenecking while the rest were barely loaded. Where as on 6-8 core setups most cores loaded at least 60-70%. This explains why overclocking yields so well. The 2 main threads, even on cryengine3 are still very heavy. The engine threads well enough, but if you cut back on the threading it will generate stronger single thread tendencies than the engine originally had to begin with. Interesting, nonetheless.

I think these are inherent limitations imposed by using the current API (i.e. DX11 or current Open GL). Both API's have the ability for quite a bit of multi-threading, however they both rely on a small number of 'main threads' that can do the big stuff like issue draw calls. From my (limited) understanding of all this- Mantle as well as the upcoming DX12 and OGL-Next should remove this limitation.

 

[-Fran-]

This very much mirrors the performance envelope MWO is getting from Cry3. Their engine runs on, at most, 4 cores. I think they have done what PGI did by making an attempt to make cry3 better on more entry systems. Of course, ubi is horrible at optimizing, so take that as you will. However, in cry3 the game scaled well to 8 cores, beyond that, loads were not equal enough and you had 2 cores bottlenecking while the rest were barely loaded. Where as on 6-8 core setups most cores loaded at least 60-70%. This explains why overclocking yields so well. The 2 main threads, even on cryengine3 are still very heavy. The engine threads well enough, but if you cut back on the threading it will generate stronger single thread tendencies than the engine originally had to begin with. Interesting, nonetheless.

I think these are inherent limitations imposed by using the current API (i.e. DX11 or current Open GL). Both API's have the ability for quite a bit of multi-threading, however they both rely on a small number of 'main threads' that can do the big stuff like issue draw calls. From my (limited) understanding of all this- Mantle as well as the upcoming DX12 and OGL-Next should remove this limitation.

And since an available API that can alleviate that "pain" exists, but they chose not to use and instead went with Gameworks, makes you say out loud, but to yourself: "damn you, Ubicrap". Hahaha.

I really doubt Ubicrap will be willing to re-work these titles under DX12 now that they're out. Makes no business sense. At most, announce MANTLE compatibility at some distant point when no one remembers they committed to that.

Cheers!

 

[jdwii]

FX 8350 is doing horrible in far cry 4 competing with an I3 and Amd is performing better on a Nvidia game?
http://www.techspot.com/review/917-far-cry-4-benchmarks/page6.html

What I find most interesting is in page 5: Both the i7-4770k and FX-8350 gain performance almost linearly with clockspeed. This indicates, even OCd to 4.5GHz, the game is still being limited to some extent by the CPU. The Intel chips are clustered together less because of a GPU bottleneck, and more because all their chips offer comparable performance.

What's also interesting is the rate of increase. The i7 is getting about 3 FPS per 500MHz. The FX is getting about 2 FPS per 500MHz. That's the IPC difference between the two showing up again, and accounts for all of Intels performance lead.

In the end, it looks like two strong cores simply outperform 8 weak ones. I'd be interested to see latency graphs, as the i3 is likely a lot less stable performance wise, but it's quite clear where Cry Engine 3's performance characteristics lie at this point.

It's bizarre - from what I've read, in most multi-threaded benchmarks the a8-7600 and i3-4330 (I think I got the number right) share around the same compute capability (usually within 10%), so would it be reasonable to infer that an FX-8350 being beaten by it probably means the game doesn't scale well off multiple cores? But then again in the consoles aren't they programming for 8 (light) cores?

I'm sure there's an ubisoft-blaming answer that satisfies the demands of reason somewhere here...

Games on consoles can use 6 cores. There's no evidence that FC4 uses all of them on consoles. Actually threads structure are usually the same on consoles and PC, that's why most games from past-gens use scale to only 3-4 cores (with drivers and system overhead).

Look at the Intel results too. There's 5 fps difference between i3 3220 and 3930k, yet 13fps difference between 2.5ghz 4770k and 4.5ghz 4770k. 3770k and 4690x is 2fps, within margin of error and 4690x has 100 extra mhz.

I would have liked to see Pentium G results just to see core scaling on a dual core without HT, and the unlocked one would be nice to see frequency scaling.

This looks like the only CPU benchmark so far that I can find. Reading reviews it looks like Ubisoft didn't do much to favor AMD hardware at all.

I'd assume scaling is similar to Crysis 3, which had two heavy threads and about 12 smaller rendering threads account for most of the work, I'd need to see someone look at a GPUView analysis or something though, since I have no plans to personally buy the game. It is quite possible though Ubi just did a really shoddy coding job, which given AC:U, may be the simpler, but correct, explanation.

To me, these results simply are the result of the Intel/AMD IPC difference.

I tested 4 using 8 cores vs 4 cores down below 8 without a doubt felt faster although was still skippy on my setup when driving.
4 cores
http://s7.postimg.org/5awex57q3/Far_cry_4_using_4_cores.png
8 cores
http://s29.postimg.org/wi0uwp6mf/Far_cry_4_8_Cores.png

If i turn off AA my GPU usage stays around 80% usage while my fps stays low even running V-sync off.

 

[Reepca]

https://www.youtube.com/watch?v=I6zX2IqBI7A&list=PL7DEF8194A6268E0E
("Preview AMD's next gen APU (Carrizo) with John Byrne")

Sometimes I wish non-informational marketing wouldn't use titles that make me want to watch it, hoping for any sort of meaningful information. I mean, I guess I learned... it will... be on time... and... power efficient... and... yeah that's all.

Entertainment value is definitely there, though, just because of the accent.

 

[con635]

Carrizo?
http://browser.primatelabs.com/geekbench3/compare/1279533?baseline=1279533

 

[szatkus]

Carrizo?
http://browser.primatelabs.com/geekbench3/compare/1279533?baseline=1279533

"Processor Codename Richland"

Seems legit.

 

[Fearnot]

Carrizo?
http://browser.primatelabs.com/geekbench3/compare/1279533?baseline=1279533

http://browser.primatelabs.com/geekbench3/619181 So it would be tying and beating Kaveri in almost all benchmarks while being clocked almost a whole ghz slower? Something seems off.

 

[anxiousinfusion]

Especially that 8350rocks said Carrizo is mobile only.

We will see Carrizo beyond mobile. In fact, AMD has already announced officially that Carrizo will be used in some "All-in-One systems".

"The 2015 desktop roadmap remains unpublished so far. AMD’s perception of a mobile-focused strategy would tend to suggest that the mobile comes first, with desktop following behind, although at this point it is unclear. A number of AMD’s marketing materials with this launch gave examples of the use of Carrizo and HSA for the prosumer, indicating that a desktop version should be announced in due course." - http://www.anandtech.com/show/8742/amd-announces-carrizo-and-carrizol-next-gen-apus-for-h1-2015

"Some rumors do point to the desktop variant being called Carrizo-D but AMD didn’t even mention about the desktop platform in the video but we already have some details as to what’s it all about." -http://wccftech.com/amd-officially-shows-carrizo-apu-sample-chip-carrizol-carrizo-apus-planned-1h-2015/

Also the fact that John Byrne discusses Carrizo as being a top-to-bottom product.

 

[gunrave]

Due to the newest AMD pressconference, leave me to think that, "Did AMD scare to give all they have in the competitive price just to discard the competition with Intel?"
I think AMD have all the tech we discussed about all entire thread but refuse to bring it on the consumer.
Even if they have outperform Intel in some segmen, they make the price higher which make them look expensive(eg. 7850K vs i3-4xxx).

Why AMD why..
CMIIW off course

 

[de5_Roy]

AMD Announces Major Technology Partnerships at Future of Compute Event
http://www.techpowerup.com/207362/amd-announces-major-technology-partnerships-at-future-of-compute-event.html

Numascale, Supermicro, and AMD Announce the World’s Largest Shared Memory System to Date
http://www.numaq.com/numascale-supermicro-amd-announce-worlds-largest-shared-memory-system-date-2/

semiaccurate's mantle tests:

part1
http://semiaccurate.com/2014/11/17/mantle-showdown-part-1/
part2
http://semiaccurate.com/2014/11/18/mantle-showdown-amds-a10-7850k/
part3
http://semiaccurate.com/2014/11/19/mantle-showdown-amds-r9-270x/
part4
http://semiaccurate.com/2014/11/20/mantle-showdown-amds-athlon-5350/


@anxiousinfusion: fix the anandtech link in the hyperlink.

@gunrave: amd eventually reduces cpu/apu prices which is why kaveri apus are as cheap as, if not cheaper and better than haswell/refresh core i3 cpus.

 

[esrever]

semiaccurate's mantle tests:

part1
http://semiaccurate.com/2014/11/17/mantle-showdown-part-1/
part2
http://semiaccurate.com/2014/11/18/mantle-showdown-amds-a10-7850k/
part3
http://semiaccurate.com/2014/11/19/mantle-showdown-amds-r9-270x/
part4
http://semiaccurate.com/2014/11/20/mantle-showdown-amds-athlon-5350/

I think it would have been interesting if they tuned the resolution and settings to playable levels for each card and game and then seen the difference. Although the kabini might not have been able to play any of the games even at minimal resolutions. The gains in Civilizations is a lot more than other sites maybe they tested differently.

 

[gunrave]

@gunrave: amd eventually reduces cpu/apu prices which is why kaveri apus are as cheap as, if not cheaper and better than haswell/refresh core i3 cpus.

Thx for reply,
Yeah the 7850k is definitely Haswell refresh core i3 killer, but in the beginning the price are really not make sense, unless AMD know the their high end kaveri are outperform Intel counterpart and adjust the price to add the margin.

The last price-cut is, i think, the most attracting for kaveri to sale.

When the price get in to my country, i will buy it..haha

 

[con635]

Carrizo?
http://browser.primatelabs.com/geekbench3/compare/1279533?baseline=1279533

"Processor Codename Richland"

Seems legit.

Compare scores to other cpus though, as fearnot shows it nearly beats out kaveri at 1 ghz more, in fact compare with i5 2400s 2.5ghz and its close, real close.