AMD CPU speculation... and expert conjecture

[mayankleoboy1]

Found this interesting:
I also asked Huang how investors should think about margins longer term, assuming a snap-back in Tegra revenue, given that Tegra brings lower gross margin than the company’s GPU parts such as “Kepler.”

“It will have an effect, but Tegra is not that much lower, it’s close to 50%” gross margin, Huang pointed out to me. “But as that takes off, we’d be very happy to have the volume” in Tegra sales, he added.
http://blogs.barrons.com/techtraderdaily/2013/05/10/nvda-bulls-raise-targets-on-second-half-product-cycle/?mod=BOLBlog

If Nvidia cant accelerate Tegra lineup, it is doomed. Already the T4 is late, and no OEM worth his name is interested in using it. We only have HP and ZTE with a possible Tablet. And there are no Smartphone wins at all.
Despite the hype, Tegra is slowly failing, mostly due to delivering lesser than promised.

 

[griptwister]

I prefer "hasbeen"

I like "hasbeen". I still prefer "failwell" because resembles the typical "faildozer" that you could read in so many forums when AMD released Bulldozer.

Haswell isn't anything remotely close to the Bulldozer fail.

Minor IPC gain, large graphics gain, and better power savings is what they were after. It has little benefit for desktop enthusiasts but the laptops will love it.

No... Haswell is pretty bad bro. I have an i7 laptop and I'm pretty sure that i7 with HD 4000 Graphics will not suffice for most FPS. IMO, the APU is where people should be going for a budget gaming laptop. Anything Marked Intel on a lap top with out a GPU, I will be sure to stay away from. Thank God my laptop was a gift. Cause honestly, I'd return it.

Someone used this math with the APU cores, so now I'm going to use it with the Intel HD Graphics: 2 x Crap = 2ce the Crap!

Note: Websites and sales men will sell you the name, not the performance.

Also, I guess AMD made their own XMP (Or is it AMP) profiles for their Radeon branded RAM? They've also claimed up to a 22% more FPS with gamer series RAM.

http://www.radeonmemory.com/

http://www.newegg.com/Product/Product.aspx?Item=N82E16820403002

 

[juanrga]

Also, I guess AMD made their own XMP (Or is it AMP) profiles for their Radeon branded RAM? They've also claimed up to a 22% more FPS with gamer series RAM.

http://www.radeonmemory.com/

http://www.newegg.com/Product/Product.aspx?Item=N82E16820403002

XMP is Intel profiling. AMP is AMD profiling. The AMD radeon gamer memory supports both XMP and AMP for compatibility with a larger number of mobos.

 

[mayankleoboy1]

Also, I guess AMD made their own XMP (Or is it AMP) profiles for their Radeon branded RAM? They've also claimed up to a 22% more FPS with gamer series RAM.

22% better over an APU coupled with a 1333 MHZ ram. Though IIRC, the best price/perf for RAM speeds was 1600 MHZ. Below it, you lose too much perf on the table. Above it, the gains are less than the large amount of money you have to spend on a 2133 RAM.

 

[noob2222]

Also, I guess AMD made their own XMP (Or is it AMP) profiles for their Radeon branded RAM? They've also claimed up to a 22% more FPS with gamer series RAM.

22% better over an APU coupled with a 1333 MHZ ram. Though IIRC, the best price/perf for RAM speeds was 1600 MHZ. Below it, you lose too much perf on the table. Above it, the gains are less than the large amount of money you have to spend on a 2133 RAM.

im guessing you haven't checked prices lately.
www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=100007611%2050008476%20600006069&IsNodeId=1&bop=And&Pagesize=100
1600 easily had for 64.99, 2133 is a large ammount of 69.99, $5 isn't that much especially if you consider the entire system cost. as a $500 build, that's a measly 1% cost for 22% faster graphics.

 

[The Q6660 Inside]

Also, I guess AMD made their own XMP (Or is it AMP) profiles for their Radeon branded RAM? They've also claimed up to a 22% more FPS with gamer series RAM.

22% better over an APU coupled with a 1333 MHZ ram. Though IIRC, the best price/perf for RAM speeds was 1600 MHZ. Below it, you lose too much perf on the table. Above it, the gains are less than the large amount of money you have to spend on a 2133 RAM.

im guessing you haven't checked prices lately.
www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=100007611%2050008476%20600006069&IsNodeId=1&bop=And&Pagesize=100
1600 easily had for 64.99, 2133 is a large ammount of 69.99, $5 isn't that much especially if you consider the entire system cost. as a $500 build, that's a measly 1% cost for 22% faster graphics.

On an Intel system, anything above 1600 is a waste, on an AMD FX system, 1866+ is the way2go.

 

[noob2222]

^ and for that reason, run all benchmarks with 1333. cant let AMD runwith their top specs, might give them an edge.

kinda irritating that its not acceptable to run faster memory on a review until intel "officially" supports it well after AMD did, even though supporting it is only a technicality, all you have to do is tell the bios how fast to run the memory.

 

[The Q6660 Inside]

^ and for that reason, run all benchmarks with 1333. cant let AMD runwith their top specs, might give them an edge.

kinda irritating that its not acceptable to run faster memory on a review until intel "officially" supports it well after AMD did, even though supporting it is only a technicality, all you have to do is tell the bios how fast to run the memory.

By the time Intel has "official" support for 1866, AMD will be crusing with "official" support for DDR3-3200 I remember the days of the good ol FSB and Memory controller on the chipset (ala EVGA 790i DDR3-2000).

 

[juanrga]

On an Intel system, anything above 1600 is a waste, on an AMD FX system, 1866+ is the way2go.

Agree! I have seen lots of 'fair' reviews of the AMD FX 8350 where the chip is forced to run underclocked memory @ 1600 instead of @ stock speed. I have seen one review where this happens whereas some Intel chips such as i7-3930K and i7-3960x Extreme were running memory near 2000 MHz (and with less latency)

http://www.bit-tech.net/hardware/2012/11/06/amd-fx-8350-review/2

I have also seen a first review of Haswell i7-4770K using 2600 MHz ram!!!

 

[The Q6660 Inside]

On an Intel system, anything above 1600 is a waste, on an AMD FX system, 1866+ is the way2go.

Agree! I have seen lots of 'fair' reviews of the AMD FX 8350 where the chip is forced to run underclocked memory @ 1600 instead of @ stock speed. I have seen one review where this happens whereas some Intel chips such as i7-3930K and i7-3960x Extreme were running memory near 2000 MHz (and with less latency)

http://www.bit-tech.net/hardware/2012/11/06/amd-fx-8350-review/2

I have also seen a first review of Haswell i7-4770K using 2600 MHz ram!!!

+1, I am guessing the Steamroller 8550 (Name implication) will have a DDR3-2400 "official" memory controller, and this quoting system is awful, my post was stolen .

 

[jdwii]

On an Intel system, anything above 1600 is a waste, on an AMD FX system, 1866+ is the way2go.

Agree! I have seen lots of 'fair' reviews of the AMD FX 8350 where the chip is forced to run underclocked memory @ 1600 instead of @ stock speed. I have seen one review where this happens whereas some Intel chips such as i7-3930K and i7-3960x Extreme were running memory near 2000 MHz (and with less latency)

http://www.bit-tech.net/hardware/2012/11/06/amd-fx-8350-review/2

I have also seen a first review of Haswell i7-4770K using 2600 MHz ram!!!

+1, I am guessing the Steamroller 8550 (Name implication) will have a DDR3-2400 "official" memory controller, and this quoting system is awful, my post was stolen .

Lets not forget about the double posts either.

 

[cowboy44mag]

With Haswell releasing soon I wonder how long AMD will wait until they release Steamroller. I'm assuming that part of the silence from AMD regarding Steamroller was so they could see what Haswell had to offer before they had to show their "hand". Really if AMD was able to beat Ivy Bridge with Steamroller then they have probably surpassed Haswell as well. The reviews for Haswell are downright bleak. I'm just thinking that if Steamroller doesn't need many "tweaks" to best Haswell than we may see an earlier than expected release.

Also does anyone know how they will be releasing the new line? For example will it be APUs to get Steamroller first, FX series, or will they release about the same time?

 

[mayankleoboy1]

I dont expect SR to beat Failwell in single threaded, stock per stock basis. That is work for Excavator.

 

[noob2222]

kaveri will have the first SR cores.

as for benchmarks, depends. cinebench is impossible for amd to catch intel (sse3 vs avx coding). games will still slightly edge towards intel (AMD needs the on chip pcie contoller wont see till excavator).

I thing the main delay for sr is the 28nm fabs, if thats the correct node.

 

[palladin9479]

kaveri will have the first SR cores.

as for benchmarks, depends. cinebench is impossible for amd to catch intel (sse3 vs avx coding). games will still slightly edge towards intel (AMD needs the on chip pcie contoller wont see till excavator).

I thing the main delay for sr is the 28nm fabs, if thats the correct node.

On-die PCIe controller doesn't do much if anything for system performance. HT is more then fast enough and we're not even using it's full specified bandwidth. If anything AMD's IMC needs work though it's still not an issue when talking to PCIe devices, notably GPUs. This may change in the next three years, but there is no realistic improvement needed in the next two years.

 

[juanrga]

+1, I am guessing the Steamroller 8550 (Name implication) will have a DDR3-2400 "official" memory controller, and this quoting system is awful, my post was stolen .

My bet for stock speed would be 2133 MHz, now that AMD has just released AMP Radeon memory with that speed, but I wait mobos to support 2400 and above via OC.

kaveri will have the first SR cores.

as for benchmarks, depends. cinebench is impossible for amd to catch intel (sse3 vs avx coding). games will still slightly edge towards intel (AMD needs the on chip pcie contoller wont see till excavator).

I thing the main delay for sr is the 28nm fabs, if thats the correct node.

Cinebench is one of those unfair benchmark using the Intel compiler, which forces AMD chips to run the slowest possible code, even if they can run more faster one.

 

[mayankleoboy1]

Cinebench is one of those unfair benchmark using the Intel compiler, which forces AMD chips to run the slowest possible code, even if they can run more faster one.

Cinebench is based on a software (Maxon) that professionals use for their work. So if Cinebench favours Intel, then the professional will buy an Intel processor. So for that software, intel is the better CPU. If you want to use Cinebench you buy intel, or you go home.

If you want completely fair benchmarks, then say goodbye to Windows. Take up a linux distro, and build all the opensource softwares with -O3 -march=native flags. This way you can be assured that no hanky-panky is being done by Intel.

Because no closed source company is going to reveal how they build and code their program.

Edit :

which forces AMD chips to run the slowest possible code, even if they can run more faster one.

Its not that simply specifying -/AVX to the compiler will automagically make the program faster. It is extremely seldom that happens. 85% of the time, there is no effect. 12% of the time, there is perf regression. 3% of the time, there is a gain, in the 4-5% range.
Also, unless you have Cinebench compiled by the most AMD friendly compiler ever, you cant say Cinebench is unfair, simply because you dont know what is the perf of Cinebebch on pro-AMD compiler.

Edit : This is a good illustration : http://www.h-online.com/newsticker/news/item/Processor-Whispers-About-latencies-and-compilers-1232290.html?view=zoom;zoom=4

Edit2 : Also, compiler optimisations are for wimps and sissy's. Real coders use assembly code or compiler intrinsic's for CPU intensive code.

 

[truegenius]

-_- and we have discused that intel compiler produce fast code than amd's own compiler !!!

if amd hardware is capable but they are not doing anything to force softwares to use its capabilities then its capability is of no use

its like saying that apple iphone 5 is near htc sensation because it uses only 1.2ghz cortex a15 style dual cores and 1GB ram only
but iphone 5 belongs to s4 teritory

 

[cowboy44mag]

I dont expect SR to beat Failwell in single threaded, stock per stock basis. That is work for Excavator.

I agree that Haswell will more than likely still beat Steamroller in single thread, stock per stock benchmark, although Steamroller IMO will lessen the gap a lot. There is more to computing than just single core performance though, and that is where I think Steamroller will surpass Haswell by a wide enough margin that even Intel will have to pay attention.

 

[sarinaide]

Again we are at that age where software writes are trying to use extentions and code to scale hardware and that never works, my best example is if you try emulate a PS3 no tweaks nor the fastest Intel Xeon 20 Thread CPU and two of them can muster up playable framerates or experiences, simply because software trying to scale is inneficient. It is time for a change, AMD provides that chance for hardware to flourish and scale without the need of API's and the works.

Right now nothing can manipulate AMD's core count or HSA architecture, but prominant game developers are saying that will change headlong in 2014 with more codings for this, some have gone on to say this has made developing simpler and scale better. In pure x86 Intel can pump 7BN dollars into strain testing silicon and have dense nodes, for intel to achieve higher performance is not really that outstanding of an achievement to think that they still struggle at times to definitively beat AMD despite as much as 10x the resources in R&D alone.

This is like Formula One racing Redbull have a near unlimited budget due to the label it represents Redbull is probably in the top 10 brands currently and turning over, this is like Intel in the semi conductor world. AMD is more like Marussia Racing whom are basically the back markers that finish +1 or 2 laps down the road. Pure and simple money = dedication to resources and development which ultimately produces results, RBR crush Marussia definitively without even an arguement being made, Intel still doesn't do that despite having all the leverage in the world. So before we go on bitching about AMD not delivering Intel beating performance just think about how things have changed, how even when AMD had a good arch they were phased out the market by Intel's lust for control. All things said if they continue to make the strides over previous generations then its good. Excavator will be a new architecture and one where the likes of Keller can influence, but Keller is not the only top industry engineer and designer AMD reeled in under the radar.

 

[Cazalan]

Low end Richlands now in stock.

http://www.cpu-world.com/news_2013/2013051301_AMD_A4-4000_is_in_stores_now.html

Says 70 in stock. A4-4000 (Dual core 3.2Ghz)

 

[8350rocks]

True...the talent they've amassed in the last 1.5 years is mind bending in it's own right:

John Gustafson
Jim Keller
Raja Kudori

Those names are household names among technophiles...people who haven't just contributed...but made industry wide impacts. I have no doubts that the rewards for those hirings are still a few years out, and AMD is competitive now! In the next few years...it will be a really interesting series of events unfolding. I can't fathom what leaps and bounds forward will come from minds like those under one roof with CPU and GPU technology at their fingertips.

 

[kettu]

Looks like it's actually a 3.0GHz part with turbo upto 3.2GHz. Unless CPU-world got it wrong. Shops seem to be selling it as a 3.2GHz part though. In my opinion that's false advertising. Perhaps an honest mistake that hopefully gets corrected soon.

 

[Cazalan]

I think Intel started that trend when their stock frequencies were starting so much lower than AMD FX.

Like most marketing it's whatever sounds better. That new 36mpg car won't get 36mpg unless you drive it just right.

 

[kettu]

As far as I can remember Intel never implied that turbo frequency was the operating frequency of their chips. Besides, it's not AMD doing this but the shops (unless AMD is telling them to do that, but there is no evidence of that). Either they don't understand the specsheets or they are misrepresenting them on purpose. Or perhaps CPU-world is wrong and the chip is actually a 3.2GHz. In any case some other industry doing shady marketing doesn't make it right. It just makes the other guys wrong. Supposedly EU atleast is investigating the automobile industry and their fuel consumption figures.