Vishera spotted running at 5GHz

[MLSCrow]

G'head and prosecute me for starting a new Vish thread, I don't care, but at least at this point we have something new and interesting to discuss.

Vishera spotted running at 5GHz watercooled with stock AMD watercooler. Looks like that's what I'm gonna get.

http://www.theinquirer.net/inquirer/news/2204765/amd-vishera-chip-running-at-5ghz-is-spotted

 

[Yuriman]

Doesn't really tell us a whole lot. You can hit 5ghz with most modern chips if you put enough volts through them.

However, it does look promising that (source) for the first time, a manufacturer will be shipping a 4ghz+ CPU from the factory, and the TDP isn't outrageous either.

EDIT: Looking at this chart, something seems fishy. AMD will be selling a 2 module chip at 4ghz/4.2 turbo with no iGPU with a 95w TDP, while the A10 is 3.4ghz/4ghz with an HD7660D and has a 65w TDP.

To get from 3.4ghz to 4ghz, even with deleting the iGPU, TDP goes up by almost 50%.

 

[MLSCrow]

What IS interesting about the way the article is worded is that the chips is an AMD demonstration unit. Not some OEM or third party vendor, but 5GHz capable strait from AMD's mouth, but perhaps there is something lost in translation. I'm hoping for the best though.

EDIT: Looking at this chart, something seems fishy. AMD will be selling a 2 module chip at 4ghz/4.2 turbo with no iGPU with a 95w TDP, while the A10 is 3.4ghz/4ghz with an HD7660D and has a 65w TDP.

To get from 3.4ghz to 4ghz, even with deleting the iGPU, TDP goes up by almost 50%.

Keep in mind that the desktop version will have 8MB of included cache, which could contribute to a large portion of that increased TDP.

 

[NTMBK]

EDIT: Looking at this chart, something seems fishy. AMD will be selling a 2 module chip at 4ghz/4.2 turbo with no iGPU with a 95w TDP, while the A10 is 3.4ghz/4ghz with an HD7660D and has a 65w TDP.

To get from 3.4ghz to 4ghz, even with deleting the iGPU, TDP goes up by almost 50%.

Don't forget, AMD has some pretty loose "bins" on their TDPs. Take a look at their current Llano desktop numbers: http://en.wikipedia.org/wiki/AMD_Fusion#Llano_.2832.C2.A0nm.29 Notice how there's a sudden jump between A8-3820 and A8-3850? Basically the *20 will have a real world TDP just under 65W, so it will get the lower TDP rating, whereas the *50 has a real world TDP just over 65W, which means it gets bumped up to the next rating. The whole range doesn't neatly happen to have precisely either 65W or 100W TDPs, these are just loose approximations I wouldn't read too much in to. Wait until we see actual test numbers.

 

[Ancalagon44]

Also remember Bulldozer clocked at 8+ Ghz, and look at a turd that turned out to be.

 

[SPBHM]

cool, but I wonder if it uses as much power as BD...

 

[ShintaiDK]

Also remember Bulldozer clocked at 8+ Ghz, and look at a turd that turned out to be.

You forgot "dancing in the aisles" :awe:

 

[Pilum]

However, it does look promising that (source) for the first time, a manufacturer will be shipping a 4ghz+ CPU from the factory,

Intel started shipping 4.4 GHz Xeons last spring. And of course IBM shipped 4+ GHz POWER6 five years ago. But hey, nice that AMD is finally catching up.

 

[guskline]

What is stable? Able to boot into Windows without crashing or able to run all 8 cores flat out on prime95 or pass IntelBurn Test with temps within acceptable ranges? I opt for the latter definition. That said, my 8150 can run at 4.5 Ghz while running prime 95 all day without problems and run IntelBurn test all day without problems. I can see a PileDriver with improved IPC and Resonant Mesh technology doing the same thing at nearly 5 Ghz if AMD made the improvements that have been claimed.

The question is how does a 5 Ghz Vishera stack up to Intel CPUs of similar price?

 

[guskline]

Also remember Bulldozer clocked at 8+ Ghz, and look at a turd that turned out to be.

Correction: 1 of the 4 modules (2 cores) was clocked at that speed. To achieve this "record" they had to disable 3 of the 4 modules, use UBER exclusive cooling and every tweak known to man. Frankly, a silly demonstration. I own a 8150 and have fun with it but what is the record for a STABLE OC'd 4 module 8 core 8150?

 

[inf64]

I think 8150 can run stable at 4.6-4.7Ghz on good air cooler. But it burns a lot of power at this clock under full load(good thing that power saving features work even when OC so it can clock/volt down).

5Ghz for Vishera is not bad but until we see retail samples running @ 5Ghz under same conditions,stable, in end user rigs ,we can dismiss this news(or take it with a grain of salt).

 

[Hubb1e]

Doesn't really tell us a whole lot. You can hit 5ghz with most modern chips if you put enough volts through them.

However, it does look promising that (source) for the first time, a manufacturer will be shipping a 4ghz+ CPU from the factory, and the TDP isn't outrageous either.

EDIT: Looking at this chart, something seems fishy. AMD will be selling a 2 module chip at 4ghz/4.2 turbo with no iGPU with a 95w TDP, while the A10 is 3.4ghz/4ghz with an HD7660D and has a 65w TDP.

To get from 3.4ghz to 4ghz, even with deleting the iGPU, TDP goes up by almost 50%.

AMD's chips have tools that allow them to estimate power usage and clock the chip to hit a particular TDP. It's similar to what they do with their graphics cards when running a "power virus" such as furmark. The 65W Trinity chips will likely not be able to sustain clocks at full load while the 95W chips will be able to sustain full load longer at a high frequency.

 

[Kaldorine]

I don't have an educated opinion on the actually cpu. But what is going on with the names here? Piledriver/Steam roller/Bull dozer? Really? when i think of a cpu i'm not thinking of heavy machinery.

 

[SPBHM]

I don't have an educated opinion on the actually cpu. But what is going on with the names here? Piledriver/Steam roller/Bull dozer? Really? when i think of a cpu i'm not thinking of heavy machinery.

it's funny...
but when you think about CPUs you are thinking about Conroe? :biggrin:

 

[thilanliyan]

I like the F1 track naming of the server products.

 

[Soulkeeper]

I like the F1 track naming of the server products.

AMD used to sponsor a Formula 1 car

"Rival semiconductor manufacturers Intel and AMD sponsor BMW Sauber and Ferrari respectively. "

I remember they even had a ferrari racing edition laptop at one point.

Not sure if they still have the money for this ...

 

[SPBHM]

AMD used to sponsor a Formula 1 car

"Rival semiconductor manufacturers Intel and AMD sponsor BMW Sauber and Ferrari respectively. "

I remember they even had a ferrari racing edition laptop at one point.

Not sure if they still have the money for this ...

they still have some sort of deal with Ferrari, there is a AMD logo on the rear wing (side),
I remember back when the Phenom was released in 2007 they even had a Phenom logo on the helmets..

Ferrari needs a lot of CPUs for their CFD and other simulations, so that's probably part of it...

I want one AMD Nordschleife :biggrin:

 

[nenforcer]

Probably still at 200+ Watts....but still nice it can reach such speeds.

 

[Ancalagon44]

I don't have an educated opinion on the actually cpu. But what is going on with the names here? Piledriver/Steam roller/Bull dozer? Really? when i think of a cpu i'm not thinking of heavy machinery.

Because they are a steaming pile of bull....

Without the driver, roller and dozer.

 

[NTMBK]

Because they are a steaming pile of bull....

Without the driver, roller and dozer.

 

[NostaSeronx]

I don't have an educated opinion on the actually cpu. But what is going on with the names here? Piledriver/Steamroller/Bulldozer? Really? when i think of a cpu i'm not thinking of heavy machinery

It is something from the design phase that didn't make it to the actual architecture.

Every version of Bulldozer, Piledriver, Steamroller, Excavator has 4 AGLUs that each compute 4 Macro-ops per cycle which are an arithmetic/logic ops + memory op. Do to the 1 cycle latency it would allow the retire unit to flip between the core and the FPU every other cycle. Leading to a total of 12(8 really but we all know on the whitepaper it will say 12) macro-ops being executed per module.

Reality:
Bulldozer -> 1 macro-ops per core/2 macro-ops per FPU(unless the code is only core specific)
Piledriver -> 1 macro-ops per core/2 macro-ops per FPU(unless the code is only core specific) <-- improved scheduling which leads to a 5 to 20% improvement.
Steamroller -> 2 macro-ops per core/4 macro-ops per FPU(unless the code is only core specific)
Excavator -> 2 macro-ops per core/4 macro-ops per FPU(unless the code is only core specific) <-- improved scheduling(perf)/increased density(cost).

Not until the fifth generation of Bulldozer fam will we see the original design goal to be implemented.

 

[guskline]

NostaSeronx, thank you for that explanation. Are you a CPU engineer? Your knowledge and ability to explain the design changes are impressive.

 

[NostaSeronx]

NostaSeronx, thank you for that explanation. Are you a CPU engineer? Your knowledge and ability to explain the design changes are impressive.

I am not an engineer of any sorts. I just have a lot of free time to look between 2006 and 2008.

The High IPC design that was meant to be became a High Frequency design because they couldn't implement all that power in 32-nm by 2009/2011. At the time I also believe they didn't want to make a Pentium 4 mistake or make a CPU to powerful to market.

High IPC and High Frequency designs appear then vanish over the eras of 1985 to 201x. Everything is median focused or it is deemed a failure. This generation of Bulldozer fam is near the HF median. While the Core fam is actually is near HF but not as near as Bulldozer.

The absolute reason why Bulldozer sucks is programs don't tend to upgrade code overnight. 2009 vs 2011-wise Bulldozer couldn't have come any sooner in a marketing standpoint. Also, another thing to point out for all the bragging about FPFMA/MUL/ADD it kinduh sucks compared to FPMMA/MAL.

You can shoot off 2 * Integer FMA/Mul + 2 * Integer Add each in 128-bit width(512-bit in aggregated FMA+ADD or MUL+ADD) in Bulldozer but you have issues trying to shoot out 2 * Floating Point FMA/MUL/ADD. Is Bulldozer meant for FP, nope.

What is most game physics/etc in oh that is right Floating Point it's no wonder why Bulldozer can get low frame rates. If the game runs on Integer logic/math you can see a measly FX-4100 destroying an i7-3960X(most games have really poor scaling). You can make any game run Integer logic/math if you pump up that resolution and defer that physics to the GPU.

OpenCL+CUDA+C++ AMP /+/ C/C++ = Bulldozer+GCN or Kepler > Core+GCN or Kepler. But, the change to the hetero ideology is taking its time.

--Back to Vishera--
OR-C0 = PH-E0 * 1.13

Vishera @ ~5 GHz should have the same power consumption of Thuban @ ~4.5 GHz.

OR-C0 is power consumption fixed
with logic that was disabled or wasn't functioning properly -> with logic that is enabled and is functioning properly.

OR-C0 = TN-A1 in compute performance. TN-A1 has "more" if you consider them more because it has a GPU attached to it's donkey butt. Trinity has to deal with memory bandwidth being swept away by the GPU. Orochi doesn't have to deal with that so it can get the data when it wants and how it wants.

So, you have a 1090T or 1100T at 4 GHz and want to upgrade to something that has similar power consumption but more performance well you can get a FX-8350 or FX-6300* @ ~4.5 GHz.

*It will have lower power consumption than the 1100T @ 4 GHz.

 

[AtenRa]

Nice CineBench score for the FX8350 (4.8GHz or 5GHz), faster than Core i7 2600K at 4.5GHz.

FX8350

FX8150@4.7GHz

Core i7 2600K @ 4.5GHz

FX8150 and Core i7 2600K source :
http://forums.anandtech.com/showthread.php?t=2241712&highlight=

 

[Makaveli]

G'head and prosecute me for starting a new Vish thread, I don't care, but at least at this point we have something new and interesting to discuss.

Vishera spotted running at 5GHz watercooled with stock AMD watercooler. Looks like that's what I'm gonna get.

http://www.theinquirer.net/inquirer/news/2204765/amd-vishera-chip-running-at-5ghz-is-spotted

There is no need to prosecute.

You posted a link from the Inquirer!

It would be like me posting a wiki link as fact!

Will wait until Anand benchs it and their are proper reviews before getting a hard on