The Kaveri Pre-Launch Thread (A10-7800 and A10-6800k @3,5 Ghz)

[Fjodor2001]

They don't mean anything. I don't mean to be rude, but have you been living under a rock?

Then why are they still mentioned, and taken as a sign of Intel's process tech advantage by some?

Also, they must have used some math and methodology when arriving at the nm numbers, right? Or are you saying they just pick any random number...

But no, GloFo's 28nm process is not denser than Intel's 22nm process. They have minimum SRAM cell sizes of 0.120 µm2 and 0.092 μm2 respectively.

But from what we've seen GloFo's 28 nm process has higher transistor density than Intel's 22 nm process.

 

[Fjodor2001]

I think AMD is kinda out of the game when it comes to mobile processors if you discount mobile gaming laptops, or anything performance oriented for graphics. Haswell's TDP ranges anywhere from 11.5w to 57w across the entire mobile range from the 2961Y Celeron notebook processor to the 4930MX beast with the HD4600 graphics. In a time where 9 hour+ battery lifes are a deciding factor in netbook or small form factor laptop purchases, AMD just cant compete still. The deciding factor behind my purchase of my dell inspiron 11 was exclusively the fact that it used the haswell 15w 2955U, which made the battery life absolutely amazing for a 3 cell lithium ion.

Like i said though, alot of this is just speculation until we get some solid reviews from good sites about what its capable of. I for one am very excited to see what the results are and my hope is that kaveri puts AMD back within some range of competition with intel again. We are once again in a situation where Intel having zero competition has allowed them to price their parts exceptionally high. I have never in my life paid more then 250 dollars for a processor for any of my PC's including my gaming PC's.

Check out the AMD Kaveri battery life here:

 

[Homeles]

Today 250$ gives you about the fastest desktop CPU money can buy for consumer usage. 10 years ago when X2 ruled, 250$ wouldnt even buy you the cheapest X2. Looking back further, 250$ would give you even less. And thats even without accounting for inflation.

Yeah, he's a bit out of his mind. An AMD K6 233MHz launched at $469.

Then why are they still mentioned, and taken as a sign of Intel's process tech advantage by some?

Compare transistor performance between the two. Enough said.

Also, they must have used some math and methodology when arriving at the nm numbers, right? Or are you saying they just pick any random number...

No, they pretty much do pick a random number. It's based on ITRS's roadmap, and has nothing to do with feature size at this point.

But from what we've seen GloFo's 28 nm process has higher transistor density than Intel's 22 nm process.

No, from what we've seen, one specific chip on 28nm is denser than one specific chip on Intel's 22nm.

Here, allow me to put an end to your silliness:

 

[ShintaiDK]

The fastest "mainstream" processor right now is the 4770k i7 which is 339.99 on newegg while the i7-3960X Extreme Edition is $1,069.99. Not much has changed about intels lineup on pricing in 10 years. There were always high end mainstreams priced between 300-500 dollar's and "extreme edition" processors priced between 650-1000 dollars.

http://techreport.com/review/8616/amd-athlon-64-x2-3800-processor

Is it cheaper now or back then?

And then remember the CPI as well. You have to add something like 25% to those old prices to get todays cost.

In todays prices, and moving back to 1998. For 250$ you might just be able to buy a Celeron 300A. The cheapest CPU from Intel.

 

[Spawne32]

Check out the AMD Kaveri battery life here:

too many variables there with that graphic to make an assumption about how it compares to haswell one way or the other yet

Yeah, he's a bit out of his mind. An AMD K6 233MHz launched at $469.

A bit out of my mind? I said the past 10 years, IE 2004-2014. Not 1997. Intel's pricing segments have been around the same price ranges for mid range, mainstream, and high end for the past 10 years. With only times when AMD was very competitive were the prices driven down further.

Heres an article from 2006 quoting intels price schema vs AMD which was dominate up until the launch of C2D.

Moreover, Intel's Core 2 Duo prices are, for now, undercutting their competition. Intel's 2.67-GHz E6700 chip will be priced at $530, while the 2.6-GHz AMD Athlon 64 X2 5000+ is currently priced at $696, a 31 percent premium. Intel's Core 2 Extreme chip sits at the high end; the 2.93-GHz X6800 is priced at $999.

Intel's 2.4-GHz E6600 will ship for about $316, while the 2.13-GHz E6400 and 1.86-GHz E6300 will be priced at $224 and $183, respectively. Meanwhile, AMD's 2.4-GHz Athlon 64 X2 4800+ is priced at $645, while the 2.2-GHz 4200+ and 2.0-GHz 3800+ are priced at $365 and $303, respectively. AMD does not offer a sub-2.0-GHz Athlon 64 X2 processor. Both the E6600 and E6700 contain a total of four megabytes of level-2 cache, while the E6400 and E6300 contain just two.

No one argues the fact that over time gradually the prices do come down as technology allows it, having a computer, let alone building your own in the 90s was damn near not even heard of. My parents spent 2000 dollars on my first 155mhz PC back in 1995 and it was refurbished to save money.

 

[Spawne32]

http://techreport.com/review/8616/amd-athlon-64-x2-3800-processor

Is it cheaper now or back then?

And then remember the CPI as well. You have to add something like 25% to those old prices to get todays cost.

In todays prices, and moving back to 1998. For 250$ you might just be able to buy a Celeron 300A. The cheapest CPU from Intel.

Let's see, pentium extreme edition still priced at $999, mainstream desktop processor priced between $316 and $530. Mid range priced around $200 dollars. Yeh seems to be about the same as it is now.

 

[inf64]

Guys please stay on topic of Kaveri/Richland comparison and leave the intel/amd pricing from 10 years ago for some other topic.

 

[ShintaiDK]

Let's see, pentium extreme edition still priced at $999, mainstream desktop processor priced between $316 and $530. Mid range priced around $200 dollars. Yeh seems to be about the same as it is now.

But does the extreme cost 1250-1300$ today? No it cost 1059$.
Does the mainstream/performance cost between 380 and 650$ today? No, they cost between 187 and 350$.

In short, CPUs have never been cheaper.

 

[Spawne32]

But does the extreme cost 1250-1300$ today? No it cost 1059$.
Does the mainstream/performance cost between 380 and 650$ today? No, they cost between 187 and 350$.

In short, CPUs have never been cheaper.

I can see that im clearly arguing with a child who wasnt around to buy processors in the past decade. Ill concede the argument to you for the sake of the thread, since now you just want to make up numbers to prove your point after posting evidence to the contrary.

Insulting other members is not allowed here.
Markfw900
Anandtech Moderator

 

[ShintaiDK]

Kaveri will help a lot on the mobile side wich is currently
not that bad at all...

http://www.pcper.com/reviews/Graphi...rinity-and-Richland-Integrated-Graphics/Perfo

It shows its very bad. Intel leads with over 2x on the performance/watt arcording to your article. But we need to see what clocks and at what TDPs Kaveri can substain in the mobile front. The GPU part was bad for AMD there previously, since it had to be reduced alot compared to desktop variants.

 

[inf64]

45W Kaveri performs better than 100W Richland when iGPU is in question(in 3dmark fire strike):

1125pts for 6800K Richland vs 1287pts for 45W kaveri- Kaveri has 15% higher score while having more than 2x less TDP rating. That is 2.55x more perf./watt in 3dmark fire strike! Now the 35W will probably perform a little bit lower in GPU benchmarks than 45W but this time around 35W mobile parts will perform well Vs previous gen 100W desktop counterparts.

 

[9enesis]

45W Kaveri performs better than 100W Richland when iGPU is in question(in 3dmark fire strike):.

(facepalm) its just one picked benchmark....

Now the 35W will probably perform a little bit lower in GPU benchmarks than 45W but this time around 35W mobile parts will perform well Vs previous gen 100W desktop counterparts.

thats just a guess .... at the moment it is not possible to say what the TDP/performance scalling will be

 

[inf64]

True and I stated that in the brackets . I do think that 45W Kaveri will perform close to 100W Richland since it will have same number of SP (384) which will be way more efficient (GCN >> VLIW4) and which will run at similar clock speed as desktop Kaveri variant(~720Mhz as per AMD). Basically AMD is expecting that iGPU in 45W part will have ~1.2x or around 20% lower iGPU performance than the 95W Kaveri part which is in line with what we know about Kaveri. 512SP part won't perform at its peak anyway due to memory BW bottleneck so I wouldn't be surprised to see 6CU version throw in some very good numbers in comparison with 95W (8CU) part.

 

[9enesis]

@inf64

i think you forget one thing - kaveri is not competing against richland - and the reason why i bring this up is the mainstream memory being used in the mobile sector , that would be: DDR3-1600 , which is fine by intel but not good for amd

ETA: i think that both mobile APU GPUs be it richland(384) or kaveri(384-512) are fast enough to saturate DDR3-1600 BW limit , therefore, there will be almost no performance gain from kaveri whatsoever,though not sure what MANTLE will add....just my wild guess

 

[Fjodor2001]

No, from what we've seen, one specific chip on 28nm is denser than one specific chip on Intel's 22nm.

Yes, but since the A10-7850K and 4770K are quite similar chips (about the same percentage of die area allocated to CPU cores vs iGPU), we can also assume the GloFo 28 nm process has higher transistor density than Intel's 22 nm in general for that type of chips (APUs).

Here, allow me to put an end to your silliness:

Isn't that because the Dothan CPU has a much larger L2 cache? It seems to occupy more than 50% of the die area when looking at the die image. I think it's 2 MB (Dothan) vs 512kB (Prescott) if I recall correctly. If so, they are not really comparable, since cache memory has a much higher transistor density than other parts of the chip.

 

[mrmt]

Yes, but since the A10-7850K and 4770K are quite similar chips

Is it? Intel devotes 33% of the 4770k area do the iGPU, while AMD devotes 47%.

 

[Fjodor2001]

Is it? Intel devotes 33% of the 4770k area do the iGPU, while AMD devotes 47%.

Yes, only 47-33=14% difference. I.e. that alone does not explain the difference in transistor density between GloFo 28 nm vs Intel 22 nm. As already mentioned here before.

Also, A10-7850K has only 4 MB cache, compared to 8 MB for 4770K. That should also help raise the tranistor count in Intel's favor in a similar way, since cache memory has higher transistor density than the rest of the chip. But apparently it is still not enough.

 

[Abwx]

Yes, but since the A10-7850K and 4770K are quite similar chips (about the same percentage of die area allocated to CPU cores vs iGPU), we can also assume the GloFo 28 nm process has higher transistor density than Intel's 22 nm in general for that type of chips (APUs).

On Haswell if you count the GPU and L3 cache you ll
end with more than 50% die area being devoted to high
density circuitry, a number that is about the same as
Kaveri s high density ratio.

 

[ShintaiDK]

Since we continue along the transistor/density talk. I would like to know why Kaveri is 2.41B transistors and Richland 1.3B. Since the cores is not radically changed, not more of them either. Same cache too.

So what are those transistors used on? Assuming its not a wrong number. It would mean something like ~1.25B transistors are not used for the GPU if we look on Hawaii. Please note, a 4M/8T PD with 4x2MB L2 and 8MB L3 is 1.2B for the entire chip.

 

[inf64]

I can't say much but let me just state that when it comes to iGPU, stock 7850K will do good Vs solidly OCed 6800K .

 

[Homeles]

Isn't that because the Dothan CPU has a much larger L2 cache? It seems to occupy more than 50% of the die area when looking at the die image. I think it's 2 MB (Dothan) vs 512kB (Prescott) if I recall correctly. If so, they are not really comparable, since cache memory has a much higher transistor density than other parts of the chip.

No, that's exactly why they're a perfect example. Even two chips on the same process can have wildly different densities.

You seem to be confusing the idea of a process being denser with the idea of a design being denser. There are no ifs, ands or buts -- Intel's 22nm process is denser. Period.

The design decisions made by AMD led to them having a denser chip. Put the same design on Intel's 22nm, and it would be significantly smaller.

Yes, only 47-33=14% difference. I.e. that alone does not explain the difference in transistor density between GloFo 28 nm vs Intel 22 nm. As already mentioned here before.

That is not how math works. 47% is a 42% larger number than 33%.

 

[ShintaiDK]

Another example for densities is Hawaii vs Tahiti vs Pitcairn. 6.2B/438mm2 vs 4.2B/352mm2 vs 2.8B/212mm2. Tahiti in this case was designed for better properties with its larger area. Both Hawaii and Pitcairn are smaller designs, but with worse properties. And for Hawaii it was a bad decision.

 

[NostaSeronx]

~600 Million CPU (~200 M for Logic and ~100 M for SRAM)
~1.4 Billion GPU
~400 Million NB/PCIe/DDR (Mostly Logic and Tx/Rx)

Rough Estimates.

 

[ShintaiDK]

~600 Million CPU (~200 M for Logic and ~100 M for SRAM)
~1.4 Billion GPU
~400 Million NB/PCIe/DDR (Mostly Logic and Tx/Rx)

Rough Estimates.

1.4B for the GPU doesnt add up with the dGPUs.

Even if we include memory controller and PCIe. Hawaii is 1.1B for 512SP, Tahiti is 1.05B and Pitcairn is 1.12B.

600M for the CPU would also make it half the size of an entire 4M/8T chip. With L3 cache, NB, memory controller and hypertransport. Or making the 2M/4T Kaveri CPU part almost as big as the entire 4M/8T chip.

 

[Homeles]

1.4B for the GPU doesnt add up with the dGPUs.

Even if we include memory controller and PCIe. Hawaii is 1.1B for 512SP, Tahiti is 1.05B and Pitcairn is 1.12B.

600M for the CPU would also make it half the size of an entire 4M/8T chip. With L3 cache, memory controller and hypertransport.

Nevermind. I see what you're doing.

Cape Verde is a 1.5 billion transistor part, and that's with extra memory controller logic, PCI-E links, shaders, etc.. There's no way Kaveri's GPU all by its lonesome is even close to 1.5 billion.