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

[Enigmoid]

Not really unless you got some new info you can share. When an Iris Pro based Core i3 is exactly the same price as the Kaveri based CPUs then maybe.

Perhaps you want to show us the socketed Core i5 R series CPUs for desktop?? Or even the DIY soldiered motherboards which we can buy??

I am not seeing any,and how much will they cost?

I have been looking for them for months in the UK or even at German and French retailers which. I can only see a version of the Brix with half the eDRAM of the one Anandtech tested and it looks like it will be expensive. It will be close to £400 for the Core i5 with taxes and that is without memory or storage too,which will add another £100. So at least £500. It looks like the CPU on the cheapest version will be the Core i5 4570R with only 2/3 the L3 cache of the Core i7 4950Q,no HT and lower clockspeeds(2.7GHZ to 3.2GHZ) and only 64MB of eDRAM instead of 128MB.

Have you got some insider information from Intel you want to share??

I am looking for forward to the GT3e version of the Core i3 4330.

It would make a good secondary system to my main system with a low power Xeon E3.

Using US pricing.

Looking at the i7 model (4770R). This chip is basically as fast as the 4770.

Cost: $649

Buying components:
CPU: $ 316
So that is $333 for mobo, PSU, and case. Is there a little markup? Yes. However, its far smaller than you would think when you consider just how custom form factor is.

Comparing it to kaveri desktops. Iris will be comparable (+/-15%) but the 4770R is going to have roughly twice the CPU power.

Its expensive but you are paying for the form factor.

 

[PPB]

The usual iris pro sequence seems to be like this :

- usual funders bring up that iris pro beats igp performance of AMD apus
-another poster comes in with the usual "but it is a lot pricier"
-fuder counterattacks by claiming that iris pro can be had for just xxx price, which corresponds to the cheapest SKU with gt3e, but conveniently omits that this cheaper SKU performs worse than the iris pro SKU that was referred at first, neither that you probably won't ever be able to find one in the channel as it is a very limited SKU or is OEM only.
- another poster explains what I just did to the fuder, who then proceeds to hide under a rock for some time till most casual readers forgets the entire argument (or amds new product with igp gains come back to the discussion) only to come back AGAIN with the same repertoire of false claims and fallacies.

So to be clear, Iris Pro is a codename for a repertoire of SKUs that share the same number of EUs and a stacked piece of edram to act both as a be bottleneck alleviator for the igp and as a last level cache. Doest mean they share the same end performance because that is largely Dependant of clocks and turbo behavior, which actually is DIFFERENT between the iris pro skis mentioned.

So for god's sake, drop the iris pro FUD already, its price automatically makes it irrelevant for the larger part of the customers in the segment AMD usually places their apus. Nobody gives a damn about some BGA only part that can only be found in a couple of AIOs made by some company that is known to make overpriced products (which contradicts entirely the argument of the R series i5/7 being cheap). Iris pro is a niche product as the principal customer that made Intel designed such sku and bring it to the market in the first place.

Hope this isn't brought back ever again, at least till some broad well leaks show up and the iris pro hyping machine lights up back again.

 

[sniffin]

same could be said for amd or nvidia, they scale their cus up with new manu. process.

GCN and Kepler were not "scaled up" VLIW4 and Fermi. Nor will AMDs next gen be a scaled up GCN as we know it now, and nor will Maxwell be a scaled up Kepler.

Iris pro is just good.

No it isn't. No graphics parts that Intel makes are good. The CPUs they are attached to are excellent, but thats it.

snip

Pretty much

 

[monstercameron]

^that isn't quite fair, iris pro is a direct competitor, so comparisons a very valid. We know that kaveri is 10-20% faster over richland while iris was ~20-40% faster...this inst taking into account intels decent cpu and b/w lead. Kaveri might still not be as fast...

 

[Ventanni]

The usual iris pro sequence seems to be like this :

- usual fuders bring up that iris pro beats igp performance of AMD apus
-another poster comes in with the usual "but it is a lot pricier"
-fuder counterattacks by claiming that iris pro can be had for just xxx price, which corresponds to the cheapest SKU with gt3e, but conveniently omits that this cheaper SKU performs worse than the iris pro SKU that was referred at first, neither that you probably won't ever be able to find one in the channel as it is a very limited SKU or is OEM only.
- another poster explains what I just did to the fuder, who then proceeds to hide under a rock for some time till most casual readers forgets the entire argument (or amds new product with igp gains come back to the discussion) only to come back AGAIN with the same repertoire of false claims and fallacies.

So to be clear, Iris Pro is a codename for a repertoire of SKUs that share the same number of EUs and a stacked piece of edram to act both as a be bottleneck alleviator for the igp and as a last level cache. Doest mean they share the same end performance because that is largely Dependant of clocks and turbo behavior, which actually is DIFFERENT between the iris pro skis mentioned.

So for god's sake, drop the iris pro FUD already, its price automatically makes it irrelevant for the larger part of the customers in the segment AMD usually places their apus. Nobody gives a damn about some BGA only part that can only be found in a couple of AIOs made by some company that is known to make overpriced products (which contradicts entirely the argument of the R series i5/7 being cheap). Iris pro is a niche product as the principal customer that made Intel designed such sku and bring it to the market in the first place.

Hope this isn't brought back ever again, at least till some broad well leaks show up and the iris pro hyping machine lights up back again.

This. This. And this again.

Iris Pro is a fantastic product, not arguing that, but it's essentially a high price, low volume OEM solution who we all know its principal customer is. For the average Joe shopping parts at Fry's, waltzing into Best Buy to buy a desktop, or even browsing a selection of your typical laptop choices online, the Iris Pro isn't even an option.

 

[monstercameron]

This. This. And this again.

Iris Pro is a fantastic product, not arguing that, but it's essentially a high price, low volume OEM solution who we all know its principal customer is. For the average Joe shopping parts at Fry's, waltzing into Best Buy to buy a desktop, or even browsing a selection of your typical laptop choices online, the Iris Pro isn't even an option.

the same could said for many products with amd components, still doesn't invalidate it for comparison. I am saying this as an openly biased AMD apu fan. If AMD can't surpass intel with their massive gpu tech advantage then they are blind...then again the vast majority of the market won't care either way.

 

[bullzz]

@sniffin
"GCN and Kepler were not "scaled up" VLIW4 and Fermi" - same can be said for intel's "EU"s. they make arch changes to them and broadwell is going to see more than changes than before

"No it isn't. No graphics parts that Intel makes are good" - pretty sure AMD fans were saying this abt intel CPUs before conroe. iris pro is really neat concept and if intel can make edram mainstream with broadwell, the tables are going to turn

i have said this before. intel has no choice but to greatly increase igpu performance. with tablet processors doubling gpu capabilities every year, intel has to keep increasing. if not they will end will being a laughing stock when a smartphone performs better than their desktop igpu

I am actually surprised that 4670k can run BF4 at 19fps. thats not great performance but a big jump for intel's igpu capability even from 2 yrs ago

 

[ShintaiDK]

The Brix has 64MB:

http://techreport.com/news/25866/gigabyte-brix-pro-mini-pc-has-quad-core-haswell-with-gt3e-graphics.

Actually, it seems that the Brix Pro has 64MB eDRAM.


http://www.anandtech.com/show/7648/gigabyte-brix-pro/5

Odd, but as you say 128MB was overkill for Haswell anyway. Maybe halving the cache size will help lower costs.

Gigabyte statement is wrong. There are only 128MB models.

 

[ShintaiDK]

there are multipule, the die shot ( unless you think DDR3 interfaces have magically changed). or the fact that its stated that FM2 its using memory channels DCT 0 and 3 enabled.


But how about you answer my question!!!!!!! :whiste::whiste::whiste:


edit: also look at PS4 die shot from chip works..... the memory phy looks very similar........

edit2:

http://linux-kernel.2935.n7.nabble....ecoding-for-newer-Fam15h-models-td696358.html

Ever considered support for DDR4 for mobile SKUs?

 

[bullzz]

ya. imac uses the same chip and anand mentioned all had 128mb. athough im surprised how can OEMs get it wrong

 

[ShintaiDK]

^that isn't quite fair, iris pro is a direct competitor, so comparisons a very valid. We know that kaveri is 10-20% faster over richland while iris was ~20-40% faster...this inst taking into account intels decent cpu and b/w lead. Kaveri might still not be as fast...

Yep, and its all due to bandwidth.

AMD desperately needs eDRAM. They cant wait for stacked memory.

With eDRAM Kaveri would most likely be faster than Iris Pro. The DDR3 2133Mhz already add a 20-30$ premium vs DDR3 1600. So its not exactly cheap to configure a Kaveri/Richland design with reduced bandwidth issue.

 

[Abwx]

According to Hardware.fr Kaveri s memory controller
is GDDR5M compatible but because of Elpida s bankruptcy
there s no way to implement the technology since a second
source is necessary for supply reliability reasons.

http://www.hardware.fr/news/13530/ces-gddr5m-victime-deboires-elpida.html

 

[Homeles]

GDDR5M was a perfect fit for Kaveri. Shame it didn't happen.

 

[monstercameron]

Yep, and its all due to bandwidth.

AMD desperately needs eDRAM. They cant wait for stacked memory.

With eDRAM Kaveri would most likely be faster than Iris Pro. The DDR3 2133Mhz already add a 20-30$ premium vs DDR3 1600. So its not exactly cheap to configure a Kaveri/Richland design with reduced bandwidth issue.

I havent looked at ddr3-1600 but I bought a 8GB kit ddr3-2400 for ~$70USD

 

[ShintaiDK]

I havent looked at ddr3-1600 but I bought a 8GB kit ddr3-2400 for ~$70USD

1.5V DDR3 2400? Overvolted 1.65V memory doesnt count.

http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=100007611&IsNodeId=1&Description=DDR3%202400&name=Desktop%20Memory&Order=PRICE&Pagesize=20&isdeptsrh=1

DDR3 2400 will never be officially supported either since its not a JEDEC standard.

http://www.newegg.com/Product/Produc...CE&PageSize=20

As you can see, DDR3 2133Mhz 1.5V quickly passes the 100$ mark.

http://www.newegg.com/Product/Produc...CE&Pagesize=20

While there is a large selection of the 75$ DDR3 1600.

 

[AtenRa]

The Baidu preview of the AMD Kaveri mentioned that software drivers would be forthcoming to unlock the performance and the latest info indicates how much. Previous Musemage image manipulation bench of Kaveri and A10-6800 was basically tied at around 5300.

Now with update the performance is boosted by 30%-40% to 7529!

http://wccftech.com/amd-kaveri-a10-...g-pictured-bios-update-fm2-boards-compulsory/

The performance advantage keep growing in AMDs favor with a litho disadvantage. ouch

 

[AtenRa]

1.5V DDR3 2400? Overvolted 1.65V memory doesnt count.

http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=100007611&IsNodeId=1&Description=DDR3%202400&name=Desktop%20Memory&Order=PRICE&Pagesize=20&isdeptsrh=1

DDR3 2400 will never be officially supported either since its not a JEDEC standard.

http://www.newegg.com/Product/Produc...CE&PageSize=20

As you can see, DDR3 2133Mhz 1.5V quickly passes the 100$ mark.

http://www.newegg.com/Product/Produc...CE&Pagesize=20

While there is a large selection of the 75$ DDR3 1600.

Where AMD says that it only supports 2133MHz DDR-3 at 1.5v ???

 

[NTMBK]

Gigabyte statement is wrong. There are only 128MB models.

Both Anand and TechReport believe Gigabyte's statement, so I'm going to trust their judgement over yours. The Ark page for the 4770R doesn't specify L4 cache size sadly.

Besides, is it really such a crazy idea? Intel already admitted that they overspecified the eDRAM and Haswell doesn't need 128MB, why wouldn't they half the physical size of the eDRAM die and save themselves money without killing performance?

 

[ShintaiDK]

Where AMD says that it only supports 2133MHz DDR-3 at 1.5v ???

JEDEC standard is 1.35V and 1.5V.

 

[ShintaiDK]

Both Anand and TechReport believe Gigabyte's statement, so I'm going to trust their judgement over yours. The Ark page for the 4770R doesn't specify L4 cache size sadly.

Besides, is it really such a crazy idea? Intel already admitted that they overspecified the eDRAM and Haswell doesn't need 128MB, why wouldn't they half the physical size of the eDRAM die and save themselves money without killing performance?

They dont save anything since there is only a single eDRAM die. Also Apple say its 128MB. Do you think Apple get some other R chips than Gigabyte? Because then you are missing an S spec listing.

 

[Shivansps]

It makes no sence direct compare BGA CPU ark prices, end product price is what matters, for example, iMAC upgraded to GT3e whiout increasing the price.

You just cant say that Kaveri is good because Iris Pro its expensive, that is just people looking for excuses, specially on mobile, who cares about cpu price if im not buying a cpu im buying a end product?

So until Broadwell-K LGA is launched Kaveri cant be price compared to a Iris Pro, and on mobile you must compare products and them check the price/performance.

 

[ame132]

actually, if BRIX cpus really have only 64MB eDRAM, this could mean that Intel is eager to make best fitting L4, which means that broadwell can have more than 128MB if broadwell GT4 needs it

 

[AtenRa]

JEDEC standard is 1.35V and 1.5V.

Again, where does AMD says that it only supports DDR-3 2133MHz at 1.5v ???

 

[ShintaiDK]

A minor update on DDR4. And the future usage with IGPs. Its anything but pretty.

http://techreport.com/news/25893/micron-provides-ddr4-details-for-servers-high-end-desktops

The first DDR4 parts targeted at high-end desktop systems won't arrive until later this year, probably the fall.

At least initially, the desktop modules will come in 4GB and 8GB capacities. Micron has parts with 2133 MT/s and 2400 MT/s transfer rates right now.

Looking forward, Micron expects DDR4 memory to move into notebooks and more conventional desktop rigs in late 2015. That time frame could move forward or back, the company says.

 

[ShintaiDK]

Again, where does AMD says that it only supports DDR-3 2133MHz at 1.5v ???

Nobody talked about DDR3 2133 at any voltage.

Plus you know as well as everyone else that no OEMs will touch anything that isnt a standard. Dell/HP/Lenovo/Apple is not gonna sell office and home PCs with overvolted memory. And only a fool would buy overvolted memory anyway. And that includes 1.35V chips overvolted and overclocked to handle 2133. OEMs dont want they either.

To use (proper) 2133Mhz memory you simply have to pay extra. And its not cheap.