Kabini Rumors

[Abwx]

@AbwX

Think he ment Dirt 3 showdown. From this link to the 650m review:
http://www.notebookcheck.net/NVIDIA-GeForce-GT-650M.71887.0.html

If you hover you mouse cursor over the text, it ll show resolution and settings.

650M GPU:

Dirt 3:
(low) 1024x768 Ultra low preset = ~89 fps
(med) 1366x768 Medium preset = ~65 fps
(high) 1366x768 High preset + 2xMSAA = ~51 fps
(ultra) 1920x1080 Ultra preset + 4xMSAA = ~18 fps

Its not Apples-to-Apples.

The "video" Intel has of Haswell running Dirt3, is at 1920x1080 + High preset + no AF/AA.

Haswell:
http://www.youtube.com/watch?feature=player_embedded&v=VPrAKm7WtRk

Anyways at 1920x1080 + High Quality + no AF/AA,
a A10-5800k with its IGP@1050mhz can do ~45-55 fps too.

Thanks for the clues , i guess that we ll have to wait for extensive
benchs to sort out who is capable of what....

 

[CTho9305]

Jaguar 2.0 is expected to launch 9 months after Jaguar. Then 5-6 months after that we will see Leopard.

Source?

 

[Pilum]

Anything with a GT3e should be priced solidly out of Richland's range. The CPUs with it are all quadcore monsters, plus it's Intel. I wouldn't expect the two to compete at all.

AFAIK there will be i5 and i3 SKUs with the normal GT3, 40EU without embedded RAM. These might compete with Richland at all price ranges. But we'll have to wait for details and of course performance with final drivers.

 

[Sleepingforest]

Oh yeah. The Wikipedia article only lists the i7 Haswells, so you are right AFAIK.

 

[Enigmoid]

At what res.?..

Their claim is BS , typical marketing.

At least AMD gives 3DMark scores of their future products
while with intel we have to believe their sayings at face value
without even the slightest number just to make people think
that this is really wonderfull....

3D mark is hardly a worthwhile metric.

If you look at trinity (mobile a10-4600m) you will notice that in games it performs on average about 20% worse than its 3dmark score would suggest (because in 3d mark the cpu load is very low so it can turbo quite a bit, in games the gpu is lower clocked). Hopefully Richland can fix this.

(Note that the a10-4600m gets around 1150 3dmark gpu score and the desktop gets around 1500 but on average the a10-4600m is only about 2/3 the desktop performance because playing games the gpu is downclocked relative to 3dmark).

Thats why I want to see real benchmarks, not more 3dmark scores.

(Though the fact that they release the numbers is nice).

 

[Enigmoid]

AFAIK there will be i5 and i3 SKUs with the normal GT3, 40EU without embedded RAM. These might compete with Richland at all price ranges. But we'll have to wait for details and of course performance with final drivers.

I don't think gt3 will be available for the desktop outside of 'special' chips.

 

[IntelUser2000]

I'm surprised that a 1 Ghz Bobcat was actually competitive to 1.8 Ghz Atom in application performance (a lot of help from a very fast SSD), I thought it would be a lot worse. They only did test PC-mark though, which makes me suspicious, given it's very storage-sensitive.

Actually, the Z2760 is 30% faster in Cinebench.

 

[IntelUser2000]

Also here confirmed by Intel: http://www.mobilegeeks.com/intels-clover-trail-heading-tablets/

It is, but only for Android, just like how Medfield is in a Tablet(relatively unknown though).

For Windows, its Clover Trail until Bay Trail.

 

[IntelUser2000]

50-60% higher overall performance with a 40% increase in clock speed and a move from 2 cores with HT to 4 cores without would actually suggest a major drop in IPC. If we're talking about typical performance scenarios, HT on Atom will tend to give you somewhere around 40% better performance with two fully loaded threads where two full cores will have given you 100% better performance.

Well, the most recent Intel statement says(official too!) 2x or more for Bay Trail. Even if we assume Bay Trail is at 2.1GHz, the clock speed gain is actually only 10-20% because Burst Mode on Clover Trail allows for 1.8GHz frequency, and it can sustain that quite well.

The better comparison seems to be what they are claiming for Merrifield. 50-60% gain over Clover Trail+, but its known to be a dual core initially.

 

[MightyMalus]

Last I checked, GT3 won't be in any desktop at all. Unless you count All-In-Ones as Desktops?

 

[Sleepingforest]

You are correct. There are no plans for GT3 in desktop Intel chips, since there is basically no reason to do so (at least, for enthusiasts).

 

[monstercameron]

kabini soc is ft3? wait I though only temash was soc.

http://www.engadget.com/2013/04/23/amd-reveals-g-series-x-embedded-chips/

The low-power x86 microprocessor class includes: GX-420CA @ 25W TDP (scored 19); GX415GA @ 15W (25), GX217GA @ 15W (17), GX210HA @ 9W (20), G-T56N @ 18W (12), G-T52R @ 18W (7), G-T40N @9W (14), G-T16R @ 4.5W (19), Intel Atom N270 @ 2.5W (20), Intel Atom D525 @ 13W (9), Intel Atom D2700 @ 10W (12) & Intel Celeron G440 @ 35W (5). Performance score based on an average of scores from the following benchmarks: Sandra Engineering 2011 Dhrystone ALU, Sandra Engineering 2011 Whetstone iSSE3, 3DMark® 06 (1280 x 1024), PassMark Performance Test 7.0 2D Graphics Mark, and EEMBC CoreMark Multi-thread.

Calculation based on performance of GX-420GA GPU running at 600MHz = 0.6 GHz. 0.6 x 256 FLOPs = 153.6 GFLOPS

15W kabini gpu competes with trinity ulv (The 18W E2-1800 has much lower perf than the a6-4455m in both cpu and gpu at the same tdp)

Models and pricing

Models available at launch include:

• GX-420CA SOC with AMD Radeon™ HD 8400E Graphics
  • Quad-core, 25W TDP, CPU freq. 2.0GHz, GPU freq. 600MHz
• GX-415GA SOC with AMD Radeon™ HD 8330E Graphics
  • Quad-core, 15W TDP, CPU freq. 1.50GHz, GPU freq. 500MHz
• GX-217GA SOC with AMD Radeon™ HD 8280E Graphics
  • Dual-core, 15W TDP, CPU freq. 1.65GHz, GPU freq. 450MHz
• GX-210HA SOC with AMD Radeon™ HD 8210E Graphics
  • Dual-core, 9W TDP, CPU freq. 1.0GHz, GPU freq. 300MHz
• GX-416RA SOC
  • Quad-Core, 15W, CPU Freq. 1.6GHz, No GPU

Pricing ranges from $49 - $72 for the SKUs.

super cheap pricing...weren't prices >$100 mentioned on these forums?

http://www.youtube.com/watch?v=LCFLqBgX4Ac&feature=youtu.be

intro video by amd

 

[Haserath]

Last I checked, GT3 won't be in any desktop at all. Unless you count All-In-Ones as Desktops?

There were rumors about GT3 chips being sold as soldered desktop chips, I think.

 

[Exophase]

Well, the most recent Intel statement says(official too!) 2x or more for Bay Trail. Even if we assume Bay Trail is at 2.1GHz, the clock speed gain is actually only 10-20% because Burst Mode on Clover Trail allows for 1.8GHz frequency, and it can sustain that quite well.

Still way too ambiguous for my tastes. Is it single threaded, peak throughput, or maybe something in between? If it's full on peak throughput in something well threaded you would expect at least a 40% improvement moving from HT to real cores. Add 17% clock minimum and you get 66%. You only need another 20% boost from IPC to get you up to 2x, which is definitely within the realm of expectations (I'd expect a bit higher).

The opposite extreme is full single-threaded where the IPC gain needs to be 43% coming from 1.5GHz (pushing past high end of expectations) to 71% for 2.1GHz (way past realistic for me).

For now my gut says something in between. Some benchmark with moderate but not ideal scaling.

This is of course taking twice as fast to really mean something very close to just that and not a very rough approximation, as such things often are.

On clock speed, I'm still remembering those slides that compared Clover Trail with Bay Trail and listed frequency as 1.5GHz vs 2.1GHz, who knows what really applies for the comparison they made...

 

[piesquared]

kabini soc is ft3? wait I though only temash was soc.

http://www.engadget.com/2013/04/23/amd-reveals-g-series-x-embedded-chips/

15W kabini gpu competes with trinity ulv (The 18W E2-1800 has much lower perf than the a6-4455m in both cpu and gpu at the same tdp)

super cheap pricing...weren't prices >$100 mentioned on these forums?

http://www.youtube.com/watch?v=LCFLqBgX4Ac&feature=youtu.be

intro video by amd

http://www.engadget.com/2013/04/23/amd-reveals-g-series-x-embedded-chips/

AMD reveals G-Series X embedded chips, drops a little ARM-powered bombshell

Roy Taylor was right, this could be another Apple style turnaround. AMD have some outstanding products on and coming to the market that largely mitigate any process node advantage of intel (although that node advantage could be narrowing significantly). If these little silicon gems are any inndication of what is coming, it looks awesome.

 

[NTMBK]

Source?

It's Seronx, he doesn't do sources... :awe:

 

[AtenRa]

http://www.amd.com/us/press-releases/Pages/amd-embedded-gseries-2013apr23.aspx

Models available at launch include:

• GX-420CA SOC with AMD Radeon™ HD 8400E Graphics
  • Quad-core, 25W TDP, CPU freq. 2.0GHz, GPU freq. 600MHz
• GX-415GA SOC with AMD Radeon™ HD 8330E Graphics
  • Quad-core, 15W TDP, CPU freq. 1.50GHz, GPU freq. 500MHz
• GX-217GA SOC with AMD Radeon™ HD 8280E Graphics
  • Dual-core, 15W TDP, CPU freq. 1.65GHz, GPU freq. 450MHz
• GX-210HA SOC with AMD Radeon™ HD 8210E Graphics
  • Dual-core, 9W TDP, CPU freq. 1.0GHz, GPU freq. 300MHz
• GX-416RA SOC
  • Quad-Core, 15W, CPU Freq. 1.6GHz, No GPU

Pricing ranges from $49 - $72 for the SKUs.

And here are the the Brazos 2.0

http://www.amd.com/us/products/notebook/Pages/consumer-notebooks.aspx#7

C-60, 1.333/1.0GHz, 2 cores, 9W, 1MB, 80 @ 400/275MHz

E2-1800, 1.7GHz, 2 cores, 18W, 1MB, 80 @ 680/523 MHz

So, let me get this straight.

The new 28nm Kabini Dual Core GX-210HA will be slower than the 40nm Brazos 2.0 C-60 at the same TDP??
And the dual core GX-217GA will have the same performance with the E2-1800 ??

Also,

GX-416RA SOC

• Quad-Core, 15W, CPU Freq. 1.6GHz, No GPU

15W TDP without the iGPU at only 1.6GHz ?? WTF ???



It seems to me those are the base frequencies only. If turbo frequencies doesn’t scale much higher then i don’t see those APUs have a bright future (unless they are holding them back)

 

[Abwx]

So, let me get this straight.

The new 28nm Kabini Dual Core GX-210HA will be slower than the 40nm Brazos 2.0 C-60 at the same TDP??

Not that straight , you did forget that Kabini has integrated
chipset wich add to the TDP so theses numbers are not
directly comparable....

 

[Arkadrel]

The new 28nm Kabini Dual Core GX-210HA will be slower than the 40nm Brazos 2.0 C-60 at the same TDP??

Have to remember it has the FCH integrated into it.
The C-60 had a "FCH A50M (Hudson M1)" that took up a additional ~4watts or something.

So its more like:

C-60 = 9w + 4watts : 13w total
GX-210HA = 9w : (-44% less power)

Both are Dual cores 1ghz, but the GX-210HA is supposed to have like a ~15% IPC increase.
So the GX-210HA will be the faster of the two, CPU wise.

GPU wise.... theyre all supposed to have 2CU of GCN right?
128x300mhz (GX-210HA) vs 80x276mhz (C-60) = winner GX-210HA by large margin.

I dont know I still think it ll be good.

 

[Abwx]

Doesnt kabini have a dual channel memory controler btw ?..

Brazos had a single channel IIRC.

Edit : Overall TSMC s bulk looks quite average
not to say crappy , there s no doubt that
the much trashed GF has better process.

 

[Abwx]

C-60 = 9w + 4watts : 13w total
GX-210HA = 9w : (-44% less power)

C-60 = 9w + 4watts : 13w total = 44% more power than 9W
GX-210HA = 9w = 30% less power than 13W

 

[Dresdenboy]

super cheap pricing...weren't prices >$100 mentioned on these forums?

I didn't read every related thread in full, but I can't imagine someone expecting such high prices for a evolutionary Bobcat successor. For Bobcat I remember price speculations in the $20-$30 range.

 

[AtenRa]

Not that straight , you did forget that Kabini has integrated
chipset wich add to the TDP so theses numbers are not
directly comparable....

Have to remember it has the FCH integrated into it.
The C-60 had a "FCH A50M (Hudson M1)" that took up a additional ~4watts or something.

So its more like:

C-60 = 9w + 4watts : 13w total
GX-210HA = 9w : (-44% less power)

Both are Dual cores 1ghz, but the GX-210HA is supposed to have like a ~15% IPC increase.
So the GX-210HA will be the faster of the two, CPU wise.

GPU wise.... theyre all supposed to have 2CU of GCN right?
128x300mhz (GX-210HA) vs 80x276mhz (C-60) = winner GX-210HA by large margin.

I dont know I still think it ll be good.

Haven’t forgotten the FCH but, have a look at the iGPU naming,
Assuming the HD8400E has 128 (GCN) Radeon Cores @ 600MHz, the HD8210E must have 64 (GCN) Radeon cores @ 300MHz.

Now if that’s the case, the 28nm dual core Kabini @ 1GHz with half the iGPU(64) will have the same TDP as a 40nm Dual core @ 1GHz + 80 (VLIW5) Radeon Cores. The performance difference in CPU will only be 15% and something about the same in iGPU.

This is hardly what you expect from a node shrink + Architectural enhancements (unless we're talking Intel).

Ps: Even if you had a separately FCH, at 28nm it would have half the TDP of the 40nm FCH(or A50M was it 65nm ??).

 

[MightyMalus]

I wish AMD would just go massively into cores. 16 core Jaguar non BGA(!) 100W-125W, anyone?

Someone mentioned a "Leopard", as the next step for the cat family. Any news or source on that? Did a quick search and didn't find anything.

 

[lagokc]

I wish AMD would just go massively into cores. 16 core Jaguar non BGA(!) 100W-125W, anyone?

Someone mentioned a "Leopard", as the next step for the cat family. Any news or source on that? Did a quick search and didn't find anything.

Probably no Leopard, don't want to violate Apple's trademark. After all everyone knows Apple invented Leopards. :awe:

A 16 core Jaguar would be pretty terrible at most modern workloads which aren't highly parallel. On the other hand a 4-core Jaguar should be both enough for non-gamer/non-powerusers and also a very tiny chip that would be very cheap for AMD to manufacture.