AMD Carrizo Pre-release thread

[Abwx]

Those are 15W TDP scores and not 35W TDP ???

Do we have a link for that ??

I just edited my previous post...

Note 6.

http://wccftech.com/amd-bristol-ridge-apu-computex-launch/

 

[Masochrist_]

Those values don't add up. Checking this page shows that a 15 W 8800P scores around 207 in C15 MT. What that image shows is in fact the MT performance for 15W BR and Carrizo, not 35W. A 8800P with that TDP scores 277.

Furthermore, check out the sixth endnote. It's clear they only compare 15W SKU's when it comes to multi-threaded performance.

However, the ST performance seems to add up from checking the first link. Quite a bit higher than Carrizo, and that is without considering that the top of the line BR (9830P) has got a higher base and turbo clock than the tested 9800P (if they indeed compared the 9800 against the 8800 in ST performance).

Could be wrong, but doesn't any 8800P SKU max out at 3.4 GHZ, no matter its TDP ?

 

[AtenRa]

I just edited my previous post...

Note 6.

http://wccftech.com/amd-bristol-ridge-apu-computex-launch/

thanks

 

[AtenRa]

Those values don't add up. Checking this page shows that a 15 W 8800P scores around 207 in C15 MT. What that image shows is in fact the MT performance for 15W BR and Carrizo, not 35W. A 8800P with that TDP scores 277.

Furthermore, check out the sixth endnote. It's clear they only compare 15W SKU's when it comes to multi-threaded performance.

However, the ST performance seems to add up from checking the first link. Quite a bit higher than Carrizo, and that is without considerig that the top of the line BR (9830P) has got a higher base and turbo clock than the tested 9800P (if they indeed compared the 9800 against the 8800 in ST performance).

Could be wrong, but doesn't any 8800P SKU max out at 3.4 GHZ, no matter its TDP ?

thanks, somehow i missed this

 

[DrMrLordX]

Carrizo has Cinebench R10 scores similar to ~3.4 GHz Haswell. No surprise that it comes close in R15 as well. R11.5, however . . .

 

[frozentundra123456]

Just realized that 35W Bristol at 28nm HDL has the same performance as the 22nm Haswell Core i3 4000M at 37W TDP in Cinebench R15 both Single and Multi.

http://www.hardwarecanucks.com/foru...md-details-bristol-ridge-am4-performance.html

Core i3 3110M 35W TDP 22nm

http://www.notebookcheck.net/Intel-Core-i3-3110M-Notebook-Processor.74459.0.html
Cinebench R15
Single = 81,5
Multi = 209

Core i3 4000M 37W TDP 22nm

http://www.notebookcheck.net/Intel-Core-i3-4000M-Notebook-Processor.93560.0.html
Cinebench R15
Single = 94,2
Multi = 239,8

That i3 is bottom of the barrel in both efficiency and performance. i5 6500u scores 120 in a 15 watt tdp. Maybe AMD is catching up though. At least now you are comparing to haswell instead of ivy bridge.

 

[dark zero]

That i3 is bottom of the barrel in both efficiency and performance. i5 6500u scores 120 in a 15 watt tdp. Maybe AMD is catching up though. At least now you are comparing to haswell instead of ivy bridge.

M tier is NOT a bottom of the barrel compared to the U chips.

Make it worse? THERE IS NOT ANY i5 6500U, is an i7 6500U! Which cost way more that the i5 one.
The i5u SKYLAKE only gets up to 112 and is NOT sustained compared to M tier chips.

 

[deasd]

Enhancement in cinebench might due to much efficient power management.
Looking forward to gaming performance test using dual-channel DDR4......

 

[AtenRa]

That i3 is bottom of the barrel in both efficiency and performance. i5 6500u scores 120 in a 15 watt tdp. Maybe AMD is catching up though. At least now you are comparing to haswell instead of ivy bridge.

Well, the performance is extremely close to 14nm Skylake Core i3 6100U. This is an engineering marvel, 28nm CPU almost equals 14nm CPU at the same TDP in a very Intel optimized benchmark. If BristolRidge was made at 14nm LPP the performance/watt would skyrocket.

http://www.notebookcheck.net/Intel-Core-i3-6100U-Notebook-Processor.149441.0.html

Core i3 6100U 15W TDP 14nm

Cinebench R15
Single = 97
Multi = 247

 

[The Stilt]

Well, the performance is extremely close to 14nm Skylake Core i3 6100U. This is an engineering marvel, 28nm CPU almost equals 14nm CPU at the same TDP in a very Intel optimized benchmark. If BristolRidge was made at 14nm LPP the performance/watt would skyrocket.

http://www.notebookcheck.net/Intel-Core-i3-6100U-Notebook-Processor.149441.0.html

Core i3 6100U 15W TDP 14nm

Cinebench R15
Single = 97
Multi = 247

Yeah, but Carrizo has significantly more resources. In Cinebench XV has ~91.62% CMT yield, while the SMT in Broadwell (and most likely Skylake) have ~26.5% yield.

Skylake = 2.53 cores effectively
Excavator = 3.83 cores effectively

Carrizo should score < 51% higher in multitheaded CB R15 than Skylake 2C/4T part at the same clocks, to be as fast (which it definitely isn't).

If you wish to really compare the apples to the apples, then you must test with CMT and SMT technologies disabled on both designs.

However I do agree that those results posted for Carrizo are completely *ucked up:

15/25W = 215
20/25W = 217
25/25W = 238
30/30W = 255
35/42W = 267

Those are the correct values for XV.

 

[frozentundra123456]

Well, the performance is extremely close to 14nm Skylake Core i3 6100U. This is an engineering marvel, 28nm CPU almost equals 14nm CPU at the same TDP in a very Intel optimized benchmark. If BristolRidge was made at 14nm LPP the performance/watt would skyrocket.

http://www.notebookcheck.net/Intel-Core-i3-6100U-Notebook-Processor.149441.0.html

Core i3 6100U 15W TDP 14nm

Cinebench R15
Single = 97
Multi = 247

Those cinebench results in the hardware canuks article are at 35watts. It is shown as 35 watts on the graphs and stated very clearly again in the text. Additionally, they were not even verified by the author, but were numbers provided by AMD.

 

[The Stilt]

The number in HardwareCanucks write-up are otherwise fine and correct, besides the Cinebench R15 MT results.

Neither Bristol Ridge nor Carrizo are running at 35/42W limit in that one. If they are then the chip themselves or their motherboards are overheating and the chip are throttling. Those numbers they provided for "35W cTDP" configuration are too well matched with the numbers these chips should provide at the default 15/25W limit for it to be a coincidence.

 

[AtenRa]

Yeah, but Carrizo has significantly more resources.

A single Skylake Core + HT is as wide or wider than a single Excavator MUDULE.

If you wish to really compare the apples to the apples, then you must test with CMT and SMT technologies disabled on both designs.

So apples to apples will be to compare a 2x ALUs of a single Excavator CORE vs 4x ALUs of a single Skylake Core ??

Apples to apples is comparing a single SMT Core vs single CMT Module.

Intel Skylake SMT has the Single Thread advantage and AMD CMT Module has the Throughput advantage.

But the eye opener here is the SINGLE THREAD CB R15 performance at 15W TDP. Excavator scores almost the same as Skylake Core i3.

However I do agree that those results posted for Carrizo are completely *ucked up:

15/25W = 215
20/25W = 217
25/25W = 238
30/30W = 255
35/42W = 267

Those are the correct values for XV.

AMD slides says @15W TDP for BristolRidge.

http://images.anandtech.com/galleri...e-007.jpg?_ga=1.7364530.1459516107.1438938042

 

[AtenRa]

Those cinebench results in the hardware canuks article are at 35watts. It is shown as 35 watts on the graphs and stated very clearly again in the text. Additionally, they were not even verified by the author, but were numbers provided by AMD.

The CB R15 scores are for 15W TDP taken from the AMD slide bellow.

http://images.anandtech.com/galleri...e-007.jpg?_ga=1.7364530.1459516107.1438938042

 

[The Stilt]

AMD slides says @15W TDP for BristolRidge.

http://images.anandtech.com/galleries/4796/7th%20Generation%20AMD%20A-Series%20Processors%20Pre-Announce%20Press%20Deck%20%281%29-page-007.jpg?_ga=1.7364530.1459516107.1438938042

On all Excavator based designs with default TDP of 15W the boost limit is always 25W, unless STAPM is disabled (by OPN config). All Excavators have STAPM enabled by default and the STAPM period is 200 seconds.

That's why there is a drop in performance if the workload / benchmark takes more than 200 seconds to complete.

Likewise the Bristol Ridge parts which come with 35W TDP will have 42W boost limit for 200 seconds.

 

[The Stilt]

To be exact, Excavator designs can be programmed for six different power configurations from factory. There are also three different cTDP mode options, depending on SKU.

cTDP modes are: Unlimited (10-35W), Low (12-15W) and High (25-45W).

All Carrizo (CZ-A1) based mobile & embedded OPNs use the "unlimited" cTDP (10-35W) option. Bristol Ridge xx00 models use "Low" (12-15W) configuration and the xx30 models use "High" (25-45W) configuration.

TDP / cTDP configurations are:

- 10W (TDP), 16W (boost)
- 12W (TDP), 20W (boost)
- 15W (TDP), 25W (boost)
- 25W (TDP), 42W (boost)
- 35W (TDP), 42W (boost)
- 45W (TDP), 54W (boost)

The default STAPM duration is 200 seconds (boosted power limit), but it can be increased if the system manufacturer wishes to do so.

 

[frozentundra123456]

The CB R15 scores are for 15W TDP taken from the AMD slide bellow.

http://images.anandtech.com/galleri...e-007.jpg?_ga=1.7364530.1459516107.1438938042

That is not what the article says or the graphs show.

Directly copied and pasted from the posted link:

"There is however a small caveat here: all of the solutions were pushed to their 35W maximum so it is quite likely they’d still be overcome by the 15W i7-6500U from previous benchmarks which levels out around the 127 mark."

But in any case, I dont think you can accurately compare AMD provided benchmarks on who knows what kind of platform to Notebook check data from a real world shipping platform.

 

[Abwx]

That is not what the article says or the graphs show.

Directly copied and pasted from the posted link:

"There is however a small caveat here: all of the solutions were pushed to their 35W maximum so it is quite likely they&#8217;d still be overcome by the 15W i7-6500U from previous benchmarks which levels out around the 127 mark."

But in any case, I dont think you can accurately compare AMD provided benchmarks on who knows what kind of platform to Notebook check data from a real world shipping platform.

They are wrongly interpreting the slides note, 35W is only for PCMark comparison with a Richland A10 5750M chip and this is specified in note 7.

The CB tests are related by note 6 wich state explicitely 15W, but of course you are free to believe Hardwarecanucks spin if ever AMD s stated numbers dont fit your understanding...

http://images.anandtech.com/galleri...e-007.jpg?_ga=1.7364530.1459516107.1438938042

 

[The Stilt]

Here's an illustration about the STAPM behavior.

At ~200 second mark you see the power consumption starts to taper down, and after 20 seconds it has reached it's final level.

The clocks naturally taper in a similar manner, from ~ 2800MHz down to 2300MHz.

I recorder it during X265 (1.9 build 125) encoding process.

Completely default configuration, 15/25W TDP, 200s STAPM duration and 100% scalar.

 

[Abwx]

How is the power measured in the graph above..?.

 

[Dresdenboy]

The Stilt, thanks for all your detailed data like this chart or the GB scores (1ch/2ch delta for the same system)! This helps understanding the chips better incl. the chance to derive something for Zen features.

 

[The Stilt]

How is the power measured in the graph above..?.

They are read from the PMC using tools intended for evaluation. There are hundreds of parameters that can be monitored and logged, and they are accurate. That's the reason why Excavator can be as power efficient.

 

[monstercameron]

They are read from the PMC using tools intended for evaluation. There are hundreds of parameters that can be monitored and logged, and they are accurate. That's the reason why Excavator can be as power efficient.

What generation of intel hardware is this close to in terms of efficiency?
Also do have any idea how much changed with br eith the tweaked manu process?

Sent from my SM-G930T using Tapatalk

 

[AtenRa]

nice info, thanks :thumbsup:

 

[The Stilt]

What generation of intel hardware is this close to in terms of efficiency?
Also do have any idea how much changed with br eith the tweaked manu process?

Sent from my SM-G930T using Tapatalk

The designs are way too different for any kind of proper comparison. I haven't worked too much on recent Intel designs, but I would say AMD has better power management than the most recent Intel designs do. That's the result of being so far behind in both the process technology and the design itself, however that's definitely not a bad thing.

Once AMD has access to smaller manufacturing process (such as the 14nm LPP) their power efficiency will be pretty hard to compete with. Let's just hope the new architechture lives up to the expectations.

While the power management in Carrizo for example is extremely advanced and effective, there are still some things which need to be improved heavily. Basically none of the features of the power management are documented, let alone properly. Because the lack of documentation, the system manufacturers are unable to tune the power management properly to provide the best possible performance on their system.

I have personally spent maybe < 800 hours on the SMU, PMU and PMC on Steamroller and Excavator. Without the breakthroughs I made on Steamroller (i.e learnt how to program and configure the SMU) I couldn't change a single setting either on Steamroller or Excavator, despite having access to the same documents as the manufacturers do. Most of the time the stuff just isn't documented at all and when it is, it is done either poorly or just wrong.