AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

[frozentundra123456]

People eyes hurt when they read AT ipc test because it doesnt fit the dominating discourse. But it should be mandatory reading even if its kind of sad.

The reality is IPC have hit the wall with haswell and i am pretty sure nearly all design choices now centers around better efficiency.

Edit: and if someone wonders what the 6% ipc uplift from hsw to skl did for discrete card gaming AT also did test that:
http://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/10
"There is no easy way to write this...decreases"

Yes, people's eyes hurt when they read it because it was one of the worst reviews ever put out by AT. A brand new architecture gimped with slow ram, and "cpu" gaming tests that were mostly gpu limited.

 

[Arachnotronic]

Yes, people's eyes hurt when they read it because it was one of the worst reviews ever put out by AT. A brand new architecture gimped with slow ram, and "cpu" gaming tests that were mostly gpu limited.

Yep. Frankly, a better way to have tested IPC would have been to run Geekbench 4 and look at the sub-test results. I have actually done that and compared Haswell, Broadwell, and Skylake -- Skylake is a bigger jump over Broadwell than Broadwell was over Haswell.

 

[itsmydamnation]

Skylake is *definitely* bottlenecked by DDR4-2133. I think it was RS who posted some review showing huge gains in moving even to DDR4-4000(!).

Its not bandwidth its Latency and every core in the world will perform better with faster memory access times ,
ddr4 4000 with CAS 19* = 9.5ns
ddr3 3400 with CAS 16 *= 9.5ns

ddr4 2133 ( anadtech test) with CAS14* = 15ns
ddr3 1866 (anandtech test) with CAS9* = 10.5ns

Thats why the results didn't look good. To get a fair test uarch to uarch access times need to be the same. But in that anandtech review every cache miss on the DDR3 system was being fetched around 33% faster then on the DDR4 ssytem

*All CAS latency numbers taken from "best" corsair memory of that type+speed

 

[Nothingness]

Yep. Frankly, a better way to have tested IPC would have been to run Geekbench 4 and look at the sub-test results. I have actually done that and compared Haswell, Broadwell, and Skylake

Yeah GB4 is good for that kind of comparison. Did you extend your comparison to older chips? Or did you stick to CPU you own?

Skylake is a bigger jump over Broadwell than Broadwell was over Haswell.

This is to be expected since Skylake is a micro architecture step while Broadwell was more a process step (and not a good one).

 

[DrMrLordX]

ddr4 2133 ( anadtech test) with CAS14* = 15ns
ddr3 1866 (anandtech test) with CAS19* = 10.5ns

Sure you don't mean CAS9 here? I don't think I've ever seen DDR3 with CL/CAS 19.

 

[itsmydamnation]

Sure you don't mean CAS9 here? I don't think I've ever seen DDR3 with CL/CAS 19.

why yes, yes i do

 

[Arachnotronic]

Yeah GB4 is good for that kind of comparison. Did you extend your comparison to older chips? Or did you stick to CPU you own?

I have SandyB and IvyB chips that I've been meaning to test for some time, but haven't gotten around to it! I'll be buying an AM4 board and Bristol & Summit Ridge chips whenever they happen to become available as boxed CPUs to test, too.

This is to be expected since Skylake is a micro architecture step while Broadwell was more a process step (and not a good one).

Yep, Broadwell was just a big mess for Intel in general. Meagre IPC gain, clock regression, late to market, etc. Just terrible execution from Intel on Broadwell and the 14nm process, too.

 

[krumme]

I know its difficult to wake up.
Surely there is a latency part here but it doesnt play the part your brain wants it to.

May i remind you the skl performed at average better with the ddr4 system than with the ddr3?

Broadwell to Skylake (DDR3): Average ~2.4% Up
Broadwell to Skylake (DDR4): Average ~2.7% Up

Hello. Difficult facts call brain.

Its relatively fast ddr3 and surely expensive ddr4 3000 low latency will widen the gab but its also a different cost and not comparable.

 

[Arachnotronic]

I know its difficult to wake up.
Surely there is a latency part here but it doesnt play the part your brain wants it to.

May i remind you the skl performed at average better with the ddr4 system than with the ddr3?

Broadwell to Skylake (DDR3): Average ~2.4% Up
Broadwell to Skylake (DDR4): Average ~2.7% Up

Hello. Difficult facts call brain.

Its relatively fast ddr3 and surely expensive ddr4 3000 low latency will widen the gab but its also a different cost and not comparable.

Uh huh. Can't possibly be that AT's testing doesn't give us a true picture of the Skylake core's perf/MHz improvement over Haswell.

 

[Thunder 57]

Yep. Frankly, a better way to have tested IPC would have been to run Geekbench 4 and look at the sub-test results. I have actually done that and compared Haswell, Broadwell, and Skylake -- Skylake is a bigger jump over Broadwell than Broadwell was over Haswell.

Well that should be expected. Haswell and Skylake are both tocks. Broadwell was not much of an improvement; Even less so than Ivy Bridge. I still think it may be safe to say Ivy -> Haswell was the last double digit gain. Broadwell -> Sky Lake looks like less an IPC increase.

 

[krumme]

The review is from aug 2015 and since then ddr4 prices have fallen quite a lot. At that time the comparison gave good sense. And for our purpose its perfectly fine as it shows us what ipc benefits the core gives us on the same memory type. 2.4% on ddr3 1866 cas 9.
One can say thay today ddr4 3000 is low price so the platform should be compared using that. Fine. You dont buy a cpu you buy performance but dont fool yourself thinking its solely the new ice lake giving the benefit so to speak.
The point is the ipc of the cpu have more or less stopped and we get our gain from everywhere else. Memory subsystem. More cores due to efficiency gains. Fine.
I just posted in the bf1 thread under games that people should remember to use dual channel and 1866 and up ram. We have a lot of people running single channel on eg skl i5. Go look at the new 1070 laptop review at AT. Single channel. Thats what comes on the market when focus is dead wrong.

 

[Whaeveva]

I hope to goodness the OP is a lie. I have no plans to buy any new Intel technology, because they serve it out in teaspoons.I'm just tired of it.

Also all the newer stuff post Ivy-Bridge is very fragile and prone to failure.

 

[Sweepr]

I know its difficult to wake up.

There's plenty of reviews that showed bigger gains for Skylake than what we saw in AnandTech's review. Instead of telling people to 'wake up' you should be looking at more than a single piece of data.

- Eurogamer:

Core i5 6600K vs Core i5 4690K (same 3.5-3.9GHz base/turbo)
- 17% faster @ The Witcher 3
- 11% faster @ GTA V
- 10% faster @ Battlefield 4

- Hardware.fr

- PCLab

Core i5 6600K vs Core i5 4690K at fixed 4.5GHz, Skylake-S is:
16.2% faster @ Battlefield 4 MP
10.9% faster @ Counter Strife Global Offensive
6% faster @ Crysis 3
22.7% faster @ Far Cry 4
13% faster @ GTA V
12.5% faster @ The Witcher 3
9.2% faster @ Watch Dogs
14.7% faster @ Project Cars
14.8% faster @ Starcraft 2
24.5% faster @ Total War Attila

Also AnandTech's results didn't match those of any other website:

7-Zip 9.2
- AnandTech: Skylake 1% slower than Haswell per clock
- Hardware.fr: Skylake 7% faster than Haswell per clock
- Hardware Canucks: Core i7 6700K 11.8% faster than Core i7 4790K with lower turbo clocks

POV-Ray
- AnandTech: Skylake 9.7% faster than Haswell per clock
- HardOCP: ''POV-Ray is a ray tracing for creating high quality graphics. We are using the benchmark included in the software and using its multicore ability. Again we see Skylake stretch its legs. Skylake rewards us with a 16% decrease in render time compared to Haswell, a 21% decrease compared to Ivy Bridge, and a 27% decrease compared to Sandy Bridge.'' (per clock, fixed 4.5GHz)

WinRAR
- AnandTech: Skylake 0.1% slower than Haswell per clock
- Hardware.fr: Skylake 10.85% faster than Haswell per clock
- Lab501: Core i7 6700K 9.4% faster than Core i7 4790K with lower turbo clocks

 

[Whaeveva]

Bah. I bet my buddy tests AMD before any leaks even happen.

They might just do that where he is. I'll wait until Fl4nk3r posts something, or The Stilt, then make a decision.

My money is ready.

 

[krumme]

AT test was fine. They even contacted Intel about the discrete gpu results and there was no corrections.
They just tested using the same memory.

 

[itsmydamnation]

There's plenty of reviews that showed bigger gains for Skylake than what we saw in AnandTech's review. Instead of telling people to 'wake up' you should be looking at more than a single piece of data.

- Eurogamer:

Core i5 6600K vs Core i5 4690K (same 3.5-3.9GHz base/turbo)
- 17% faster @ The Witcher 3
- 11% faster @ GTA V
- 10% faster @ Battlefield 4

- Hardware.fr

- PCLab

Core i5 6600K vs Core i5 4690K at fixed 4.5GHz, Skylake-S is:
16.2% faster @ Battlefield 4 MP
10.9% faster @ Counter Strife Global Offensive
6% faster @ Crysis 3
22.7% faster @ Far Cry 4
13% faster @ GTA V
12.5% faster @ The Witcher 3
9.2% faster @ Watch Dogs
14.7% faster @ Project Cars
14.8% faster @ Starcraft 2
24.5% faster @ Total War Attila

Also AnandTech's results didn't match those of any other website:

When you limit your search to before end of year 2015 and remembering Anandtech did clock for clock having 15% base clock advantage against a 4770k can be deceiving.

guru3d doesn't seem that different
toms doesn't look that different ( they run 2133 ddr4 in many cases if you normalize for cpu clock 4770k is winning (ddr3 2133))
techspot had 2133 DDR4 and again looks simlar vs haswell DDR3 2400 (4790K almost always winning)
ARStech has DDR4 2666 vs DDR3 1866 and it shows skylake winning by ~10% ( but a tie in games)

can you provide links to the tests so we can see testing setups?
without seeing if memory access speeds are comparable we wont know if it really is a valid apples to apples comparison

this anandtech graph shows it perfectly between DDR3 1866 CAS 9 and DDR4 2133 CAS 14 :

The interesting thing from the graph is even with Comparable memory in many cases skylake is able to deliver lower latency.

 

[Atari2600]

There is some amount of sh!te in this thread regarding clocks.

You cannot compare clocks across processes. You cannot compare clocks across architectures. Any such comparisons are essentially pissing into the wind. Even if you end up approximately right, its no different from a broken clock being right twice a day.


We have 3 numbers of relevance. The first two being 3.2GHz base and 3.5 GHz turbo (both gleamed directly from the CPU string). The last being 4.0 GHz on air as noted a few pages back.

I personally don't expect anything beyond the 3.2/3.5 at launch. Indeed, the ceiling may not even be that high.


I distinctly recall getting badly stung by rumours of a 3.0 GHz Barcelona several months before launch crushing all Conroes before it.

 

[dark zero]

I have SandyB and IvyB chips that I've been meaning to test for some time, but haven't gotten around to it! I'll be buying an AM4 board and Bristol & Summit Ridge chips whenever they happen to become available as boxed CPUs to test, too.




Yep, Broadwell was just a big mess for Intel in general. Meagre IPC gain, clock regression, late to market, etc. Just terrible execution from Intel on Broadwell and the 14nm process, too.

Broadwell looks decent... If you compare to Cherry Trail. That is a Prime example of Idiot Programming. And then there is Qualcomm Snapdragon 810.

 

[Timmah!]

There's plenty of reviews that showed bigger gains for Skylake than what we saw in AnandTech's review. Instead of telling people to 'wake up' you should be looking at more than a single piece of data.

- Eurogamer:

Core i5 6600K vs Core i5 4690K (same 3.5-3.9GHz base/turbo)
- 17% faster @ The Witcher 3
- 11% faster @ GTA V
- 10% faster @ Battlefield 4

- Hardware.fr


- PCLab

:

Just proves my point that Broadwell is way closer to Skylake than Haswell.
It is just generally disliked by people cause it does not clock as high (as either Haswell or Skylake), even though especially in case of Haswell it more than makes up for this defficiency with its improved IPC. But clearly most people would take 4,6GHz chip with worse IPC over equally powerful 4,2GHz one with better IPC, just because cosmetically it looks better - therefore Haswell gets a pass and Broadwell does not.

 

[JordanMihailov]

I hope those aren't the final scores. Really want to see more competitiveness from AMD.

 

[mikk]

May i remind you the skl performed at average better with the ddr4 system than with the ddr3?

With faster memory speeds DDR4 should perform better. At the same clock, DDR3 with its lower latency should do better.

Just proves my point that Broadwell is way closer to Skylake than Haswell.

You didn't. Exactly the opposite. If Broadwell requires 128MB of edram to match the IPC gain of Skylake, then it clearly shows that the Broadwell core brought less IPC gains to us.

 

[KTE]

Performance has moved very little since Vth scaling has ended. IPC increases with successive generations have been offset by thermal/power problems limiting clockspeeds. Tackling this is impossible to very expensive. This has forced many players to leave this market (like IBM).

Then, IPC shoots up massively each generation where tricks are already known but yet to be implented. The 'low hanging fruit', so they say. Currently, that's close to dried out.

Clockspeeds have hit a dead end a decade back.

So has doubling caches.

So has doubling transistors.

So has doubling width.

So has the multi-core bubble.

Now, nearly done is the 'on-chip' period.

That's your last 30 years of performance gains.

Ultimately, thermals/power limit everything.

Any new CPU MFG with adequate funding can now come in and reach this 'already tackled' stage. This isn't surprising. All these obstacles have already been heavily tackled. But what about going forward into the unknown, what about the risk taking, the innovating?

That is the reason no one but Intel/AMD has kept this market for oh so long. Only they seem to pull this off repeatedly without catastrophes in this market.

Sent from HTC 10
(Opinions are own)

 

[LTC8K6]

I hope to goodness the OP is a lie. I have no plans to buy any new Intel technology, because they serve it out in teaspoons.I'm just tired of it.

Also all the newer stuff post Ivy-Bridge is very fragile and prone to failure.

Fragile and prone to failure? Bull...

How much new CPU technology has AMD spooned out for you up to today? Is Zen even "new technology"?

 

[Phynaz]

Also all the newer stuff post Ivy-Bridge is very fragile and prone to failure.

Citation needed.

 

[TheELF]

Citation needed.

You just did (provide his post as an citation to his starting argument)
Although I guess he is referring to the cpus being thinner and being stressed (lightly bending) by the heavier coolers.