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

[bjt2]

HSW to BDW-E is 3.3% IPC jump according to AT. Do you think that difference really matters much?

And BDW-SKL/KBL is another 3.5-4%
So +10% clock is enough to compensate...

 

[swilli89]

AMD stock is likely rising because of a recent column in Barron's saying that the stock could double in a year.

http://www.barrons.com/articles/this-time-amds-revival-is-for-real-1486188068

The article is worthless. Almost every single "financial analysts" writing these articles has a bachelor's degree in business management from their local college and think reading forums and a wiki page makes them qualified to write these articles.

No serious investor bought into AMD because of this article.

Might have been the rumor getting recirculated that AMD is going to license it's GPU IP to Intel.

 

[swilli89]

If AMD has a great server chip*, and if yields are good, and if the market accepts AMD server chips, and if Intel drops the ball when they release their next server chips this summer, then AMD has huge potential gain. The server market is massive and AMD has been cut down to a lean company without much excess cost. If all of those are true, then AMD could easily be worth $25/share (or more). But, that is a lot of "ifs".

AMD always shoots up before a launch. Most of the time, if falls back down after a short period. Either the AMD launch wasn't as hyped, or Intel responded (often by putting in one more top processor and knocking all other processors down one price range). I think AMD isn't $25/share now due to this risk.


* I speak of server chips because Ryzen seems to be geared for that at launch.

Ryzen is 0% geared towards server at launch. Naples 32c part in 2Q Wil target the server/enterprise market.

 

[KTE]

HSW to BDW-E is 3.3% IPC jump according to AT. Do you think that difference really matters much?

If it really is ~3% then Ryzen should have no problem being Broadwell-E level.

We'll see.

Sent from HTC 10
(Opinions are own)

 

[dullard]

Ryzen is 0% geared towards server at launch. Naples 32c part in 2Q Wil target the server/enterprise market.

16 threads at launch is not a gaming CPU, nor is it a typical business CPU, nor is it a cheap home CPU. The 4 core Ryzen won't come until much later this summer. At best you could call it a high-end workstation CPU which is basically the same as a low end server.

 

[inf64]

16 threads at launch is not a gaming CPU, nor is it a typical business CPU, nor is it a cheap home CPU. The 4 core Ryzen won't come until much later this summer.

AMD said they are having top to bottom launch, meaning 4C/6C/8C variants launching at the same time in March. Also they said it won't be "paper" launch either.

 

[dullard]

AMD said they are having top to bottom launch, meaning 4C/6C/8C variants launching at the same time in March. Also they said it won't be "paper" launch either.

That isn't what the rumors have been saying. Nor were the AMD Blender benchmarks geared towards gamers.

 

[inf64]

That isn't what the rumors have been saying. Nor were the AMD Blender benchmarks geared towards gamers.

Rumors are one thing, official information is another. Also Blender was used to demonstrate that 16T Ryzen is competitive with intel's highest end 16T SKU. Gaming performance will be dictated by clock speed and IPC.

 

[swilli89]

That isn't what the rumors have been saying. Nor were the AMD Blender benchmarks geared towards gamers.

What have the rumors been saying? And yes 8c cpu is gaming/content creation for consumers.

 

[dullard]

What have the rumors been saying? And yes 8c cpu is gaming/content creation for consumers.

I've heard August, but that is just a rumor. We'll know in a month.

And you are right, 16 threads does so well in gaming (roll eyes):
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/9

 

[Magic Hate Ball]

I've heard August, but that is just a rumor. We'll know in a month.

And you are right, 16 threads does so well in gaming (roll eyes):
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/9

Cores can help in several newer AAA games http://www.techspot.com/review/1180-overwatch-benchmarks/page5.html and it's not always about averages but 0.1% frame times that become visible with some lower core/thread count CPUs in other games as well. And that's at the stock speeds, which tend to leave a large percent of overclocking headroom vs the lower core variants.

Also, running a single player benchmark is not the same as playing on a 64-player BF1 server with lower core counts.

 

[inf64]

I've heard August, but that is just a rumor. We'll know in a month.

And you are right, 16 threads does so well in gaming (roll eyes):
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/9

Try more CPU limited review like this one:
http://www.hardware.fr/articles/946-6/performances-jeux-3d.html

6900K @ stock offers ~89% of 6700K's performance in games. That is great showing for 16T CPU that has 13.5% lower max. ST Turbo clock.

 

[Atari2600]

6900K @ stock offers ~89% of 6700K's performance in games. That is great showing for 16T CPU that has 13.5% lower max. ST Turbo clock.

13.5% lower clock = 11% lower performance?

I fail to see the great showing to be honest!

 

[dullard]

Cores can help in several newer AAA games http://www.techspot.com/review/1180-overwatch-benchmarks/page5.html and it's not always about averages but 0.1% frame times that become visible with some lower core/thread count CPUs in other games as well. And that's at the stock speeds, which tend to leave a large percent of overclocking headroom vs the lower core variants.

Also, running a single player benchmark is not the same as playing on a 64-player BF1 server with lower core counts.

You'll need to spoon feed me with that link. The top performer is a 4-core part. But, that is a far newer part. So, looking at the next two performers as they were both the same generation and the 6-core is basically identical to the 4-core. Again, I just don't see that as proving that we need 16 threads to game.

Try more CPU limited review like this one:
http://www.hardware.fr/articles/946-6/performances-jeux-3d.html

6900K @ stock offers ~89% of 6700K's performance in games. That is great showing for 16T CPU that has 13.5% lower max. ST Turbo clock.

So, the processor with more cores was slower, and triple the price, so I'm supposed be convinced to buy the processor with more cores to game?

 

[DrMrLordX]

I have found a link to a PDF explaining the AVFS, behind a paywall, on IEEExplorer. Since i am a researcher, my organization has access to this article and i read it.

In short, for Carrizo (and almost surely for all subsequent products) there are 10 units spread across the chips each of which have 50 replica of critical path circuits (for a total of 500) and with a complex circuit collect statistics on its delay at various Vcore, to calculate Fmax at those Vcores. This Statistic is calculated on demand by microcode or SMU and allow to precisely know what Vcore apply to a certain frequency. Obviously in Zen this will be useful also to calculate the Fmax, given the current situation (temperature, chip age, and so on)...

Hmm. Only in the case of XFR, it looks like there is no limit to the clockspeeds that can be reached in this fashion. Carrizo/Bristol Ridge still seem to be limited (or biased strongly towards) p-states.

 

[bjt2]

Hmm. Only in the case of XFR, it looks like there is no limit to the clockspeeds that can be reached in this fashion. Carrizo/Bristol Ridge still seem to be limited (or biased strongly towards) p-states.

Exactly. Probabily there is a range of acceptable vcore and these units, on demand, and probabily some times per second, collect statistics on units delay, from which you can derive Fmax for the current state (temperature, age and quality of the chip)...

 

[bjt2]

Moreover, from the graphs that i have seen, the statistics collected are very much finegrained. i will not be surprised if the XFR can be configurable in the "strongness" of the OC, with server class CPUs with a default "strongness" very low and consumer class with higher tolerance...

 

[bjt2]

This is an extract: "The SMU summarizes the statistics of the data and creates a VFT table, essentially the part-specific optimal voltage for any frequency-temperature combination. P-state changes during normal operation reference the VFT table to determine the optimal voltage."

This for carrizo. Probabily for Zen the SMU decides also the current frequency based also on core occupancy...
Probabily it's another algorithm, but with substantially the same input data.

 

[LTC8K6]

Would we be able to get a worthwhile manual overclock above the XFR auto-overclock? If we can only get 100Mhz, for example, it wouldn't be worth the effort.

 

[Magic Hate Ball]

You'll need to spoon feed me with that link. The top performer is a 4-core part. But, that is a far newer part. So, looking at the next two performers as they were both the same generation and the 6-core is basically identical to the 4-core. Again, I just don't see that as proving that we need 16 threads to game.

So, the processor with more cores was slower, and triple the price, so I'm supposed be convinced to buy the processor with more cores to game?

4 core / 8 thread vs 4 core / 4 thread has a huge gap. This shows that the game is using the extra threads available. The 6 core is significantly lower in clock speed vs what is capable with an OC (large % OC overhead).

SMT logical cores aren't as effective as the equivalent number of physical cores. Hence, if this game is able to utilize 8 logical threads, then there is room for improvement having 8 physical cores. Probably best seen in maximum frame times, which equates to less stuttering, and smoother gameplay experience.

 

[bjt2]

Would we be able to get a worthwhile manual overclock above the XFR auto-overclock? If we can only get 100Mhz, for example, it wouldn't be worth the effort.

The problem with manual overclock is that usually you disable turbo and seek a fixed frequency that must be stable under all conditions. The only creative OC modes that can be better than XFR are FSB OC leaving P-state unchanged or P-state editing. But XFR is better because adjust frequency dynamically based also on CPU temperature... So after long idle with fresh CPU it will give higher clock...

 

[dullard]

4 core / 8 thread vs 4 core / 4 thread has a huge gap. This shows that the game is using the extra threads available.

SMT logical cores aren't as effective as the equivalent number of physical cores. Hence, if this game is able to utilize 8 logical threads, then there is room for improvement having 8 physical cores. Probably best seen in maximum frame times, which equates to less stuttering, and smoother gameplay experience.

Did you link to what you intended to link to? Here is what I'm seeing in your link:

• 6700K (4 Core / 8 Thread) beats all. Doesn't make me want more threads for gaming.
• 4960X (6 Core / 12 Thread) is essentially tied with the lower clocked 4770K (4 Core / 8 Thread). Doesn't make me want more threads for gaming.
• 5960X (8 Core / 16 Thread) is the slowest of all i7 processors. Doesn't make me want more threads for gaming.

What I think you are seeing is the cache effect, since threads didn't help.

• The 4770k (8 MB Cache) does beat the identically clocked 4690k (6 MB cache).

 

[inquiss]

The problem with manual overclock is that usually you disable turbo and seek a fixed frequency that must be stable under all conditions. The only creative OC modes that can be better than XFR are FSB OC leaving P-state unchanged or P-state editing. But XFR is better because adjust frequency dynamically based also on CPU temperature... So after long idle with fresh CPU it will give higher clock...

I agree with this bolded part - of the many people saying that they will purchase and manually overclock - what is the benefit if the turbo here is governed by only a thermal cap. Wouldn't leaving a lower base clock allow for more thermal headroom when the CPU power is needed as XFR kicks in. Is the delay too slow?

 

[Magic Hate Ball]

Did you link to what you intended to link to? Here is what I'm seeing in your link:

• 6700K (4 Core / 8 Thread) beats all. Doesn't make me want more threads for gaming.
• 4960X (6 Core / 12 Thread) is essentially tied with the lower clocked 4770K (4 Core / 8 Thread). Doesn't make me want more threads for gaming.
• 5960X (8 Core / 16 Thread) is the slowest of all i7 processors. Doesn't make me want more threads for gaming.

What I think you are seeing is the cache effect, since threads didn't help.

• The 4770k (8 MB Cache) does beat the identically clocked 4690k (6 MB cache).

There are some pretty significant clockspeed variations involved. And likely the GTX1080 becoming part of the bottleneck. But if you want proof that we're moving towards more heavily threaded games, look a few of the bigger releases of 2016.

Look at the graph, 4 core / 4 thread is already becoming a liability. I don't buy hardware just for todays games, I like to hold onto a chip for at least 3 years, and if the current trend is rapidly engaging more and more cores... then I'm going to buy that way.

This isn't counting my propensity to have obese applications running in tandem on a 2nd screen *cough*Chrome*cough* to satiate my media addiction.

 

[CentroX]

^^ why are AMD stocks rising so fast it is becouse of Ryzen if it is how does unreleased and untested CPU rises stocks so high?

lol neogaf think stock rises because of scorpio, pathetic.

http://www.neogaf.com/forum/showthread.php?t=1341408