AMD RyZen CPU Architecture for 2017

[Psycho]

"No, this is strictly CPU scheduling within the game." Robert Hallock, talking if a gpu driver can fix SMT issues on games.

I think it would be obvious to add a checkbox in the radeon drivers (since they are doing a lot of game detection/profiles and manipulation anyway) to disable SMT for certain game and cpu configs.

Did any of the reviewers mention if they tried testing with power saving off?

At least toms did: http://www.tomshardware.com/reviews/amd-ryzen-7-1800x-cpu,4951-6.html

 

[Anarchist4000]

Some suggest that big part of the performance issues (where they exist) are due threads being thrown (too easily) from CCX to CCX, and cache-snooping the other CCX's L3 while making memory access in case that's a miss too doesn't help. I wonder if they can evolve the design so that each CCX share same big L3 cache instead of dedicated L3 cache per CCX? It could solve at least part of the problem

our SMT scheduling topology for Fam17h systems is broken, because the ThreadId is included in the ApicId when SMT is enabled. So, without further decoding cpu_core_id is unique for each thread rather than the same for threads on the same core. This didn't affect systems with SMT disabled. Make cpu_core_id be what it is defined to be.
https://git.kernel.org/cgit/linux/k.../?id=08b259631b5a1d912af4832847b5642f377d9101

That kind of sounds like this. Wouldn't be surprising if Windows hadn't patched it yet.

Most benchmarks I've seen seem to indicate Intel being better with single threaded and AMD with multi-threaded with not a lot of games that can strain that many cores. Performance is ultimately still limited by a single thread in most tests.

 

[Silent_Buddha]

At least toms did: http://www.tomshardware.com/reviews/amd-ryzen-7-1800x-cpu,4951-6.html

Yeah, it's rather interesting that Battlefield 1 had a 10.1 FPS (6.7%) increase in rendering performance just by switching from a Balanced Windows power profile to the High Performance power profile for the 1800x. Minimum FPS went up a slightly larger 11 FPS (7.9%).

It basically went from being CPU limited (balanced power profile) to being GPU limited (high performance power profile).

I don't think I've seen anything that drastic before in a game just by changing Window's power profile. Although it's complicated by the fact that SMT is also disabled at the same time. So it might just be SMT and not the HP profile.

Regards,
SB

 

[Kyyla]

All in all Ryzen seems to be around 10% slower than 7700k on average in games. It's not bad at all.

 

[digitalwanderer]

From reading the reviews I've been noticing a trend that seems to indicate that AMD was really pushing hard to hit this launch date, and it may have been a tad early. Nothing major seems wrong with the CPUs, they really do look great...but it looks like they're still working out a few software bugs and they have been frantically all last week.

I expect it to be a great chip, but it probably won't hit its potential until the R5 launch. Just my opinion from reading reviews, still think it's gonna bring a new golden age of gaming it's just not gonna happen quite today.

 

[Malo]

From reading the reviews I've been noticing a trend that seems to indicate that AMD was really pushing hard to hit this launch date, and it may have been a tad early. Nothing major seems wrong with the CPUs, they really do look great...but it looks like they're still working out a few software bugs and they have been frantically all last week.

It's a CPU, it's not like they included a floppy disk in the box

 

[xEx]

Except from the gamer nexus review which seems to be hitting amd too hard even to the point of recommend buying an i3 before ryzen....

Enviado desde mi HTC One mediante Tapatalk

 

[Silent_Buddha]

Except from the gamer nexus review which seems to be hitting amd too hard even to the point of recommend buying an i3 before ryzen....

Enviado desde mi HTC One mediante Tapatalk

Should look at multiple reviews, and try not to tunnel vision on just one. Good or bad.

Regards,
SB

 

[xEx]

That was what I did/do. Just commenting on that specific one.

Enviado desde mi HTC One mediante Tapatalk

 

[Jay]

I know at gdc this week amd has a session on developing/optimising for ryzan, has that happened yet?

 

[xEx]

Soon. Tn

Enviado desde mi HTC One mediante Tapatalk

 

[Kaotik]

Also, worth a read https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/

 

[3dilettante]

Interesting that the voltage regulators are bypassed for the desktop platform.
Memory speed defines the data fabric clock, which influences performance outside of RAM bandwidth.

The voltage/frequency curve gets very steep past where Zen is currently being clocked to. The physical process and core implementation do matter, and currently there doesn't seem to be much room to push the silicon safely either for an overclock or another bin. Perhaps due to platform immaturity or design complexity, most of the power management and turbo functionality turns off if the chip is overclocked. Conceivably, if the whole chip cannot bump up to the desired clocks and voltages, overclocking can lead to a single-threaded regression--nevermind the likely blowup in power consumption.

 

[sebbbi]

All in all Ryzen seems to be around 10% slower than 7700k on average in games. It's not bad at all.

That's awesome, considering 7700K is 4.2 GHz, while Ryzen 1800X is 3.6 GHz. Intel has 17% clock advantage. Intel's 8-core chips aren't any better in games.

High clocked quad is currently the best gaming CPU. Similarly, the highest clocked Core 2 Duo was better than Core 2 Quad for games when it launched, but a few years later the tables had turned. Now that we have consumer 8-cores (1000$ i7 wasn't really a consumer product), game developers will start utilizing 8-cores on PC. Intel's forthcoming Coffee Lake also has mainstream (non HEDT) 6-core (12 thread) models. After that 4-core is no longer the consumer champion on either side. Future games (1-2 years) will certainly benefit from 6+ core CPU. Let's check i7 7700K vs Ryzen 1800K in 2019. I'd expect to see different results.

Ryzen R1800X seems to be a very good CPU for game programmers. It clearly beats Intel in both perf/W and perf/$ in compiling C/C++ code (VS/GCC). Source: http://www.hardware.fr/articles/956-12/compilation-visual-studio-mingw-w64-gcc.html. I

Ryzen offers around 35% higher performance than fastest Intel quad core and is only 7% slower than 2x more expensive Intel 8-core (6900K). 8-core chips also greatly improve data cooking times and shader recompile times in big projects. I am certain that many game developers are buying Ryzens. This is great for AMD, since no professional developer bought their previous FX chips. Practically no programmer had them, meaning that practically nobody analyzed code perf on them. Ryzen will change this. As a result we will see more games that scale properly to 8-cores and more games that are optimized for AMD CPU.

I am surprised that AMD didn't release a 6-core Ryzen 7 with higher clocks and same TDP. They could have released a 6-core with 3.8 GHz base clock / 4.2 GHz turbo with 95W. That's already 6% advantage in lower threaded apps (games). I am sure there would have been enough cores that reach 200 MHz higher clocks, but have a single core broken. If AMD had this product, and they had priced it at 7700K price, Ryzen would have received better press coverage regarding to gaming. Now many reviewers compared 500$ 1800X to the cheaper 7700K and it lost both performance and price in games. Slightly higher clocked and cheaper 6-core would have been perfect addition to the Ryzen 7 launch lineup. It would have been very close to 7700K in games, but have still significant lead in multithreaded apps.

 

[Gubbi]

It offers around 35% higher performance than fastest Intel quad core and is only 7% slower than 2x more expensive Intel 8-core (6900K)

The 6900K is also 140W TDP, AMD could fit a 16 core device clocked at R7 1700 speeds in that thermal envelope.

Outright single-thread performance might not beat Intel's best, but Ryzen is better across the board on perf/W, this bodes well for future server and laptop SKUs.

Cheers

 

[sebbbi]

The 6900K is also 140W TDP, AMD could fit a 16 core device clocked at R7 1700 speeds in that thermal envelope.

Also 16x Ryzen cores at ~2.7 GHz should fit to 95W TDP. That would be nice

 

[DavidGraham]

It clearly beats Intel in both perf/W

but Ryzen is better across the board on perf/W

From the reviews I've read, 1800X and 6900K are about even in power consumption.(Does a difference of 5 watts qualify as better perf/w?).
http://www.hardware.fr/articles/956-9/consommation-efficacite-energetique.html
http://www.techspot.com/review/1345-amd-ryzen-7-1800x-1700x/page6.html
https://www.pcper.com/reviews/Proce...Now-and-Zen/Power-Consumption-and-Conclusions

 

[Alexko]

Also 16x Ryzen cores at ~2.7 GHz should fit to 95W TDP. That would be nice

For reference (unknown clocks, though):

https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/

 

[xEx]

I think the biggest thing about ryzen is that it is AMD base for the future, yes it may not be perfect but damn AMD couldn't wish for a better base to have. It will be very interesting to see a ryzen+ with all the problems solved, better support from games and some improvements.

I think we only saw 8core because AMD was at the very edge of what they could do, new cpus, boards, systems and having to core cores and validated and all of that was too much more and AMD decided to focus in a single chip.

 

[sebbbi]

From the reviews I've read, 1800X and 6900K are about even in power consumption.(Does a difference of 5 watts qualify as better perf/w?).
http://www.hardware.fr/articles/956-9/consommation-efficacite-energetique.html
http://www.techspot.com/review/1345-amd-ryzen-7-1800x-1700x/page6.html
https://www.pcper.com/reviews/Proce...Now-and-Zen/Power-Consumption-and-Conclusions

Keep in mind that these are whole system power consumption figures (including memory, motherboard, etc). But it seems to be that AMD measures TDP differently than Intel. AMDs TDP seems to be closer to Intel's SDP (scenario design power), which roughly means common power usage. TDP for Intel means absolute peak power consumption. Intel CPUs need to run heavy AVX2 (FMA) code on all cores to reach TDP (AVX2 code is known to reduce clocks to maintain TDP). 256 bit AVX is the main reason why Intel doubled their cache bandwidth and included quad channel memory. Obviously all this heavy data processing and data movement consumes a lot of power, but only in applications that use AVX/AVX2 extensively. Thus comparing AMD TDP directly with Intel TDP is not valid in most cases.

Comparison when both CPUs run heavy AVX2 (FMA) workload would result in maximum TDP on both CPUs. Intel would definitely consume more energy. But AMD only has 128 bit wide vector units, resulting in peak theoretical performance is only 1/2 of Intel's (real life difference is obviously less).

I think the biggest thing about ryzen is that it is AMD base for the future, yes it may not be perfect but damn AMD couldn't wish for a better base to have. It will be very interesting to see a ryzen+ with all the problems solved, better support from games and some improvements.

I think we only saw 8core because AMD was at the very edge of what they could do, new cpus, boards, systems and having to core cores and validated and all of that was too much more and AMD decided to focus in a single chip.

AMD lead engineer already said that they have a list of low hanging fruit to improve Zen IPC in the future. It is first iteration of a brand new architecture. Also many sources tell that Global Foundries 14mm process is significantly inferior to Intel's new 14mm+. Equivalent process would increase performance and lower power consumption a lot. However Intel will introduce 10mm later this year, further increasing their advantage in process technology.