MSI Drops First Hint of AMD Increasing AM4 CPU Core Counts

[Frick]

meh still sleeping, most review sites like toms, techspot, and techpowerup could not get it past the 4.3ghz base. I found guru3D gets an A for effort to get it to 4.4ghz (183 single thread) through liquid cooling.

So only like 5 CPUs are worth talking about?

Could someone explain this IPC thing to me?
If Intel hasn't upped that for years now, is it even possible? The core count is something different, there was no push for that. But IPC... why not? People wold buy such CPU like crazy.

Check out Sweclockers CPU reviews, they do tests with CPUs at the same clocks.

 

[Ruru]

Why is no one talking about how incredibly cringey the video is?!

IMO that was funny. Just like those AMD's commercial few years ago.

I don't mind if there's coming more cores to mainstream segment, it also puts pressure for Intel since they're now just milking with Skylake by fine-tuning it and adding cores. All the -lake CPUs have identical IPC, Coffee Lake (and the coming 8-core 1151-one) just added cores.

Also some people could move on from older HEDT CPUs which could mean cheaper prices of used CPUs, and I'd love to have a Broadwell-E running the cat videos on my second monitor.

 

[Deleted member 178884]

185 single thread is easily achieved with the ryzen 7 2700x - he's a troll or die hard fanboy. Even my 6600k at 4.5 can get 193 single thread so there's no reason Zen can too. My 7740x however will break 200 with ease once I can buy my x299 i9 xe motherboard, I'd wake up and buy that ryzen if 185 single thread is what you need.

Bu.. but.. but mah single thread performance! Mah gaymes fps!

It does, and in turn based game's it's essential, but I guess you kids thought I'd say csgo or something.
Hell even my 6600k scoring 193cb single thread can't hold up in turn based games with the rare crash or lag when processing which is annoying. It's also the reason I found a cheap 7740x and bought it - It's basically a binned by 100mhz 7700k and will yield a massive performance boost. Ryzen is excellent but x299 is officially my last intel extreme platform since the x58 chipset.

 

[Johan45]

I for one will certainly vouch for Johan on this being the real deal.

Thanks Doc

 

[cuneytcam]

AMD gave more IPC increase between 1st and 2nd gen Ryzen than Intel did between its past 3 generations; despite Zen and Zen+ being the same chip physically. I'm hopeful.

As an anti-intel end-user due to their business practices, Even though the charts back your statement up those IPC gains came from mostly bus improvements such as latency, memory efficiency, iFabric etc. Iirc only branching guesswork improved on execution side compared to sources of Intel cpu ipc gains. So AMD needs to keep working harder than we might be thinking after success of Ryzens. They do not have any other move that would yield this much performance this easy. (Faster memories being available on its own, relatively easy task to get done like better memory compatibility and improved imc, basically more of polishing improvements). This makes me think Ryzen+ was the initial Ryzen they wanted to release.

 

[Valantar]

As an anti-intel end-user due to their business practices, Even though the charts back your statement up those IPC gains came from mostly bus improvements such as latency, memory efficiency, iFabric etc. Iirc only branching guesswork improved on execution side compared to sources of Intel cpu ipc gains. So AMD needs to keep working harder than we might be thinking after success of Ryzens. They do not have any other move that would yield this much performance this easy. (Faster memories being available on its own, relatively easy task to get done like better memory compatibility and improved imc, basically more of polishing improvements). This makes me think Ryzen+ was the initial Ryzen they wanted to release.

You're a bit off here. The IPC improvements between Zen and Zen+ mainly came from faster caches and interconnects and lower internal latencies latencies. This yielded a ~3% gain, from tweaks so small they could make them work while refreshing Zen for the updated, process node. Faster RAM support doesn't affect IPC, and doesn't play into this. Anything internal to the CPU, whether it's cache, latencies, pipeline changes, improved branch prediction or other optimizations, affects IPC in its "purest" form.

AMD has also unequivocally stated that they know where the road forward lies in terms of improving performance, and that substantial per-generation IPC increases can be expected. Note that in this sense, Zen+ is not a generation, but a refresh. Considering that the design process of any complex IC is long enough that its designers will inevitably know where and how it falls short, I have no doubt at all that they've long since found areas for improvement. As with any product that needs to launch, they launched Zen (1) when it was ready, rather than delaying it to make it "perfect" (which would have been impossible anyhow). Zen wasn't rushed, but the AM4 platform was (hence the memory issues, which microcode updates have largely fixed).

 

[cuneytcam]

Well yeah, i think correct wording should rather be said single core performance (or per core) instead of IPC. If you increase clockspeeds IPC still remains same but single core performance increases. But memory It does affect single core performance in an inverted way. ~3% latency improvements, ~%2 faster memory support and matured production/clock speed bumps bring us to today's ryzen 1st and 2nd gen difference. However my message was rather about AMD needs to work hard to increase single core performance. They could have halved the IPC but tripled clock speeds with same power budget we should not care about how exactly they are going to pull it off and to get same increae from 1stgen to 2nd gen they need to do "more" because they used latency cards. Further improvements has to be on core-complexes and not limited some fine tunings since they done that and they got nice chunk of performance with polishing because product was ready to perform that way but i believe released early to get some cash going on. If im not horribly wrong board makers have nothing to do with microcode updates. That block of firmware of a mainboard i guess have a lot more to do with cpu producer rather than mainboard vendor itself.

 

[Adam Krazispeed]

I don't know how well a ten or even twelve core cpu would run on a four phase motherboard.

how do you even know the mbs will only be 4 phase???? seriously!!!!!! and it may not even b e until 7nm am4 Ryzen cpus before we get anything more than an 8c/16t cpu on AM4 MBs and when they are a 10c 12c and even a 16c 32 thread am4 behemoth would have to be on the 7nm node and they (GEn 3 Ryzen) (i believe) will most likely wont require a) any more than125 watt tdp??? if even that, 1800x R7 was only 95 watt, remember that, and that was on a 14nm finfet low power process

 

[Valantar]

how do you even know the mbs will only be 4 phase???? seriously!!!!!! and it may not even b e until 7nm am4 Ryzen cpus before we get anything more than an 8c/16t cpu on AM4 MBs and when they are a 10c 12c and even a 16c 32 thread am4 behemoth would have to be on the 7nm node and they (GEn 3 Ryzen) (i believe) will most likely wont require a) any more than125 watt tdp??? if even that, 1800x R7 was only 95 watt, remember that, and that was on a 14nm finfet low power process

AMD has had 65W 8-core CPUs since the first generation of Zen. There's not really any reason to think they wouldn't be able to get a 12-core (or even 16-core) chip running at or below 95W on 7nm. Sure, they might need to limit stock clocks somewhat, but that really isn't a problem. 12-16-core CPUs aren't for gaming anyhow.

And running 95W through 4 power phases isn't really an issue, given that they are made up of anything but garbage components. Let's say Vcore is 1.2V. That's 79.1A. Even cheapo bargain-basement power phases can handle >20A/phase.

 

[brian111]

how do you even know the mbs will only be 4 phase???? seriously!!!!!! and it may not even b e until 7nm am4 Ryzen cpus before we get anything more than an 8c/16t cpu on AM4 MBs and when they are a 10c 12c and even a 16c 32 thread am4 behemoth would have to be on the 7nm node and they (GEn 3 Ryzen) (i believe) will most likely wont require a) any more than125 watt tdp??? if even that, 1800x R7 was only 95 watt, remember that, and that was on a 14nm finfet low power process

The whole speculation about higher count Ryzen cpus comes from the video this article references. It's about MSI B450 motherboards which are mostly four phase motherboards. That's why I brought it up in the first place. Nobody suggested that there would "only be 4 phase" mobos.

It's almost a certainty that higher than eight core Ryzens won't arrive unitl 7nm (when six or eight CCXs are in use)- that's no surprise.

Ten and Twelve core cpus *MAY* run fine on four phase motherboards but it's not exactly being outlandish to question it. Hell some people don't recommend running eight cores (and certainly not overclocking them) on a four phase. Asrock even has some three phase motherboards.

 

[Valantar]

The whole speculation about higher count Ryzen cpus comes from the video this article references. It's about MSI B450 motherboards which are mostly four phase motherboards. That's why I brought it up in the first place. Nobody suggested that there would "only be 4 phase" mobos.

It's almost a certainty that higher than eight core Ryzens won't arrive unitl 7nm (when six or eight CCXs are in use)- that's no surprise.

Ten and Twelve core cpus *MAY* run fine on four phase motherboards but it's not exactly being outlandish to question it. Hell some people don't recommend running eight cores (and certainly not overclocking them) on a four phase. Asrock even has some three phase motherboards.

Overclocking is the key here. While newer turbo schemes do at times increase power draw well beyond TDP in short bursts (or if MCE is enabled on Intel platforms, which is essentially an automatic OC), in most cases a stock-clocked CPU stays within those limits. And, as I said, asking a single power phase to provide ~20A is really not asking a lot. Doubling that number by overclocking is another story entirely.

 

[John Naylor]

MSI & other vendors usually make cringe-worthy videos on their marketing part, though most cheesy ones is MSI... no offense.

My son must have watched this video 30-40 times ... my guess he goes back and watches it still watching it every now and then 3 years later ... just to make sure he didn't miss anything

I do use MSIs install component videos to show users how to install components

As for "wondering" ... I am wondering why core counts > 4 is something 98% of PC users need to think about. Might as well think about 8 wheels on a car. Accomplishes something on an extreme terrain vehicle or high weight tractor trailer combo but won't help me get to a job site any faster.

 

[Caring1]

As for "wondering" ... I am wondering why core counts > 4 is something 98% of PC users need to think about. Might as well think about 8 wheels on a car. Accomplishes something on an extreme terrain vehicle or high weight tractor trailer combo but won't help me get to a job site any faster.

There's ya problem, think of them as cylinders in an engine, not wheels.
The more you have, the more power at your disposal, you might not need it all the time, but it's there when you put the foot down.

 

[Valantar]

There's ya problem, think of them as cylinders in an engine, not wheels.
The more you have, the more power at your disposal, you might not need it all the time, but it's there when you put the foot down.

The whole car parts metaphor is rather ill-fitting, as a car can only go one place at a time. A computer does not do one thing at a time, so the metaphor would rather be having multiple cars. Having more cars than there are people with licenses in your household is wasteful and useless, and even having a 1:1 ratio is usually unnecessary and leaves one or more cars idle for most of the time - but makes crunch periods better and smoother.

As for 4 cores being "enough": while I see where @John Naylor is coming from, that's a very narrow view of computing. I mean, sure, for web browsing, even two fast cores with SMT can do the job without any real slowdowns. But throw something slightly demanding into the mix, and that changes dramatically. One extremely common thing to use a computer for is photo editing. While Adobe Lightroom could probably be better optimized and is a rather demanding application, it's a prime example of something that can use ally he resources you throw at it. It runs far better on my R5 1600X desktop than my i7-7600u laptop (both have 16GB RAM and NVMe SSDs). Photoshop can use these resources too. As can.... GiMP or Sketchup or various video editors or most professionally developed performance-limited applications today.

As for gaming, Intel's 8700k has shown very clearly how more threads can help even in 2-3-thread games, as compared to the 7700k there's less risk of background processes causing stuttering.

While I'd currently say that you don't need more than 6c12t for the vast majority of uses, that's also denying the fact that development today is focusing more and more on branching out into more threads as that's where there is more performance to be had. In the coming years, the need/use for more threads will only increase, and SMT is more of a band-aid than a fix. Plus, having the platform allow for more choice is purely a good thing. If you're using your PC mainly for video editing, a 16c CPU makes far more sense than 4c8t, even if the cores are slower. We're slowly but surely moving away from the era of a single SKU being "the fastest" CPU for all uses, as the need to balance fast cores, many cores, and power/cooling becomes more precarious. I don't see this as a problem, really.