AMD Ryzen 5 Processors Start At $169 and Launch on April 11th

[Deleted member 93354]

Oh the 4 core cpus are going to be a "blast" then. /facedesk

I was considering getting one to play around with because I was hoping it would be just one 4 core piece. But this? Nah. Maybe I'll reconsider and try a full 8 core cpu in the future. But I'll stick to my i7's for now. :/

You aren't going to get the fastest here. But to be honest at the highest res', few games are CPU limited. So if you are building a high end rig, a Ryzen 7 still might make sense if you do a lot of work station type task. But even a R5 will give you a great bang:buck and likely won't CPU limit you at the highest resolutions.

I know a lot of benchmarkers like Kyle and group like to max out the fastest processor they can get their hands on. Truth is there will be a relatively small percentage of difference between CPUs, as most of the time the CPUs are not being utilized 100% of the time.

At high resolutions, most of the time the processor isn't getting maxed out because the GPU is doing the lion share of the work. The CPU is only really responsible for "bundling up the scene, executing the game rules, and sending it off to the GPU" And because that is a small percentage of the total time at high resolutions, differences between a 4GHz Ryzen (Non overclocked) and 4.8 OC Kabylake aren't all that bad considering most games are designed for 4 cores, and very very few ever use more than 8 logical processors

The percentage difference between that and the next lower one down at the highest resolutions really isn't that big a deal when you consider the huge price break you get in the process and the upgrade of workstation performance. (If that is your thing)

I'm not saying Ryzen is the holy grail of processors, but it's a dang good alternative.

 

[SvenBent]

Call me stupid but i dont necessarily see 2+2 having the same issues as 4+4.. wouldn't the complexity going down in each core stack help somehow? Being the same silicon maybe they are not really killing cores but using them differently in such way that while effectively yes core count is down IPC can go up?

The problem is the access times between the two ccx's
So having a 2+2 is stil loging to have that access penalty because you still have the "slow" connection.
4 cores would have no "slow" connection and there by "faster"

consider having 4 huse on the same side of a 1persone vide brige. all house share a direct 5 lane highway to each other.
now the slow bridge doesn matter.
but if you instead have 2 house on one side of the slow brige. and 2 house on the other. trafic from one house often have to cross the slow bridge.

-- edit --
i think i understand you better now

you mean going form 4+4 to 2+2 has lesse of an issue ( but still there) because less CPU data has to go back and forth the CCX's.
I think you are right but. only the future will tell how much the issues is reduced.

 

[tunatime]

Wonder if it will be unlockable like the good old days?

 

[FrgMstr]

Kyle,
did they clarify whether at some point the 4-Core would have its own silicon and so eventually not rely upon 2+2?
There were hints that initially it would be same silicon as the 8-Core, and then later on would be its own silicon/die - but this was just some rumours.
I appreciate this would not be anytime soon even if AMD did do this, fingers crossed they will clarify but then I am not sure they will as they may think it will hurt 4-core sales with people waiting for a true 4-core dedicated CPU.
Thanks

"We do not comment on unreleased products." I am fairly sure that would be the answer.

 

[FrgMstr]

Yeah tend to agree even if they do get pressure from the press at large to clarify the development of the 4-core going forward.
I guess it all depends how it pans out in review benchmarks with games and whether it is fine or even further exacerbated with inter-ccx/data fabric use.
Thanks

I would suggest that if that was going to happen, it is already baked unless we are looking at Ryzen 8.

 

[gigaxtreme1]

Now if we can onlock some goodness on those 1600s!

 

[/dev/null]

Can't windows/linux treat each ccx like a numa node in a multi-socket system? Hasn't this problem been worked around for quite some time ( as well as it can be, at least) ?

 

[psyclist]

Im sure the engineers and ol Jimmy K knew what they were building. Lets just wait for some reviews befor writing it off. 16mb of cache may help with the latency issues if they are indeed detrimental

 

[NKD]

Well, this seems problematic, considering the cross-CCX latency issues.

Hopefully some of those issues can be solved through scheduler and other software improvements.

Or maybe AMD can increase cross CCX communication performance in a second revision of the chip.

This is fabric bandwidth issue. Higher speed memory greatly improves this. There are a few short reviews out there that show this. So only thing AMD needs to work on is supporting higher and higher memory speed. Which should be sorted out through bios updates. AMD has a forum post detailing this. By May they should have much better memory support it seems.

Since this only matters in games and it seems to reflect significant improvements as memory speeds go up which helps bandwidth between the CCXs

 

[Cali3350]

Im sure the engineers and ol Jimmy K knew what they were building. Lets just wait for some reviews befor writing it off. 16mb of cache may help with the latency issues if they are indeed detrimental

They have already done a phenomenal job! For non-gaming workloads they offer the better product most of the time.

I think its very realistic we see ~15% IPC improvements with Ryzen series 8. Its very easy to get a lot of low hanging fruit on the first major redesign after a brand new uarch.

 

[tunatime]

They have already done a phenomenal job! For non-gaming workloads they offer the better product most of the time.

I think its very realistic we see ~15% IPC improvements with Ryzen series 8. Its very easy to get a lot of low hanging fruit on the first major redesign after a brand new uarch.

They need clock speed if they could hit colse to 5ghz it would be a winner

 

[Xebec_]

I don't know if it is me, I just lost interests in PC stuff. Nothing like the glory days of 1995 to 2001!

It's just you .. 1979 through 1987 were much more exciting as engineers added custom chips that seriously enhanced performance beyond what a CPU only system could do ;-)

 

[SighTurtle]

I don't get how higher speed RAM would affect communication inside the CPU. Isn't the Infinity Fabric just a fancy word for the lines that connect the CPU to the rest of the board?

 

[NKD]

I don't get how higher speed RAM would affect communication inside the CPU. Isn't the Infinity Fabric just a fancy word for the lines that connect the CPU to the rest of the board?

No LOL. Higher speed memory increases fabric bandwidth, hence improving the communication between CCX. AMDs own Ryzen 5 launch video that kyle posted compares 2133mhz memory with 2966mhz I believe and results in 16 extra fps in F1 at 1080p. It's been written in several articles as well. That on Ryzen fabric bandwidth plays a major role on this chip especially in games, probably helps with constant swap that occurs during games.

 

[BloodyIron]

Dear god AMD is fucking tearing up this industry like a tissue at a snot party!

I am all kinds of xite.

 

[Elios]

Can't windows/linux treat each ccx like a numa node in a multi-socket system? Hasn't this problem been worked around for quite some time ( as well as it can be, at least) ?

that would been even slower as it would have to go out to RAM

 

[CrisanT]

"According to Mercury Research, 95% of all CPUs sold last year were less than $225."

This.

 

[dragonstongue]

They need clock speed if they could hit colse to 5ghz it would be a winner

Sure, and power usage would go through the roof and therefore temperatures and likely life span, not everyone wants or needs a stupid high clock speed, efficiency for the clock speed given and all around IPC(as well as more the modern chips can offload pre-complie etc etc) for what is delivered matters a great deal more than pushing crazy clock (just as example wtih the 8 core 16 thread chips, it takes a great deal of extra power to push that many cores an extra Ghz+ speed STABLE, everything has its limit, even more so when talking extremely small electronics)

Just saying, these things are very efficent for what they are delivering in work performance for the amount of power they are using(not that many years ago you were talking single core 65+w worth of power, similar to less actual speed, and pushing as much if not more heat, modern chips technically using a fraction of the same power and less actual heat given, GPU +/- though they are much faster as well) an extra few hundred Mhz does not seem like much, but it very much is when talking how much extra power needed and heat generated to drive them so, it is not a linear curve of X extra Mhz needs Y extra power, once past a certain curve it takes much much more power to "drive" it .

AMD, Intel, Nvidia etc do not make physics, but they do have to abide by their fundamental laws well maybe not as much Nv as they cut things away and do some fancy tricks to "appear" faster.

 

[Pandur]

So you'll still get a 350$+ i7 while complaining about CCX layout on a ~180$ chip. Way to make a terrible comparison there. Maybe had you said i3/low end 5?

You missed the point. I have two previous generation i7 computers that I use today, so it's not about spending money on a new one. I was playing with the idea of getting a RyZen to play with. And the 8-core design makes good enough sense. And I don't really mind the 6-core either. But the 4-core still being a 2x4-core CCX with half of it disabled... Meh. I'll pass. Because what I was hoping for was a single 4-core module which could've been spot on for a gaming setup. So it's not what I was hoping for, meaning I don't have to spend money on one either. That's all.

 

[dragonstongue]

Pandur, maybe the native 4 core 8 thread is a mix of CCX and non CCX cores, only once released will we truly know what they deliver

 

[Formosa]

im ok with 3+3 CCX but 2+2?

seriously?

Why not just a NATIVE 4 core chip with NO CCX issues?

Its like the IGFX in the intel i series crap indirectly; ie dead/useless die space taking space for TDP / forcing epple to buy a suboptimezed CPU from the bin bottom.

IMHO the real fight might be zen2 vs cannon dale/whatever intel can cook up.

 

[anubis44]

Well, this seems problematic, considering the cross-CCX latency issues.

Hopefully some of those issues can be solved through scheduler and other software improvements.

Or maybe AMD can increase cross CCX communication performance in a second revision of the chip.

I keep hearing conflicting information about the factors which affect the cross CCX speed. Apparently, faster memory improves the cross CCX bandwidth, but I thought this happened entirely on the die/module, and that system ram didn't affect this. Does anyone have any sources to expand on this question?

 

[NKD]

im ok with 3+3 CCX but 2+2?

seriously?

Why not just a NATIVE 4 core chip with NO CCX issues?

Its like the IGFX in the intel i series crap indirectly; ie dead/useless die space taking space for TDP / forcing epple to buy a suboptimezed CPU from the bin bottom.

IMHO the real fight might be zen2 vs cannon dale/whatever intel can cook up.

Simply because if allows them to just make one chip and cut it down. Probably saves them a ton of time and resources. Also they are keeping it fair. They are likely avoiding product discrepancies and making all of them equal. What if they had 4+2 or 3+1 and 4+0? Imagine the shit storm they get if one processor hits 5% higher performer and rest are below. That kinda shit will get them in trouble and likely bring legal trouble. So they probably made it fair across the board. Everyone gets the same product. Easier to make 8 core if yields are good and just cuit it down for now.

Zen APU will probably be the true 4 core chip you are looking for with 11 Vega CUs on board.

 

[anubis44]

This is fabric bandwidth issue. Higher speed memory greatly improves this. There are a few short reviews out there that show this. So only thing AMD needs to work on is supporting higher and higher memory speed. Which should be sorted out through bios updates. AMD has a forum post detailing this. By May they should have much better memory support it seems.

Since this only matters in games and it seems to reflect significant improvements as memory speeds go up which helps bandwidth between the CCXs

Other than the AMD forum, do you have any links to a decent explanation about the role system memory speed plays in the cross CCX bandwidth?

 

[NKD]

Other than the AMD forum, do you have any links to a decent explanation about the role system memory speed plays in the cross CCX bandwidth?

I can try researching it once I get home. I have read it in forums in that and post. I believe it has more to do with memory speed. I think techpowero up had a discussion on their article where faster memory increases the bandwidth and it helps with cross communication between CCXs. It doesn't show benefit under normal workload but it does show it in games.

 

[Zarathustra[H]]

Other than the AMD forum, do you have any links to a decent explanation about the role system memory speed plays in the cross CCX bandwidth?

I saw AMD themselves tout that using fast RAM (DDR4-2933+) would help, which is funny considering they don't support anything above DDR4-2666 for Ryzen.

I think they might need to go back and revisit that memory controller.

 

[xX_Jack_Carver_Xx]

Memory bus looks like Intel junk. 100Mhz bus? My 8350 is cranking at 250Mhz with DDR2. I really want to see some testing with higher cpu-memory bus speeds.

 

[Tsumi]

Memory bus looks like Intel junk. 100Mhz bus? My 8350 is cranking at 250Mhz with DDR2. I really want to see some testing with higher cpu-memory bus speeds.

There is no such thing as 100 mhz memory bus on modern processors. It is only a reference clock from which all other clocks are derived by multipliers. You need to get up to date.

 

[NKD]

I saw AMD themselves tout that using fast RAM (DDR4-2933+) would help, which is funny considering they don't support anything above DDR4-2666 for Ryzen.

I think they might need to go back and revisit that memory controller.

There is nothing wrong with the memory controller. it has been well known since initial reviews were launched and the news following they that they had set parameters for memory and there will be constant improvements on memory side of things. So expect memory stability at higher speeds to increases. Some boards already do fine with 3200mhz.

 

[Zarathustra[H]]

There is nothing wrong with the memory controller. it has been well known since initial reviews were launched and the news following they that they had set parameters for memory and there will be constant improvements on memory side of things. So expect memory stability at higher speeds to increases. Some boards already do fine with 3200mhz.

Yep, but "official" supported RAM speeds for Ryzen still max out at DDR4-2666

 

[/dev/null]

that would been even slower as it would have to go out to RAM

I'm not sure what you mean.

I'm talking about scheduling -- figure out which CCX node a process is on and it's memory locality & treat it like numa.

 

[mesyn191]

I saw AMD themselves tout that using fast RAM (DDR4-2933+) would help, which is funny considering they don't support anything above DDR4-2666 for Ryzen.

Yep, but "official" supported RAM speeds for Ryzen still max out at DDR4-2666

Intel officially only supports up to DDR4 2400 on the i7-7700K. Anything past that is considered overclocking by Intel.

edit: So if "only" DDR4 2666 is "bad" how "terrible" is "only" DDR4 2400 support supposed to be?

OK so that is pretty excessive scare quote use but still complaining about official peak DDR4 RAM speed support is kinda silly at this point.

What AMD does have a actual problem with right now is getting their memory controller to support much faster DDR4 RAM. Its quite possible to get up to DDR4 4000+ working on a Kabylake CPU. That flat out isn't possible right now with Ryzen. Even getting to DDR4 3200 is a bit of a struggle. Hopefully they'll fix that issue fast since Ryzen does actually seem to really benefit from fast RAM, which is kind've odd but hey whatever works.

 

[drescherjm]

Ryzen still max out at DDR4-2666

I think this is the fastest ram being produced. I mean when I looked at datasheets from Samsung yesterday the fastest ram they have in production is DDR4-2666 CL19. Anything faster than that is overclocked ram.

 

[Zarathustra[H]]

Intel officially only supports up to DDR4 2400 on the i7-7700K. Anything past that is considered overclocking by Intel.

Good point. I did not realize that.

That being said, they design seems less reliant on it in order to perform well (due to the lack of a CCX issue) and Intel isn't recommending users try to improve performance by using faster RAM than they officially support.

 

[mesyn191]

That being said, they design seems less reliant on it in order to perform well (due to the lack of a CCX issue) and Intel isn't recommending users try to improve performance by using faster RAM than they officially support.

True but that isn't a memory controller issue and is also something probably not easily changed for Ryzen. Maybe they'll have improved CCX bandwidth enough in Ryzen 8 but who knows at this point.

Until then with Ryzen 7 getting faster RAM appears to be a surprisingly viable solution to this issue for an enthusiast since DDR4 3200 isn't expensive. The problem is right now getting it to work at those speeds. I think only a few RAM sticks out there are known to work at those speeds and only on a handful of motherboards for Ryzen 7. If its just a BIOS issue it'll get fixed in a month or 2 for almost everyone. We'll have to wait and see at this point.

 

[DocSavage]

At stock speeds. Overclocked, it'll be about 80% of the performance at 60% of the cost. Still not a bad deal.

My Microcenter page has the i5-7600K at $199 and the i7-7700K at $299, so it'd be more like 80% the performance of an i5 for 85% of the cost. The 6 core at $220 does better at 73% the cost of the i7.

 

[mesyn191]

My Microcenter page

"available for in store pick up only"

You might as well be going by cherry picked deal prices, of which AMD's Ryzen will have some too eventually (edit) from Microcenter, if you're going to do that.

 

[DocSavage]

"available for in store pick up only"

You might as well be going by cherry picked deal prices, of which AMD's Ryzen will have some too eventually (edit) from Microcenter, if you're going to do that.

Fair enough. Is there a good supply of Ryzen motherboards available now?

Edit: Microcenter shows plenty of Ryzen motherboards from $90 to $270.

 

[mesyn191]

.....

Nope. Even the 350 chipset boards have tighter availability than I'd like and I wouldn't buy one of those unless you're trying to do a budget build on tight cash flow.

Supposedly mobo availability will pick up by around the end of the month but that is a rumor mill source.

If it were me and I HAD to buy now I'd get the X370 Asrock Taichi mobo which seems to be the best one ATM. Its still very early though for Ryzen and every BIOS update makes a difference right now. I'm personally trying to hold off until May or June before buying. I'm waiting for not only the dust to settle on Ryzen AM4 mobo's but also for Vega and better monitors so I can upgrade everything at once.

edit:

Edit: Microcenter shows plenty of Ryzen motherboards from $90 to $270.

That isn't what I get when I search their store, only 1 in stock.
http://imgur.com/a/kJX0I

 

[NKD]

Yep, but "official" supported RAM speeds for Ryzen still max out at DDR4-2666

Intel is the same way though. Official support is just a number they are willing to support, rest is up to user. Its just a way to deny shit. To have an official response to any claims lol.