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

[nismotigerwvu]

A single title or line does not a review make -- I looked through those benchmarks for Phenom II and they are getting crushed by the new i7's and Anand is happy that the Phenom II can compete with the 2 year old (at the time) Core 2 Quads, though even the 9550 is still faster. With Bulldozer its also still slower and using 2x the power. They haven't released anything useful for mid-high end in 10 years.

Did you really read the article? Here's a quote 4 paragraphs down from where I quoted, "Here's how it breaks down. The Phenom II X4 940 is usually the same speed or faster than Intel's Core 2 Quad Q9400, and priced similarly at $275. There are some areas where the Q9400 will be faster than the Phenom II X4 940, so if you happen to use an application that runs better on Intel hardware then you've got your choice made out for you. But for the most part, if you're buying a quad-core processor at around $275 today, Phenom II will tempt you". Is it really noteworthy that the $999 i7 965, $562 940 and $285 920 and their $200+ motherboards were faster than the $275 and it's $75 mobos? The C2Q 9550 was a $~300 chip at the time as well and LGA 775 boards were still more expensive than AM2(+)/3 boards. Again, no stunner than the more expensive setup was faster. It isn't exactly like people are spouting that the Power 8 chips make the i7 line noncompetitive. It all comes back to that quote I'll leave here again, "A very smart man once told me that absolute performance doesn’t matter, it’s performance at a given price point that makes a product successful". Dollar for dollar, the Phenom II was more compelling at that time. Even more so when you consider the enormous performance boost a fairly dated AM2 based system could receive from dropping a 940 down in. No new mobo, no new RAM just a multigenerational leap in performance for that $275, which again was the cost of just a decent motherboard for one of those i7's.

 

[BeepBeep2]

Even more so when you consider the enormous performance boost a fairly dated AM2 based system could receive from dropping a 940 down in. No new mobo, no new RAM just a multigenerational leap in performance for that $275, which again was the cost of just a decent motherboard for one of those i7's.

Well, if it had only worked that way. EVGA/Foxconn/XFX 590 boards didn't support it (though I have a modded bios that works), really only the ASUS M2N32 boards actually allowed this. Then, when the AM3 CPUs came out, those didn't work in AM2 at all, or if they did they were locked at the lowest p-state

I agree that price/performance is what matters most - for those arguing that Phenom II wasn't competitive, it was close enough even vs. Nehalem for a lot of people, at cheaper price. The PII X6 CPUs (when they arrived) also kept up with Nehalem in some multi-threaded tasks.

 

[rtsurfer]

Sure, if you really want to tamper with the BCLK...
Personally I appreciate reliability more, than the few additional MHz that in most cases gain you nothing but bigger figures on the screen.

This just tells me that there won't be a native 4000Mhz freq divider. Sad...

Hopefully atleast mid 3000s.

 

[FlanK3r]

DRAM OC seems up to 3000 MHz I think on few boards. I saw dividers up to 2666 MHz in most of case, but one with higher. So I think, around the 3000 MHz in practice could be limit for Ryzen. Will see soon.

 

[Tup3x]

DRAM OC seems up to 3000 MHz I think on few boards. I saw dividers up to 2666 MHz in most of case, but one with higher. So I think, around the 3000 MHz in practice could be limit for Ryzen. Will see soon.

MSI listed DDR4-2667+(OC) in their AM4 mobo specs (what ever that means) but one could interpret that it means higher frequencies than that are possible. But how... who knows.

 

[Doom2pro]

MSI listed DDR4-2667+(OC) in their AM4 mobo specs (what ever that means) but one could interpret that it means higher frequencies than that are possible. But how... who knows.

The memory controller must be up to snuff.

 

[DrMrLordX]

There should be no need to touch the BCLK in the first place.

Why not? You can't just expect multipliers to do all the work for you, all the time. bclk OC gives the overclocker lots of flexibility, and it lets them go past the listed memory ratios on the motherboard.

s754, s939, AM2, AM2+, AM3, and AM3+ all had flexible and stable bclk/htt OC. It was beautiful. You could overclock anything from a lowly Sempr0n to a big-boy FX on pretty much any motherboard that could handle it. You just had to feel out the mobo's bclk limits.

 

[DeathReborn]

The memory controller must be up to snuff.

Is the memory controller in Zen an off the shelf or custom design?

 

[The Stilt]

Why not? You can't just expect multipliers to do all the work for you, all the time. bclk OC gives the overclocker lots of flexibility, and it lets them go past the listed memory ratios on the motherboard.

s754, s939, AM2, AM2+, AM3, and AM3+ all had flexible and stable bclk/htt OC. It was beautiful. You could overclock anything from a lowly Sempr0n to a big-boy FX on pretty much any motherboard that could handle it. You just had to feel out the mobo's bclk limits.

Because on APUs & SoCs adjusting the BCLK is not a good idea, since there are various other frequencies tied to it.

 

[krumme]

Is the memory controller in Zen an off the shelf or custom design?

It was rumored in august it was Rambus tech. Dont know if its confirmed but i read it somewhere recently again...cant find it using google. Perhaps its in the canard article somewhere?

 

[iBoMbY]

Is the memory controller in Zen an off the shelf or custom design?

It is probably based on some Synopsys IP. They have several.

Edit: I would guess this one: DesignWare LPDDR4 multiPHY IP

Design in 28-nm and below, including 14/16-nm FinFET, that requires high-performance mobile SDRAM support (LPDDR4/3) up to 3200 Mbps and/or high performance DDR4/3 support up to 3200 Mbps for small memory subsystems.

 

[superstition]

It is probably based on some Synopsys IP. They have several.

Edit: I would guess this one: DesignWare LPDDR4 multiPHY IP

If the Zen controller can support DDR4 3200 and boards coming out in the first iteration don't that suggests that the board makers are using RAM speed as a planned obsolescence lever to get people to get version 2 of the boards. But this would require all board makers that put out competitive enthusiast-grade first iteration boards to be in on it. We've seen DRAM makers collude in the past so it's not like that can't happen. It just seems unlikely.

The other thing is that if that spec says it goes up to 3200 then why would anything up to and below that be labeled "OC" by board makers.

 

[jpiniero]

LPDDR4 would be nice considering Skylake and Kabylake don't support it and Coffee Lake might not either.

 

[DrMrLordX]

Because on APUs & SoCs adjusting the BCLK is not a good idea, since there are various other frequencies tied to it.

Back in the Socket 1 days (think pre-Celeron300a) that was also a problem. Motherboard OEMs had to figure out how to lock down sensitive system busses that were tied to FSB. Once they cut their teeth on bus locks, everything was fine.

Now that we have SoCs and APUs, the locks must be implemented by the CPU/APU provider rather than the mobo OEM. But the same principle applies.

 

[cytg111]

Back in the Socket 1 days (think pre-Celeron300a) that was also a problem. Motherboard OEMs had to figure out how to lock down sensitive system busses that were tied to FSB. Once they cut their teeth on bus locks, everything was fine.

Now that we have SoCs and APUs, the locks must be implemented by the CPU/APU provider rather than the mobo OEM. But the same principle applies.

Running my 300a at 504 was running the pci bus(41?) wildly out of spec too but find the components that could handle it and you were flying .. kicked intels flagship P2-450's behind real good too(fullspeed ondie l2 vs halfspeed off die).

 

[DrMrLordX]

Yeah some of the bus locks weren't in place on all the socket 1 boards. By the time I started overclocking in the s754 days (didn't OC my k6-233, the borrowed k6-2 350, or the 1.4 GHz Tbird) everything was locked down, even on a cheap Chaintech board (vnf3-250).

 

[The Stilt]

Back in the Socket 1 days (think pre-Celeron300a) that was also a problem. Motherboard OEMs had to figure out how to lock down sensitive system busses that were tied to FSB. Once they cut their teeth on bus locks, everything was fine.

Now that we have SoCs and APUs, the locks must be implemented by the CPU/APU provider rather than the mobo OEM. But the same principle applies.

Simply, no.

Completely different thing & reasons back in the day. The frequencies are linked on purpose, to simplify the design and to increase the efficiency.

 

[iBoMbY]

The other thing is that if that spec says it goes up to 3200 then why would anything up to and below that be labeled "OC" by board makers.

Maybe because the highest specified JEDEC standard is DDR4-2400, as far as I know.

Edit: Or maybe that changed. DDR4-2666 and DDR4-3200 do seem to be standardized by now. If they are using an older version of the spec, and maybe an older controller version, everything that hasn't been specified at that point, could be seen as overclocking. Or maybe it's just not stable with every module over a certain point.

 

[DrMrLordX]

The frequencies are linked on purpose

Then that is a bad thing for OC hobbyists . . .

 

[The Stilt]

Then that is a bad thing for OC hobbyists . . .

I see no issue in that.
You seem to expect that the different configurations available through ratios would be somehow insufficient?
Personally I see no merit in higher than standard BCLK on it's on.

 

[DrMrLordX]

I see no issue in that.
You seem to expect that the different configurations available through ratios would be somehow insufficient?
Personally I see no merit in higher than standard BCLK on it's on.

Raven Ridge. Unless they come through with HBM2, dual-channel DDR4 will be a bottleneck. Also, someday AMD will probably sell a Zen-based CPU or APU that can not be overclocked via multipliers, so it's bclk or bust for those chips.

 

[SpaceBeer]

Dual channel DDR4 won't be bottleneck for 8-10 CUs in Raven Ridge if it can fully utilize RAM @3000 MHz or above. And those modules are not so expensive and will probably be cheaper by the time RR is released

 

[DrMrLordX]

Dual channel DDR4 won't be bottleneck for 8-10 CUs in Raven Ridge if it can fully utilize RAM @3000 MHz or above. And those modules are not so expensive and will probably be cheaper by the time RR is released

Dual-channel DDR3-2400 already bottlenecks an A10-7700k fully overclocked, and that's only 6 CUs. DDR4-3000 will certainly be a bottleneck.

 

[SpaceBeer]

Yes, but memory controllers on old(er) APUs/CPUs were not so good, comparing to Intel's. There is/was a lot of room for improvements in that area also. If GCN IGPs were in Skylake CPUs, it would be much faster

 

[bjt2]

I remember that locked CPUs, at least for AMD, had only the cpu multiplier locked. The memory and even NB multiplier were unlocked. The problem was that most BIOSes lacked the advanced settings for the memory... Do i remember well?