AMD Ryzen Gen 2 Set For Q2 2018

[CatMerc]

Can't view the HTML pages, anything particularly interesting there?

Just the writeup, nothing of note really that wasn't already known.

 

[IRobot23]

You cannot. Having the multiplier doesn't mean any CPU/RAM/Mobo combination can run it.

Well some guys got 4000MT/s on single channel, but my point was that it actually supports those clocks, but can it run stable at normal V...

 

[DisEnchantment]

https://videocardz.com/75381/amd-ryzen-7-2700x-ryzen-5-2600-review-posted-ahead-of-launch

2700X

2.2-3.7-4.2GHz (22x-37x-42x) [+9% rated, +11% turbo]

If this is all, is AMD trying to pull a fast one? I mean if the fastest SKU is 2700X and we are not comparing against the fastest SKU from last gen and then claim we have +10% faster clock over the last gen... seems weak to me even by Intel standards.
What about that comment from Papermaster, tock tock tock.
Compared to 1800X (right now it cost like what, USD 329)
+0.1 GHz Base, +0.2 GHz boost, but +10 W TDP, +40 USD (at launch, if leaks are true, but they seems to be true now)


... Unless there will be a 2800X.
XFR 2.0 Enhanced/Precision Boost 2.0 Overdrive will be the deal here ?

 

[CatMerc]

2700X


If this is all, is AMD trying to pull a fast one? I mean if the fastest SKU is 2700X and we are not comparing against the fastest SKU from last gen and then claim we have +10% faster clock over the last gen... seems weak to me even by Intel standards.
What about that comment from Papermaster, tock tock tock.
Compared to 1800X (right now it cost like what, USD 329)
+0.1 GHz Base, +0.2 GHz boost, but +10 W TDP, +40 USD (at launch, if leaks are true, but they seems to be true now)


... Unless there will be a 2800X.
XFR 2.0 Enhanced/Precision Boost 2.0 Overdrive will be the deal here ?

How is AMD pulling a fast one when it's sisoftware testing?

And it's not priced like last gen's flagship, it's priced like the 1700X, so...

 

[DisEnchantment]

I could just shut up and not say anything and wait until next April, but since we are discussing...
I'm speculating... Let me just drop the 'last gen flagship'
If I am buying new, what more can I get from 2700X compared to 1800X which I can get at USD 329? I mean so far the leaked slides have been correct despite the mistakes.

 

[jpiniero]

I could just shut up and not say anything and wait until next April, but since we are discussing...
I'm speculating... Let me just drop the 'last gen flagship'
If I am buying new, what more can I get from 2700X compared to 1800X which I can get at USD 329? I mean so far the leaked slides have been correct despite the mistakes.

Probably 5-10% faster.

 

[Mech0z]

I could just shut up and not say anything and wait until next April, but since we are discussing...
I'm speculating... Let me just drop the 'last gen flagship'
If I am buying new, what more can I get from 2700X compared to 1800X which I can get at USD 329? I mean so far the leaked slides have been correct despite the mistakes.

Better memory compatibility, maybe higher OC, but thats about it I suppose

 

[moinmoin]

I could just shut up and not say anything and wait until next April, but since we are discussing...
I'm speculating... Let me just drop the 'last gen flagship'
If I am buying new, what more can I get from 2700X compared to 1800X which I can get at USD 329? I mean so far the leaked slides have been correct despite the mistakes.

Better multicore turbo. 1xxx essentially stops the boost after two cores, while with Precision Boost 2 (as seen already in Raven Ridge) the decrease from the max frequencies whenever more cores are used will be more gradual.

 

[formulav8]

What about that comment from Papermaster, tock tock tock.

I believe he was referring to Zen, Zen 2, and Zen 3. Not this refresh part (Ryzen 2).

 

[Dayman1225]

I believe he was referring to Zen, Zen 2, and Zen 3. Not this refresh part (Ryzen 2).

I mean you can't really exclude one and keep calling it Tock Tock Tock. It'd be like Intel excluding Coffee Lake and Kaby Lake then claiming Tick Tock was still alive

 

[piesquared]

He was referring to performance. I bigger increase compared to inel's tick tock scheme.

 

[formulav8]

I mean you can't really exclude one and keep calling it Tock Tock Tock. It'd be like Intel excluding Coffee Lake and Kaby Lake then claiming Tick Tock was still alive

I believe he was referring to Zen cores and not a Ryzen refresh.

 

[DrMrLordX]

Ok, I found another geekbench entry



Also, both are same geekbench version, same amount of RAM and frequency, but 1800X at 4000 MHz.
Not sure about kit latencies though.

Also that ASUS BIOS 0401 on X370 is already fairly stable.

Damn... the spread on these geekbench scores ...

Sorry to reply to this same message again, but I figured hey, why not run GB4 on my own Ryzen to see how it stacks up? Turns out the latest version is now 4.2.2 (could not find 4.1.1 available for download), but I'm not sure the version change makes a big difference. My score:

https://browser.geekbench.com/v4/cpu/7524482

ST 4890
MT 28715

Something tells me that RAM speed/timings made a difference. FWIW Aida64 reads lower latency and higher mem bandwidth scores than GB4.

 

[CatMerc]

Sorry to reply to this same message again, but I figured hey, why not run GB4 on my own Ryzen to see how it stacks up? Turns out the latest version is now 4.2.2 (could not find 4.1.1 available for download), but I'm not sure the version change makes a big difference. My score:

https://browser.geekbench.com/v4/cpu/7524482

ST 4890
MT 28715

Something tells me that RAM speed/timings made a difference. FWIW Aida64 reads lower latency and higher mem bandwidth scores than GB4.

Of course it does, it's quite literally part of the score, even ignoring the workloads that benefit from faster memory.

The memory scores are 20% of the score in their weighted average system.

 

[Hans Gruber]

I have the MSI B350 gaming pro Carbon. Latest bios update made it so that my Team Dark Pro 3200mhz 8GB kit boots at default settings (XMP) in Bios or the AMD equivalent. The Ryzen architecture scales well with higher memory. Intel CPU's do not improve based on fast memory. So those memory problems of the past for AMD have been ironed out in the newest bios.

I assume that with the Ryzen refresh (2) will not have memory issues.

 

[DrMrLordX]

Of course it does, it's quite literally part of the score, even ignoring the workloads that benefit from faster memory.

The memory scores are 20% of the score in their weighted average system.

Point being, I think IMC, cache, and IF performance differences are likely the culprit in the score delta between the posted 1800x and 2700x results. 1800x chips can certainly get much, much better scores than the ones posted.

Even taking into account that memory performance is 20% of the GB4 score total, you're still talking a ~700 point spread (+17% ST) between my 1800x and the one posted when the mem scores alone should only account for a ST score delta of 144 actually I think I'm extrapolating that score delta backwords; without seeing more details on that posted 1800x run, I can't tell for sure what were the actual memory scores.

edit: In the ST segment of the program, my total memory score was 6646; assuming that's 20% of my ST score, the ST contribution would be 1329. The 1800x with DDR4-2400 running identical timings would (assuming perfect downward scaling) score 4985, contributing 997 to the final score of that chip. That would be a delta of 332. Note that I am making several assumptions here. That still doesn't explain the 700 point difference in total, nor does it touch on the MT results . . .

The MT score delta is massive. I have to wonder, how much does GB4 feature interthread communication between CCXs or thread reassignment between CCXs?

I have the MSI B350 gaming pro Carbon. Latest bios update made it so that my Team Dark Pro 3200mhz 8GB kit boots at default settings (XMP) in Bios or the AMD equivalent.

That's . . . really interesting! I wonder if newer UEFI revs from other OEMs are showing the same behavior? Oh the thought of my Ryzen running DDR4-3733 . . .

I assume that with the Ryzen refresh (2) will not have memory issues.

Hope you're right, especially for people with cheap(er) Hynix kits.​

 

[Hans Gruber]

Point being, I think IMC, cache, and IF performance differences are likely the culprit in the score delta between the posted 1800x and 2700x results. 1800x chips can certainly get much, much better scores than the ones posted.

Even taking into account that memory performance is 20% of the GB4 score total, you're still talking a ~700 point spread (+17% ST) between my 1800x and the one posted when the mem scores alone should only account for a ST score delta of 144.

Also, the MT score delta is massive. I have to wonder, how much does GB4 feature interthread communication between CCXs or thread reassignment between CCXs?

That's . . . really interesting! I wonder if newer UEFI revs from other OEMs are showing the same behavior? Oh the thought of my Ryzen running DDR4-3733 . . .



Hope you're right, especially for people with cheap(er) Hynix kits.​

My next step is OCing the Dark Pro 8GB Kit beyond 3200mhz. The bios is a Jan 2018 and those heatsinks on it are the heaviest, sturdiest things I have ever seen or felt on DIMM's. I have my AMD test system using a Ryzen 3 1200 OC'd to 3850mhz. It's watercooled with a 240mm cooler master AIO. My plan was always to upgrade to a 6 or 8 core Ryzen.

 

[PcPete69]

Updated my bios yesterday, gigabyte ab350m, my ram now runs at rated 3000mhz with tighter timings, cheap crucial ballistic tactical 3000mhz

 

[IRobot23]

Updated my bios yesterday, gigabyte ab350m, my ram now runs at rated 3000mhz with tighter timings, cheap crucial ballistic tactical 3000mhz

3000MT/s is relatively easy to reach with ryzen these day, maybe some stick could give you some problems... wrong (faster) timings. I bought few 2x4GB 3466MT/s CL16 sticks for budget and out of the box for everyone 3200 MT/s with XMP profile ... (R5 1400 & 1600, R3 1200) 1.05V SoC and OC to around 3,6-3,8GHz depends on VRM on the board. Boudht few cheapeer stick 3000MT/s and 2800MT/s (hynix die) and XMP working out of the box.

Some people still may have problems, but probably on their part unless they were really unlucky.

 

[piesquared]

Nearly all of the RAM is running at rated sped now, or damn close to it. I've had mine running at full speed for a long while, and it's dual ranked. AMD has sorted this RAM teething seemingly much better than when intel was having so many issues with their DDR4 controller.

 

[DrMrLordX]

Time to throw more data onto the fire!

Per the data posted by @DisEnchantment as well as my own results, I decided to rerun GB4.2.2 @ DDR4-2400 using the same timings I use at DDR4-3466. Results:

https://browser.geekbench.com/v4/cpu/7540430

ST 4526
MT 26096

Can the differences between GB4.1.1 and GB4.2.2 be that big? If I could hunt down 4.1.1 I could juxtapose where appropriate.

Of interest: the memory portion of the ST score went down from 6646 to 5009 when lowering DRAM speed from 1733 to 1200 MHz. My projection of a score of 4985 was close, but not exact. If you take 20% of the difference in scores between the 1733 and 1200 MHz runs, you get 327 points, which is almost the difference between my DDR4-3733 ST score (4890) and my DDR4-2400 ST score (4526). That leaves unaccounted a mere 37 points . . . mere statistical noise. The only other score that changed was the Crypto portion. FP and INT didn't change much.

The plot thickens. How did anyone manage to get an 1800x to score that badly @ 4 GHz? Even running DDR4-2133 @ CL14 wouldn't bring my score that low. Something tells me the 1800x system was throttling or misreporting clockspeed.

 

[Gideon]

There are also two new 2700X benches with geekbench version 4.2.2. One with god-awful 2133MHz, scoring 4650/24877 memory the other one at 3200MHz, scoring 4903/26514. Which I guess is still pretty good, considering it's running at stock.

 

[DrMrLordX]

There are also two new 2700X benches with geekbench version 4.2.2. One with god-awful 2133MHz, scoring 4650/24877 memory the other one at 3200MHz, scoring 4903/26514. Which I guess is still pretty good, considering it's running at stock.

Thanks for finding those. I am guessing that those 3.7 GHz scores indicate that the chips are running at default clockspeeds (no overclock) so boost/turbo can take over where applicable?

When I ran my 1800x with all default settings (including DDR4-2133, yay) I got this:

https://browser.geekbench.com/v4/cpu/7544232

Still better than the earlier-cited 4 GHz 1800x. It looks like the GB4.2.2 scores match up with 4.1.1 pretty well so I still can not explain why that 1800x scored so badly. Even at bone stock settings, mine is faster. I also tried running at a fixed 3.7 GHz with DDR4-2133 and got an ST score of around 4100, but MT was basically the same.

 

[IRobot23]

This is my at 3,8GHz (or 99,8*38) with 3400MT/s 2x4GB custom timings.
https://browser.geekbench.com/v4/cpu/7545340

latency is definitely lower

 

[DisEnchantment]

https://www.anandtech.com/show/680/6

10GHz by 2005 running at < 1 volt

According to Anand, Intel is at 10 GHz long way back so AMD better get those clocks up.

Reveal is on Monday
https://videocardz.com/75489/amd-ryzen-7-2700x-pictured-full-review-on-monday

XD