AMD details 2018 plans for Ryzen CPUs

[CAT-THE-FIFTH]

EDIT:

Re that CPU performance slide; the 2200G and 2400G have higher clock speeds than the 1200 and 1400, which makes those performance increases look less than stellar, IMNSHO - looks like the IPC drops off with the reduction to 4MB of L3 cache...

*EDIT 2: That would actually tie in with the leaked slide we saw showing there would be "Ryzen Gaming" APUs in the premium mobile market - if they were also an EMIB package like the Kaby-G parts.

The 1400 has a top clockspeed of 3.4GHZ and the 2400G a top clockspeed of 3.9GHZ,ie,around 15% it seems,but we don't know for sure if the 2400G will consistently hit 3.9GHZ though. The increase in CB ST scores is 11% though.

The 1400 base clockspeed is 3.2GHZ and the 2400G has a base clockspeed of 3.6GHZ,ie,12.5% but the score goes up by 12.5% so it seems at least for CB its not really affected by the reduction in L3 cache.

I also suspect for gaming having one CCX instead of two will actually help mitigate any potential loss in performance,and should make performance with a dGPU less dependent on higher speed RAM IMHO OFC.

Edit!!

Something else which is interesting too.

The Ryzen 3 2200G is a 3.5GHZ~3.7GHZ part and the Ryzen 3 1200 is a 3.1GHZ to 3.4GHZ part and the Ryzen 5 1400 is a 3.2GHZ to 3.4GHZ part.



Look at the CB ST score - the 2200G is 5% faster for 9% increase in clockspeeds(assuming it can hit is max Turbo clockspeed). The Ryzen 3 1200 and Ryzen 5 1400 are hitting their max boost clockspeeds though.

Now look at the increase in base clockspeeds of the 2200G over the 1200,ie,around 13% extra clockspeed. The increase in performance is between 7% to 15% in MT scores.

At least comparing the Ryzen 3 1200 and 2200G,it looks like the reduction to 4MB of L3 cache from 8MB has not done much.

[The Hand]

Depending on settings, probably no issue for Civ V, but Civ VI might get it hot under the collar (leave off the high beard density options!). I was under the impression nobody plays BE... Not sure what Stellaris is like, but if HoI:IV is anything to go by, you will want to turn off some of the fancy effects. Elite Dangerous is probably a step too far, if you want it to be enjoyable anyway.

Oops, I've just outed myself as a BE player lol. I don't actually have Civ 6 yet, so I suppose if the demo does struggle I could leave it out for the time being. Some people are playing Elite moderately well on a RX560 at 1080p so maybe the Vega 11 will cope ok, if it's comparable to a RX 560.

[CAT-THE-FIFTH]

Oops, I've just outed myself as a BE player lol. I don't actually have Civ 6 yet, so I suppose if the demo does struggle I could leave it out for the time being. Some people are playing Elite moderately well on a RX560 at 1080p so maybe the Vega 11 will cope ok, if it's comparable to a RX 560.

The RX560 will be faster as it has 1024 shaders and has GDDR5. I would expect the Vega11 IGP with some overclocking and very fast RAM to get closer to a RX550.

[The Hand]

The RX560 will be faster as it has 1024 shaders and has GDDR5. I would expect the Vega11 IGP with some overclocking and very fast RAM to get closer to a RX550.

Probably best if I wait for the reviews then, I was under the impression that the ryzen apus were going to be comparable to the ps4 graphics performance. Probably closer to Xbox one level.

[DanceswithUnix]

I presume these APUs will get used for the CPU to drive mining rigs as it gets a few extra CUs on tap for a slightly higher hashrate. So I won't be able to buy one.

[scaryjim]

... look at the increase in base clockspeeds of the 2200G over the 1200,ie,around 13% extra clockspeed. The increase in performance is between 7% to 15% in MT scores.

At least comparing the Ryzen 3 1200 and 2200G,it looks like the reduction to 4MB of L3 cache from 8MB has not done much.

That's a generous reading! Don't forget that the 2200G has Precision Boost 2, which should theoretically allow it to maintain higher boost speeds in multithreaded workloads. And in Cinebench and 7Zip it does manage that. But in 3 tests it's only 7% faster for a 12% higher boost clock, and in one it's barely faster at all! Plus in the one single threaded test it's only 4% faster for a 9% increase in clock speed. So in 4/6 MT tests and one ST test it's slower than you'd expect based on clock speeds. The only architectural differences are PB2, which should increase performance, and the different cache arrangements.

The single-CCX design would only help if it was missing less often in the cache. With half the cache amount per thread, you're missing more often in cache, so you're going back to main memory which is just as expensive as going cross-CCX. The 2400G has a 12.5% base clock advantage on the 1400, plus PB2, yet its only Cinebench which it gets > 10% gains, and in three of those tests its gain is < 5%.

By clock speed advantage, these should be a lot faster than they are. The only reason I can they wouldn't be is the smaller cache. And the single CCX design isn't going to help when you're going back to main memory more often.

EDIT: to be fair, there is another option than cache - they might not be hitting the target clockspeeds within TDP. But that'd be a poor bit of marketing fluff from AMD if that was the case...

[Tarinder]

Why does Hexus always use tiny images ? Usually for things like this you can click on em for a bigger picture, but I've noticed on Hexus that's never the case, they're always just embedded and crappy. Cmon lads I'm getting old my eyes don't work as well anymore!

It was because only watermarked slides were available before launch. I will update the article with high-resolution images today.

[adidan]

So Ryzen+ will be the 2000 and then 2019(?) will see Ryzen 2 (presumably 1000 again?).

[ET3D]

The really fascinating thing about this slide is that they benchmarked the 2000G series with 2667 RAM and the 1000 series with 2933 RAM. I'm used to silliness in internal benchmarking, but said silliness usually tries to give the edge to the new stuff, not the old stuff.

[Xlucine]

So Ryzen+ will be the 2000 and then 2019(?) will see Ryzen 2 (presumably 1000 again?).

More likely 3000

[CAT-THE-FIFTH]

That's a generous reading! Don't forget that the 2200G has Precision Boost 2, which should theoretically allow it to maintain higher boost speeds in multithreaded workloads. And in Cinebench and 7Zip it does manage that. But in 3 tests it's only 7% faster for a 12% higher boost clock, and in one it's barely faster at all! Plus in the one single threaded test it's only 4% faster for a 9% increase in clock speed. So in 4/6 MT tests and one ST test it's slower than you'd expect based on clock speeds. The only architectural differences are PB2, which should increase performance, and the different cache arrangements.

The single-CCX design would only help if it was missing less often in the cache. With half the cache amount per thread, you're missing more often in cache, so you're going back to main memory which is just as expensive as going cross-CCX. The 2400G has a 12.5% base clock advantage on the 1400, plus PB2, yet its only Cinebench which it gets > 10% gains, and in three of those tests its gain is < 5%.

By clock speed advantage, these should be a lot faster than they are. The only reason I can they wouldn't be is the smaller cache. And the single CCX design isn't going to help when you're going back to main memory more often.

EDIT: to be fair, there is another option than cache - they might not be hitting the target clockspeeds within TDP. But that'd be a poor bit of marketing fluff from AMD if that was the case...

Not generous in anyway - you should know the clockspeeds reading the reviews for the 1200 and 1400 and they should have no problem hitting the clockspeeds I mentioned. The fact is Ryzen does not have a large IGP drawing power and heat,yet Raven Ridge does,so that does draw power and produce more heat which will impact clockspeeds.

As mentioned later the Ryzen 1000 series CPUs are also running faster RAM,ie,2933MHZ RAM as against 2667MHZ RAM which will add a few percent to the scores(not a massive amount),so I like I said - I don't think the IPC difference will be that different,and only stuff which really requires loads of L3 cache might be affected and uses all cores will be.

Plus the other thing is for more lightly thread tasks since the Ryzen 3 2000 has two cores per CCX so has 4MB of L3 cache per set of two cores,so technically for lightly threaded benchmarks using one to two cores,the Ryzen 3 2000 series will still have access to the same amount of L3 cache,and you could argue it will be for things that use more threads where any impact in the cache reduction would be seen.The thing is with the Intel CPUs,you see reductions in L3 cache amounts going from Core i7 to Core i5 and from Core i3 to Pentium CPUs,and it does not have a massive impact AFAIK.

The main thing is this the first AMD APU to actually have L3 cache - all the others have tended to lop off the L3 cache entirely.

So even with slower RAM,the Ryzen 2000 series is faster than the Ryzen 1000 series SKUs they are replacing in CPU performance,which if anything makes them look far better!

The issue with the dual CCX design is if the game engine is not aware of it,it will lead to latency issues and this is why I would like to see some games based on older engines compared with Ryzen at the same clockspeed.

Edit!!

The Sandra test is fully synthetic,but if you average the others,you are looking at around a 9.5% increase in performance,and that actually is roughly the difference between the Ryzen 3 1200 and Ryzen 3 1300X with its extended XFR.

[scaryjim]

... even with slower RAM,the Ryzen 2000 series is faster than the Ryzen 1000 series SKUs they are replacing in CPU performance,which if anything makes them look far better! ...

That's kind of my point though - they clock like the X parts, but they're replacing the non-X parts. Until we get like-for-like systems tested across a wider range of benchmarks we won't know for sure, but the way AMD have presented this looks like the Ryzen APUs are slower, clock for clock, than the comparable CPUs.

[globalwarning]

"Ryzen Threadripper 1900X sees a $100 snip, the official SEP of the 1800X is decimated to $349, to better match up with Intel's premier mainstream chip, and the rest see a smaller reduction."

Decimate = reduce to 10%...

[kompukare]

"Ryzen Threadripper 1900X sees a $100 snip, the official SEP of the 1800X is decimated to $349, to better match up with Intel's premier mainstream chip, and the rest see a smaller reduction."

Decimate = reduce to 10%...

Well, if we want to be totally pedantic the original meaning was to reduce by 10% as in meaning 1. below



Also the etymology makes that fairly apparent.

Borrowed from Latin decimāre (“to take or offer a tenth part”), from decimus (“tenth”).[1] As a noun, via Latin decimatus (“tithing area; tithing rights”).[2]

All from the Wiktionary:
https://en.wiktionary.org/wiki/decimate

[adidan]

So they've double decimated it

Vigintimated the price?

[3dcandy]

So they've double decimated it

Vigintimated the price?

It don't quite work does it... possibly would be vigesimated