AMD RyZen CPU Architecture for 2017

I just finished testing the 1700 @ 3.0GHz (a 1.0GHz difference) with ram at 3466 LL and added the results to the previous charts, be prepared for some interesting results.
 
The results:

RotTR:
rotr1ccx60.jpg
rotr2h4zop.jpg
rotr3kvafh.jpg


FC Primal:
fcprimalj7bcd.jpg


Hitman:
hitman32bxif.jpg
hitman4qzlu0.jpg
 
Last edited:
Looking at that performance of 1GHz difference, it kinda looks to me that even 3466 14 LL settings ain't fast enough to keep 4GHz chip fully fed. Thus additional clockspeed has bad efficiency.

Ryzen core is strong, but needs stronger memory than is currently available for the platform, to run at it's full speed.
 
Looking at that performance of 1GHz difference, it kinda looks to me that even 3466 14 LL settings ain't fast enough to keep 4GHz chip fully fed. Thus additional clockspeed has bad efficiency.

Ryzen core is strong, but needs stronger memory than is currently available for the platform, to run at it's full speed.

Yup, Ryzen would probably scale up to 4000+ with tight timings. They really need to work on that IMC to get faster ram supported.
 
Yup, Ryzen would probably scale up to 4000+ with tight timings. They really need to work on that IMC to get faster ram supported.

Yeah, and this kinda answers the theoritical question, if we would have Ryzen that could run at 5GHz. We simply wouldn't see great efficiency from additional clockspeed, until AMD improves their memory subsystem even further, to simply feed the beast.

Quess this is one of the low hanging fruits for zen2/3?
 
Yeah, and this kinda answers the theoritical question, if we would have Ryzen that could run at 5GHz. We simply wouldn't see great efficiency from additional clockspeed, until AMD improves their memory subsystem even further, to simply feed the beast.

Quess this is one of the low hanging fruits for zen2/3?
I'm not sure I'd call this low hanging, although it will likely come with Zen2/3 simply from faster tech in the future.

These results do make me want to see a design with HBM. Even lower timings and say 2x the bandwidth with out accounting for future changes. Get 32GB on a Threadripper with 1TB/s and life would be good.
 
Just finished testing 1500x and 1600 (by disabling cores in the BIOS, 3+3 and 2+2 settings used) with the new BIOS and tighter settings on the 3466 LL (it's LL+ now :)). These results cannot be compared with the previous results, I'll have to retest the 1700.

1500x:
Rise of the Tomb Raider(Dx12):
rotr1h0y1w.jpg
rotr2o9zey.jpg
rotr39kz36.jpg


Hitman(Dx12):
hitman1d6b43.jpg
hitman2s3lh3.jpg


Far Cry Primal:
fcpw0z1w.jpg


1600:
Rise of the Tomb Raider(Dx12):
rotr1plag7.jpg
rotr2oaz8h.jpg
rotr3txxwn.jpg


Hitman(Dx12):
hitman1c8yqm.jpg
hitman2mcbwm.jpg


Far Cry Primal:
fcpvhasz.jpg


Edit: 3466 LL+ timings if anyone wants to try them out: https://abload.de/img/timings5by9x.png
 
Last edited:
If they have stock and wants to buy one they will try to lower the price, if they want to sell they will try to rise it. At this point I wish I could have some money to spend to buy AMD stock now that is low, although with AMD you never know but with Lisa the company seems to be a clear north, leadership and products to get there. But its still risky or not as sure as others "opportunities" but I think theres a good chance AMD avoided become the next HTC.

AMD is low? It's at 5+ year highs right now!
As for Barclays, if they really did publish negative forward looking as a way to drive down prices on options, that would be illegal. But I'm confused on terminology since options are a frozen price that's either above water or not.

It'll be interesting to see if they do increase their position after calling for a $9 share price while it's trading around $14.

The real question is whether Vega will hurt share price and how much.
 
Last edited:
Why do we think Vega will hurt AMD? (Honestly curious, I haven't been keeping up as well as I should)

Wall Street is emotional and shallow. I've seen stock prices soar or plummet based on a news article that rehashed old news about company intellectual property. Literally nothing had changed but the rehash of old news drove the share price changes for hours or even a few days.

So when Vega RX reviews hit they're likely to talk about being 1.5 years behind nvidia with higher power and heat, etc. Already one opinion piece called Vega "appears to be stillborn," though that hasn't had any impact afaik.
 
AMD has been planning 2x Zeppelin + Greenland with HBM on a MCM (only Greenland & HBM on interposer, Zeppelins communicate with them with GMI-links (or is that Infinity Fabric links now?)) for a long time already
Back to this because someone posted the famous old 'Fuad fake' in the Vega thread
Vpz54Bx.jpg

Knowing what we know now that sure looks a lot like a Threadripper & a Vega glued-together doesn't it? :runaway:
 
Back to this because someone posted the famous old 'Fuad fake' in the Vega thread
Vpz54Bx.jpg

Knowing what we know now that sure looks a lot like a Threadripper & a Vega glued-together doesn't it? :runaway:
Yep, only things wrong with it are the fact that it says Zeppelin when there should be 2 and Greenland surely can't be mere "4+ TFLOPS" if it's supposed to be "highest performnace gpu". Other than that, all the codenames and such are real, confirmed much later than Fuad leaked the slide
 
Matching a 7700k 4.7Ghz + 1080 sli setup on Hitman at 1440p using the exact same settings:

Exact same frame:
7700ktfxr3.jpg

1700i4ltf.jpg
The video:

Only difference is I'm running 3466 LL+ and he is running 3200CL16 (using XMP most probably). Seems like memory performance is a limiting factor for Intel systems as well. I'm using the 1700 @ 4.0GHz and a 1080 Ti, both should be below a 1080 Sli setup pared with an overclocked 7700k.
 
Last edited:
Yup, Ryzen would probably scale up to 4000+ with tight timings. They really need to work on that IMC to get faster ram supported.
Yeah, and this kinda answers the theoritical question, if we would have Ryzen that could run at 5GHz. We simply wouldn't see great efficiency from additional clockspeed, until AMD improves their memory subsystem even further, to simply feed the beast.

Quess this is one of the low hanging fruits for zen2/3?

"Zen2" (or maybe Zen+ since Zen2 is 7nm?) with upgraded IMC may come as early as Q4 2017, within the Raven Ridge APUs.
Notice how they mention "Next-generation Zen Core" and official support for 3200MT/s DDR4 with ECC, in the slides describing Great Horned Owl and Branded Kestrel:

NML22vt.jpg
yPNZVKG.jpg


Whereas they only mention "Zen" when talking about Snowy Owl /Threadripper that only supports 2667MT/s DDR4 ECC:

WfE37zn.jpg
 
I'm not sure I'd call this low hanging.

These results do make me want to see a design with HBM. Even lower timings and say 2x the bandwidth with out accounting for future changes. Get 32GB on a Threadripper with 1TB/s and life would be good.
The memory controller is said to be (IIRC by The Stilt) very similar to the one in previous generation's Carrizo/Bristol Ridge. It is supposedly based on Synopsys off-the-shelf IP.

A conservative memory controller design for Ryzen v1 seems logical given how much else was new (process and architecture-wise).
(Particular in light of Vega's more cutting-edge memory subsystem appearing to not pan out as hoped)
 
Back
Top