Jump to content
Sign in to follow this  
Dasa

AMD Zen

Recommended Posts

The rx500 series is not Vega, it's like Polaris v2 and not a 1080 killer.

Edited by Jeruselem

Share this post


Link to post
Share on other sites

The rx500 series is not Vega, it's like Polaris v2 and not a 1080 killer.

 

I expected as much since Polaris should last a lot longer than Tonga/Tahiti/Hawaii. But Vega is the new architecture too since it rebuilt from GCN.

Share this post


Link to post
Share on other sites

Ryzen 5 out today. I just watched this review from LinusTechTips.

 

Share this post


Link to post
Share on other sites

To be honest, the 1500X is not really worth it either. For $40 more, you're getting 2 extra cores and additional 4 threads. Plus, you're compromising more with the 1500X for the future since we assume more developers will be optimising for multiple cores.

Share this post


Link to post
Share on other sites

To be honest, the 1500X is not really worth it either. For $40 more, you're getting 2 extra cores and additional 4 threads. Plus, you're compromising more with the 1500X for the future since we assume more developers will be optimising for multiple cores.

 

True, the 1600 is better value. Given Ryzen barely overclocks above 4.1 Ghz ... the 1600X might be a waste.

Share this post


Link to post
Share on other sites

It seems the vega gpu is based on the fury x so its not a true redesign like ryzen.

Edited by Jeruselem

Share this post


Link to post
Share on other sites

ROG got the 1600x to 5.9 Ghz !!!

 

On liquid nitrogen that is.

 

Kaby lakes OC to 7.0 on liquid nitrogen.

Edited by Jeruselem

Share this post


Link to post
Share on other sites

Remember that Kaby Lake has been nothing but a tweaked derivative of Sandy Bridge really. If you look at the core architecture, it is literally Sandy Bridge but shrink and optimised. Of course, it'll clock high. Compare this to Zen, you can't expect massive clocks out of the gate. I remember the E6600 had trouble getting up to 3GHz at first then with the i7-7X0 on, it began to clock higher thanks to architecture refinement...

Edited by sora3

Share this post


Link to post
Share on other sites

Sandy Bridge? There's 4 generations & 6 years difference there.

 

I've got at least one example from about half the Core CPU generations - there's massive improvements along the line in stuff like FP performance, and noticable improvements in plenty of other areas.

The overclockabiltiy - hampered in some cases by the chipset, in others by crappy motherboards, and in some by design intention.

Share this post


Link to post
Share on other sites

I don't think I've made myself clear what I meant.

 

If you look at the inherent micro-architecture, you can see that all the cores since Sandy Bridge has been a optimised design of Sandy Bridge.

 

E.g. Let's take a look at the micro-architecture of Sandy Bridge, courtesy of Real World Technologies...

sandy-bridge-2.png

 

Now, let's jump further on into Haswell. Again, the diagram is courtesy of Real World Technologies...

haswell-1.png

 

You can see that the design is clearly a refined version of what Sandy Bridge is. If I had to guess, only the branch predictors along with the additional AVX instructions are what makes Haswell better than Sandy Bridge and the derivatives since.

 

Compare this to Zen, it is clear AMD is attempting the same as what Core did to Intel. A good architecture to build and optimise for future generations, just like Intel. So far, the signs are good and if they can refine the low overclocking ceiling and as well as new instructions, we might very well see AMD gaining better share since the Athlon64.

Share this post


Link to post
Share on other sites

There's only so many things you can change.

If we ignore stuff like clock speed, extra instructions, media codecs and litho process shrinks, then the main improvements since the 1st-gen i7 have probably been in the FP processing, with some gains in hyperthreading.

 

Given that they've been improving design on the same set of core features ever since Ram access speed fell significantly behind CPU speed (ie '286 era), there's only so far you can improve stuff like branch prediction, cache management, instruction decode, etc.

 

Also, it's not Intel needing to play catchup. Ryzen is essentially AMD finally catching up to (and overtaking) what the Core2 era CPUs were capable of.

Having models with 6, 8, 12 cores is great and all, but just have a look at the list of Xeon CPUs with insane number of cores to see what Intel has been doing for years anyway (fair enough, beyond the sense of most home buyer budgets) - https://en.wikipedia.org/wiki/List_of_Intel_Xeon_microprocessors

 

Sadly, not a great deal of agregated CPU performance being gathered these days... but referring to Tom's Hardware gives an idea of the woeful performance of the likes of the AMD FX line and their APU line.

http://www.tomshardware.com/charts/cpu-charts-2013/benchmarks,140.html

http://www.tomshardware.com/charts/cpu-charts-2015/benchmarks,187.html

Edited by Rybags

Share this post


Link to post
Share on other sites

I wouldn't mind one of those 24 core ryzens in one of our new servers at work but given the timing it's not going to happen.

Share this post


Link to post
Share on other sites

It also has a decent amount of VRAM too, most likely due to the use of HBM2. Shame that the CPU is still Intel.

 

Apple would have to redesign their motherboards for the iMacs so they aren't going to do that.

Share this post


Link to post
Share on other sites

There's only so many things you can change.

If we ignore stuff like clock speed, extra instructions, media codecs and litho process shrinks, then the main improvements since the 1st-gen i7 have probably been in the FP processing, with some gains in hyperthreading.

 

Given that they've been improving design on the same set of core features ever since Ram access speed fell significantly behind CPU speed (ie '286 era), there's only so far you can improve stuff like branch prediction, cache management, instruction decode, etc.

 

Also, it's not Intel needing to play catchup. Ryzen is essentially AMD finally catching up to (and overtaking) what the Core2 era CPUs were capable of.

Having models with 6, 8, 12 cores is great and all, but just have a look at the list of Xeon CPUs with insane number of cores to see what Intel has been doing for years anyway (fair enough, beyond the sense of most home buyer budgets) - https://en.wikipedia.org/wiki/List_of_Intel_Xeon_microprocessors

 

Sadly, not a great deal of agregated CPU performance being gathered these days... but referring to Tom's Hardware gives an idea of the woeful performance of the likes of the AMD FX line and their APU line.

http://www.tomshardware.com/charts/cpu-charts-2013/benchmarks,140.html

http://www.tomshardware.com/charts/cpu-charts-2015/benchmarks,187.html

I'd disagree with regards to the hyperthreading. They haven't really touched that area given what many tech sites have gathered from the micro-architecture diagrams from Intel.

 

If I have to see it from an engineering point of view, they've reached the limit of the Core architecture.

 

Along with this, the reason why Intel has a lot of insane core products is because of no competition.

 

And honestly, I have an APU, FX and a console CPU to some degree. All of these have done what I wanted easily as long as I didn't exceed the limits. Zen kicks the low ceiling Intel arbitrarily limited.

 

In terms of Threadripper, it'll be interesting to see how much it'll be compared to Ryzen.

Share this post


Link to post
Share on other sites

Gains in other areas can flow through to HT since it's essentially a virtual core system that shares the available resources - so there is the chance you'd have hyperthreading improvements without really doing anything in that area.

 

As a comparison, 4T vs 8T wPrime on a Bloomfield Xeon vs Haswell i7. The older machine has a 38.9% improvement, newer has 44.5% (OK, for a proper test in this regard we'd probably want to disable HT in the Bios since there's a chance that a 4T task could be dispatched such that there's threads sharing the same physical core, which would penalize the 4T results a bit. But the relative results would probably remain much the same).

 

The next step would be to do relative comparison with older P4 gear - I actually have some older Celeron and Pentium LGA775s laying around so might be able to give that a go.

 

As for the ceiling - let's just say the the x86 architecture in itself is a bloated evolution with plenty of inefficiencies. The "best" thing they could do is just start all over but that's not likely to happen.

Share this post


Link to post
Share on other sites

Gains in other areas can flow through to HT since it's essentially a virtual core system that shares the available resources - so there is the chance you'd have hyperthreading improvements without really doing anything in that area.

 

As a comparison, 4T vs 8T wPrime on a Bloomfield Xeon vs Haswell i7. The older machine has a 38.9% improvement, newer has 44.5% (OK, for a proper test in this regard we'd probably want to disable HT in the Bios since there's a chance that a 4T task could be dispatched such that there's threads sharing the same physical core, which would penalize the 4T results a bit. But the relative results would probably remain much the same).

 

The next step would be to do relative comparison with older P4 gear - I actually have some older Celeron and Pentium LGA775s laying around so might be able to give that a go.

 

As for the ceiling - let's just say the the x86 architecture in itself is a bloated evolution with plenty of inefficiencies. The "best" thing they could do is just start all over but that's not likely to happen.

AMD were thinking ARM CPUS would replace x86/x64 as servers but ARM architecture itself has it's own problems. Given we have this ryzen series and new intel cpus, the difference between pc/server and tablet/phone computing capacity is still stark.

Edited by Jeruselem

Share this post


Link to post
Share on other sites

A mixed bag of HT results with an old Pentium 4 540J 3.2 GHz (single core).

wPrime gets a 37.3% improvement running 2 vs 1 threads.

yCruncher, result too close to call it beyond normal variance over different runs.

As would be expected, Geekbench 3/4 single/multi results much closer to each other relative to proper multicore CPUs.

 

So, maybe we can say "barely any HT improvement 2008 vs 2004" although in my experience number crunching stuff has massively impoved in most generations.

Share this post


Link to post
Share on other sites

Hyper threading I would say would give max 50% boost.

Its better than nothing. Old pentium 4 ht was awful though ...

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this  

×