First Look: AMD's Threadripper 2000 Series Versus Intel's High-End Desktop Chips

Toggle sidebar Toggle sidebar
Status
Not open for further replies.
Cinebench would be like the best case scenario for ryzen specially if it is 32 cores.

What about other benchmarks ?! I don't think anyone buys CPU just to do 3d rendering/cinebench (and Ryzen seems to slightly better job in cinebench even compared to other 3d rendering benchmarks). What about video encoding, compression etc. Please do more benchmarks instead of one benchmark that AMD uses for Ryzen advertisement. One benchmark does not tell you the overall performance
 


And that 28 Cores demo was showing that? The only company using disruptive tactics lately is Intel.

By the way, one of the slide show similar pattern on 4 different benchmarks.

 
GIGASTRASH:
I guess you missed this line: "We're hard at work testing as you read this, and will provide plenty of our own benchmarks that measure performance in a broader range of workloads when we can."
 


We already knew Intel was releasing an 8 core i7 9900k so how is this a ghost announcement, the only surprising thing is supposedly making the i7 9700k an 8 core no hyperthreading part. I guess they couldnt get anything more out of 14nm+++++++++++++
 

You already knew because it was rumored, or because Intel announced it?

If they're leaking things about products that haven't been officially announced (and I didn't see any evidence that the linked leak was via them, rather than one of their partners), it could certainly be called it a ghost announcement.
 
Even more important:

Intel's processors still have "Bug Inside" as INTENDED and OS kernel relocation is still required, but AMD's don't!
 
OK. What if we dont want the hassle of running liquid nitrogen cooling?
How many watts does this dissipate at full speed? Or say 80% performance?
What level of cooling system versus day to day practical use does this require?
 
I've just started reading this site and it appears that from only reading a few columns, I'm not seeing "objective" writing. Maybe I'm not realizing that this site is all for AMD's benefit.

I'm all for AMD and Intel fighting for market supremacy because we the consumer benefit from it. Yes, one might bring a smaller process first but the other is shooting for an even smaller process that will take time but to write negative reviews about one and not the other, I'm just not seeing "Objective" writing.
 
It is somtimes said that "The truth is often somewhere in between".
That is the case here with the Intel 28C Demo Cinebench numbers presented in the chart (7344) and the text below it (7244). The actual number was 7334.
 


Oddly enough, there are other times that this site has been accused of pandering to Intel instead.
 
that 32 core is pretty slow when compared clock to clock, as predicted, bottlenecked by the quad channel, numa and by the fabric they use to connect their CCX. The more CCXs they add, the bigger the penalty and bottleneck, especially thanks to slow L3 cache in heavily multithreaded tasks.
 
7820 x0 7900x
1779 vs 2218
8 vs 10 cores

So in Intel case, adding 2 more cores nets you 439 points.

Intel 28 core @ 5.ghz 7334

AMD 32 core @ 5.1ghz 7618 points.

The difference is 284 for AMD while having the 0.1ghz advantage and 4 more cores.

That's actually terribly slow and terrible scalling.

so in heavily multithreaded and memory sensitive tasks the Intel will be faster.

The problem with Intel is they will ask some insane high price for it, but well, the price doesn't matter in this case really, if you really need a 28 core/32 core an, the cpu price itself isn't the issue, you spent 10 or 20 times more for ram.

So if that threadripper will be as much bottlenecked as I predict in memory heavy tasks, you would be wasting all the money you invested in RAM so depending on your workload even Intel costing 3 times more will be more effective and "cheaper" solution.

IDK, these threadrippers seem to be a waste and released for marketing purposes, would rather buy Epyc with its8 channel memory
 

What's the memory topology? Is in one channel-per-die or do they still connect 2 channels of 2 dies?

For one channel-per-die, then NUMA-aware tasks with high cache coherency would still do well on that thing. But I agree that 32 cores is generally too much for only 4 channels. That seemed a bit silly of them.
 
Cinebench R15 is an odd choice to boast about given its terrible variance at this level. See:

https://www.servethehome.com/cinebench-r15-is-now-a-broken-as-a-benchmark-and-11-5k-surpassed/

They would be better off using c-ray, with an 8K image and lots of ray depth on the sphract set.
 


Hehe yup, it's been a while since I switched to Opera again and nothing but great performance. It doesn't eat up 4+ GB of RAM like Chrome and also much smoother and faster overall experience. :bounce:
 


You DO know that since version 13, Opera is using the exact same tech as Chrome (or its bleeding edge, non-google, version: Chromium) Unless Opera has incorporated quite a few of its old features from version 12, They can play with Chromium all they want as I won't touch it.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom