• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Zen 3 (5000 Series), rumored 17% IPC gain.

Status
Not open for further replies.
HACO said:
These are emulated benchmarks, I seriously hope you know what emulation is.
Click to expand...

I do indeed.
And unless something drastically changes, emulation's still going to be in use.

You really think every single thing in MacOS will be ARM native? Ha...

All of a sudden you'll just have stuff breaking left right and centre.
Want to play KOTOR on your macbook on Steam? It'll either emulate it and have a performance cost (Assuming perfect emulation...) or it won't work
 
Martini1991 said:
I do indeed.
And unless something drastically changes, emulation's still going to be in use.

You really think every single thing in MacOS will be ARM native? Ha...

All of a sudden you'll just have stuff breaking left right and centre.
Want to play KOTOR on your macbook on Steam? It'll either emulate it and have a performance cost (Assuming perfect emulation...) or it won't work
Click to expand...
Can you please let them ramble on by themselves about ARM and Apple,as it will be ARM and Apple is the bestest and everything else is a failure. There is already a thread here about it and people lost interest(they were arguing in there too),now they are swamping this thread with stuff which is remotely nothing to do with AMD. They can go and swamp one of the Intel threads for a change.
 
Last edited:
humbug said:
Those EPYC CPU's are running at little over 2Ghz.

That's a Power9, The strongest version ARM have, its whats in those ARM servers you like talking about, the ones that prove ARM are as good as Intel / AMD... Not.. What secrete source are Apple doing to make them many multiple factors faster?

Or is it all just faked marketing? you know, Like how Intel like to use specific versions of Sandra to show how good their CPU's are.....
Click to expand...

Er, POWER9 is not ARM by any definition.
https://en.wikipedia.org/wiki/POWER9

Anyway, that x86 is a very poor ISA is no secret. IBM originally going with the 8088 for the IBM PC rather choosing for example a Motorola 68000 or 68008, was one of the worst decisions in computer history. Imagine all those years of having to worry about 64KB memory segment, 16-bit modes and the contortions programmers had to go through with x86, 286 and 386 mode were painful. Motorola 68k on the other hand, while a very complex CISC chip, had 32 bit address and data registers from the word go in 1979. Let me repeat: Motorola 68K had 32 bit registers back in 1979!

Now, as they say the rest is history and at the end of the day a bad design (Intel 8086) which had huge amount of resources poured into from the enormous size of the PC market was bound to eventually become a fast design. While 086, 286 and 386 were in many ways poor design where each generation mainly concentrated on getting rid of the 16 bit x86 baggage, by the time of the 486 Intel's huge revenues allowed them to design a pretty good chip.

However, economics of scale are hard to ignore which is why I think Intel's biggest mistake (and there have been many usually at a cost of a few $billion each), was refusing to take high-volume lower margin stuff seriously and being to afraid that Atom would cannibalise their high margin stuff. Ironically this is not too dissimilar to the early workstation RISC vendors who didn't take Intel with its (lower margin) high volume chips seriously until it was too late.

Anyway, the low margin high volume stuff is what enabled the ARM players (and ultimately anyone using TSMC) do is gain enough experience to become a thread. The ARM designer are pretty clever at what they do, but until recently they haven't gone after desktop stuff.

And adapting a low power design to high power desktop/server usage isn't as easy as just cranking up the power. Memory, interconnect, cache designs are all vastly different when every mA counts. On Epyc the interconnect, IO etc. (the non-core parts) take up way more power than any mobile could ever afford.

However, what we enthusiast probably like the best about PCs is that (mostly by accident) there is a DIY market, whereas Apple and a lot of ARM vendors just want to buy disposable rubbish (Huawei before their recent trouble were keen to push those silly proprietary MM card instead of microSD, Apple were of course the first to solder everything, and make total landfill fodder, etc.).

So, while I don't think there is any major technical reason why ARM can't take a large part of the market, and there are genuine reasons why x86 has overheads which ARM or RISC-V do not have, I won't be impressed if this leads to the demise of the DIY market.
 
Joxeon said:
Can't see these Renoir APUs doing 2100 Fclk myself else they have just made the 3600XT / 3800XT obsolete.
Click to expand...

It's a monolithic 7nm die, with Zen2 cores and all the I/O is 7nm, the XTs are still using the 12nm I/O hence lower FCLK. .

Also no it won't make them obsolete they are targeted and different markets and have completely different specifications.
 
Journey said:
It's a monolithic 7nm die, with Zen2 cores and all the I/O is 7nm, the XTs are still using the 12nm I/O hence lower FCLK. .

Also no it won't make them obsolete they are targeted and different markets and have completely different specifications.
Click to expand...
Will be nice for zen 3 if true.

Who would buy the XT's if the 4000 APUs out perform those parts at a similar or lower price regardless of whether you need the graphics cores.
 
Last edited:
Joxeon said:
Will be nice for zen 3 if true.

Who would buy the XT's if the 4000 APUs out perform those parts at a similar or lower price regardless of whether you need the graphics cores.
Click to expand...

I think I read that these are going to be OEM rather than boxed chips - let's see what 2pm brings with reviews..
 
Joxeon said:
Who would buy the XT's if the 4000 APUs out perform those parts at a similar or lower price regardless of whether you need the graphics cores.
Click to expand...
I think that in a lot of scenarios a monolithic design will outperform the chiplet version.
However, some of the leaks seem to suggest that while ST looks good (mostly due less latency not having to the IO chip presumably), MT doesn't which is most likely due the smaller L3 cache.

In an ideal scenario, AMD would have enough volume to not have to bother with chiplets, but by the time 7nm has enough volume (between releasing Zen2, the GPUs, the APUs, and the consoles (not sure how much of the work Sony and Microsoft do there)), 5nm will be here and the cost of masks, dies, validation etc. is meant to be even higher for 5nm so again chiplet makes sense.

Weren't the estimate for 7nm designs in the $100s millions? You'd think some of that wouldn't apply for every die revision but what we do know is that costs are very high. Even for Intel with their vastly higher volume and in-house fabs, the whole idea of Intel's EMIB and similar is to address these spiralling costs.
 
RavenXXX2 said:
Why would there be an NDA or indeed any reviews of OEM APUs?
Click to expand...

Because ultimately an NDA covers the release of product information to everyone from those that have access. As for reviews that would be based on demand I guess, and if anyone can get them, if they generate views/clicks reviewers will want them and no doubt AMD wouldn't object to the extra coverage.

Joxeon said:
Who would buy the XT's if the 4000 APUs out perform those parts at a similar or lower price regardless of whether you need the graphics cores.
Click to expand...

You are assuming they outperform them in every scenario, which will not be the case. In some cases a the 4700G might be a better option than a 3800XT, if you have specific requirements, but equally the opposite could be true if you have others.
 
Status
Not open for further replies.
Back
Top Bottom