i7-5775C: why did Intel abandon development of eDRAM?

[Veseleil]

Games. The 5775C's gaming performance is extremely strong and it exhibits the same behavioral pattern of the 5800X3D and its performance will match the substantially faster and newer i5-10600K in games which the 5800X3D also tends to exhibit strong performance gains, such as Borderlands 3:

A Broadwell Retrospective Review in 2020: Is eDRAM Still Worth It?

www.anandtech.com

In Final Fantasy XIV, you'll observe the same type of behavior:

A Broadwell Retrospective Review in 2020: Is eDRAM Still Worth It?

www.anandtech.com

It's funny, this chip was way ahead of its time. It consistently pulls Comet Lake i5 weight on games known to benefit from cache performance, all of that despite being almost 5 years older. I really recommend perusing that revisit Ian did on this CPU back in 2020. It's probably the most comprehensive, informational and useful resource on Broadwell there is.



That'd be hard to say, I don't believe Intel disclosed it at all, but it should have around 100 GB/s bandwidth, half each way. Finding this information would probably be hard as the CPUs are now 10 years old.

The 1440p result is impressive:

 

[80251]

It's too bad I bought my 9700k, I should've been looking for a 5775c!

 

[vvkvvkvvk]

Apple. They wanted stronger integrated GPUs, the Crystalwell was a neat hack, but in the end Intel couldn't deliver good enough performance so they hopped on to their ARM counterparents and "own" GPUs in the end.

 

[AusWolf]

As a wild guess, I'd say it's because it isn't fast enough to act as proper cache, but it isn't large enough to contribute to your RAM, while development costs were high.

Also, isn't the 5775C that weird mutant with an AMD iGPU? I bet licensing wasn't flawless.

 

[80251]

As a wild guess, I'd say it's because it isn't fast enough to act as proper cache, but it isn't large enough to contribute to your RAM, while development costs were high.

Also, isn't the 5775C that weird mutant with an AMD iGPU? I bet licensing wasn't flawless.

You should really look at Dr. Dro's post re: the performance of the i7-5775c as of 2020, it holds its own against CPU's two generations newer.

 

[lemonadesoda]

I “missed” 5775C first time round, since at the time I was a content 2x Xeon workstation owner. It has been an interesting read to learn about 5775C and that it still holds its own in many benchmarks/games even today. And looking at prices of second hand 5775C on ebay today, they have held their values (EUR 150-200). But that’s also unfortunately, at those prices, not a great value for money upgrade to an existing system. Very different compared to the current prices of xeon e5 series where a 14c/28t can be picked up for EUR 15. A sideways upgrade (used) from consumer to workstation is great value at the moment.

 

[AusWolf]

You should really look at Dr. Dro's post re: the performance of the i7-5775c as of 2020, it holds its own against CPU's two generations newer.

Sure, but how was it when it was new? I mean, modern games seem to be happy for large CPU caches, as it's demonstrated by Ryzen X3D CPUs.

I think if Intel pulled a similar move now, it would be way more appreciated than it was with the 5775C.

 

[Mussels]

why didn't Intel try and develop a faster and/or larger eDRAM?

If yields arent good enough, the price has to go up

And if people wont buy at that higher price, you dont make it

 

[TumbleGeorge]

That'd be hard to say, I don't believe Intel disclosed it at all, but it should have around 100 GB/s bandwidth,

Come on, come on, this is not a guess, nor is it a matter of faith, and the information is quite easily found on the internet.

 

[Wirko]

Come on, come on, this is not a guess, nor is it a matter of faith, and the information is quite easily found on the internet.

Bandwidth, sure. But is it 3200 MT/s over 128 bits in each direction, or something else? Maybe some guy with electron and x-ray eyes did take it apart, I don't know. In any case, the chip is an interesting mix between HBM, LPDDR, GDDR and DDR but not very close to any of them.

 

[P4-630]

Don't care much anymore, my i7 12700K is trashing it...

 

[TumbleGeorge]

Bandwidth, sure. But is it 3200 MT/s over 128 bits in each direction, or something else? Maybe some guy with electron and x-ray eyes did take it apart, I don't know. In any case, the chip is an interesting mix between HBM, LPDDR, GDDR and DDR but not very close to any of them.

100GB/s between eDRAM and lower level cache bidirectional at full speed. To and from RAM at the speed of the RAM itself, it can't possibly be faster.

 

[Tahagomizer]

Corporations want to develop a minimum viable product, which means the cheapest thing which is better than the last one by just enough to be able to sell it and leave room for market segmentation while minimizing cost of developing the next thing. This approach is also why if a sportsman can beat a world record by 20% he won't do it in one go, instead doing it twenty times by 1% thus getting paid twenty times.

With Broadwell Intel probably didn't have the desktop version ready by the time pencil pushers demanded for a new product to hit the market so they used a chip obviously designed for high-end laptops. It was good, I used the 5775c until I got the 9600k and never felt like it lacked performance

 

[Shrek]

Isn't the eDRAM concept what Apple is doing with the M series chip? or am I mixing things up?

Seems to be the opposite of the chiplet idea.

 

[Dr. Dro]

Come on, come on, this is not a guess, nor is it a matter of faith, and the information is quite easily found on the internet.

WikiChip and CPU-World don't have that information, and WikiChip claims that Intel didn't officially disclose the specifications for Crystal Well's eDRAM module. So idk.

 

[TumbleGeorge]

WikiChip and CPU-World don't have that information, and WikiChip claims that Intel didn't officially disclose the specifications for Crystal Well's eDRAM module. So idk.

Then read it with a grain of salt, as if it were a rumor.

 

[80251]

Corporations want to develop a minimum viable product, which means the cheapest thing which is better than the last one by just enough to be able to sell it and leave room for market segmentation while minimizing cost of developing the next thing. This approach is also why if a sportsman can beat a world record by 20% he won't do it in one go, instead doing it twenty times by 1% thus getting paid twenty times.

With Broadwell Intel probably didn't have the desktop version ready by the time pencil pushers demanded for a new product to hit the market so they used a chip obviously designed for high-end laptops. It was good, I used the 5775c until I got the 9600k and never felt like it lacked performance

Was your 9600k much faster than the i7-5775c at stock settings? The benchmarks Dr. Dro posted make it seem like the 5775c could probably beat out any Coffeelake CPU in certain games.

 

[Wirko]

Then read it with a grain of salt, as if it were a rumor.

Again, bandwidth is known, but transfer rate and bus width are not (if you knew one of them, you could calculate the other).
After reading Anand's overview, I tend to assume it's 1600 MT/s and 256+256 bits wide bus. That's very, very wide. One-half of a single HBM gen 1 stack (1000 MT/s and 512 bits) would be remotely similar to that.

 

[Toothless]

To those who keep thinking the 5775c was AMD related: it wasn't. Broadwell wasn't involved with the Devil NUC that was made a generation later.

 

[R0H1T]

It's too bad I bought my 9700k, I should've been looking for a 5775c!

Those were hard to find, at least after they went EoL. I tried to get one even second hand at less than highway robbery rates but just gave up.

 

[Wirko]

Isn't the eDRAM concept what Apple is doing with the M series chip? or am I mixing things up?

Seems to be the opposite of the chiplet idea.

The M1 implements the chiplet idea and Broadwell does too, what is supposed to be opposite here? Apple has LPDDR instead of eDRAM, but either option is just a chiplet (or a set of equal chiplets). Apple also has a very wide memory bus, up to 1024 bits on the M1 Ültra. At this point, LPDDR5 has basically converged with HBM3.

 

[R0H1T]

At this point, LPDDR5 has basically converged with HBM3.

It definitely hasn't.

 

[dragontamer5788]

It definitely hasn't.

He's using hyperbole. HBM3 is a 4096-pin, low power bus. The idea is to have more pins and run less power on each pin.

Typical RAM is somewhere from 200 to 300 pins. So seeing LPDDR5 at 1024-pins is more similar to HBM than it is to regular old RAM interfaces.

 

[Shrek]

The M1 implements the chiplet idea and Broadwell does too, what is supposed to be opposite here? Apple has LPDDR instead of eDRAM, but either option is just a chiplet (or a set of equal chiplets). Apple also has a very wide memory bus, up to 1024 bits on the M1 Ültra. At this point, LPDDR5 has basically converged with HBM3.

By chiplets I meant separate chips connected by wiring; but maybe I am misusing the terminology.

 

[Wirko]

By chiplet I meant seperate chips connected by wiring; but maybe I am misusing the terminology.

Well yes, the M1 fits this description.